diff --git a/README.md b/README.md index 5a30cad..eaf632a 100644 --- a/README.md +++ b/README.md @@ -35,9 +35,10 @@ controller is authored; shoe physics and physical transfer remain unqualified. [Hardware matched robot simulation](reports/Hardware%20matched%20robot%20simulation.md) now includes an [interactive 3D arm/gripper replay](https://research.dvidia.org/downloads/topics/physical-grounding/runs/2026-10-11-hardware-reference/replay/) -and the [upcoming hardware support map](https://github.com/Dvidia-Inference/dvidia-training/blob/93fc549a370cdd448b49380c5beff3b2641c2795/docs/hardware-roadmap.md). -YAM, OpenArm, PiPER, ARX X5 and PAROL6 are planned targets, with separate simulator, -adapter and benchmark gates. The replay displays frozen native poses and geometry; +and the [hardware support map](https://github.com/Dvidia-Inference/dvidia-training/blob/928fed79eca09a1a5081aae7720ab78c2e5b817c/docs/hardware-roadmap.md). +PiPER and SO-101 have runnable CPU simulation profiles with unloaded numerical +receipts; their object-task and physical adapter qualification remain open. +YAM, OpenArm, ARX X5 and PAROL6 remain planned integration targets. The replay displays frozen native poses and geometry; it adds no live physics, learned skill or physical-fidelity result. [DVIDIA Observation Lab pilot](reports/DVIDIA%20Observation%20Lab%20pilot.md) publishes our local video detection and review workflow: sampled object/palm proposals, editable timeline revisions and JSON export. Each recorded engineering run completed six public excerpts (19 seconds and 38 sampled frames); independent annotation accuracy remains unmeasured. [Read the paper](https://research.dvidia.org/papers/dvidia-observation-lab/), [install the versioned open-source release](https://github.com/Dvidia-Inference/dvidia-training/releases/tag/observation-lab-v0.1.0), or [try the synthetic browser review demo](https://huggingface.co/spaces/Dvidia/observation-lab). The hosted demo uses authored clips and boxes; real detection runs locally with separately provisioned upstream models. No action captions, validated depth or robot skill are established. @@ -90,3 +91,7 @@ The [agenda](reports/Physical%20skill%20research%20agenda.md) responds to Sam Pa Citation keys such as `zhang2020hands` are local to each topic's `references.bib`; the same keys appear in `topic.json`. Topic prose links keys to primary URLs. Living documents use an explicitly labelled access year when a publication date is unavailable. Cite original work for its findings and [CITATION.cff](CITATION.cff) for these protocols. Record an immutable revision when actually running code or using a dataset. Original writing and metadata use [CC BY 4.0](LICENSE); original code uses [MIT](LICENSE-CODE). Referenced code, checkpoints, media and datasets keep their own terms. Repository openness does not establish permission to redistribute a cited dataset. + +## Robot simulation library + +[Public Robot Simulation Lab](https://huggingface.co/spaces/Dvidia/robot-simulation-lab) combines the recorded UR5e contact task with separate PiPER and SO-101 CPU mechanics receipts. [Run the open-source profiles locally](https://github.com/Dvidia-Inference/dvidia-training/blob/928fed79eca09a1a5081aae7720ab78c2e5b817c/docs/robot-simulation-library.md). Additional profile support does not transfer task or hardware qualification. diff --git a/docs/downloads/reports/Hardware matched robot simulation.md b/docs/downloads/reports/Hardware matched robot simulation.md index 4631697..a5eba5e 100644 --- a/docs/downloads/reports/Hardware matched robot simulation.md +++ b/docs/downloads/reports/Hardware matched robot simulation.md @@ -28,19 +28,19 @@ The exporter verifies the frozen receipt and pinned assets, reconstructs recorde The ready-made browser replay can be viewed on another supported browser without MuJoCo. Reconstructing an original numerical receipt requires its compiled-model hash to match; native builds can differ between platforms. When they differ, generate and retain a separate benchmark and replay on the current machine alongside the original results. -## Upcoming hardware profiles +## Hardware integration roadmap -The following support list is a plan, not a purchase decision or implemented compatibility. **All five additions remain upcoming.** Each needs a pinned simulator, physical adapter and benchmark evidence tied to its exact hardware/controller revision. Arm-axis counts exclude the gripper; a future five-finger hand has a separate actuation and sensing contract. +The following list separates runnable numerical simulation from planned integration. **PiPER now has a pinned CPU simulation profile; YAM, OpenArm, ARX X5 and PAROL6 remain upcoming.** All five still need their own object-task qualification and physical adapter evidence tied to the exact hardware/controller revision. Arm-axis counts exclude the gripper; a future five-finger hand has a separate actuation and sensing contract. | Priority | Target | Intended role and remaining work | | --- | --- | --- | | 1 | **YAM — 6 arm axes + gripper** | First candidate for a full-size shoe grasp-and-place benchmark. MIT control/models include MuJoCo and teleoperation; complete open manufacturing designs are not established. Pin arm/gripper revisions and measure their physical response. [Official SDK/models](https://github.com/i2rt-robotics/i2rt). | | 2 | **OpenArm — 7 arm axes + gripper per side** | Later two-arm coordination. CERN-OHL-S-2.0 hardware and Apache-2.0 simulation; native LeRobot connection currently requires Linux. Keep v1/v2 profiles distinct and assess hand mounts, mass, power and remaining payload. [Hardware](https://github.com/enactic/openarm_hardware), [MuJoCo](https://github.com/enactic/openarm_mujoco), [LeRobot](https://huggingface.co/docs/lerobot/main/openarm). | -| 3 | **AgileX PiPER — 6 arm axes + gripper** | Alternative six-axis research/community target. Open SDK and MIT Menagerie model; full manufacturing openness is not established. Bind firmware, offsets, control modes and jaw calibration. [SDK](https://github.com/agilexrobotics/piper_sdk), [model](https://github.com/google-deepmind/mujoco_menagerie/tree/main/agilex_piper). | +| 3 | **AgileX PiPER — 6 arm axes + gripper** | Runnable pinned CPU profile with unloaded numerical receipts; object-task and physical qualification remain open. Open SDK and MIT Menagerie model; full manufacturing openness is not established. Bind firmware, offsets, control modes and jaw calibration for a physical adapter. [SDK](https://github.com/agilexrobotics/piper_sdk), [model](https://github.com/google-deepmind/mujoco_menagerie/tree/main/agilex_piper). | | 4 | **ARX X5 — 6 arm axes + gripper** | Joint position/velocity/torque research adapter and later multiple-arm control. MIT research SDK and community LeRobot plugin; exact X5 simulation and hardware licensing need qualification. [SDK](https://github.com/real-stanford/arx5-sdk), [plugin](https://github.com/villekuosmanen/lerobot-arx5). | | 5 | **PAROL6 — 6 arm axes + separate gripper** | Community-build desktop target. GPLv3 mechanics/software and ROS/Python tools; the documented open-loop version has limit-switch sensing, so actual joint feedback requires verification. Complete electronics openness is not established. [Project](https://github.com/Source-Robotics/PAROL6-Desktop-robot-arm), [specifications](https://source-robotics.github.io/PAROL-docs/page2_2/). | -SO-101 remains the accessible planned entry profile, with five arm axes plus a gripper. The current UR5e/Robotiq reference is separately characterized. Gripper opening, finger geometry, slip, workspace and usable payload drive the first shoe hardware choice. Device support advances separately from learned skill and physical qualification. ([DVIDIA hardware support map](https://github.com/Dvidia-Inference/dvidia-training/blob/93fc549a370cdd448b49380c5beff3b2641c2795/docs/hardware-roadmap.md)) +SO-101 is the accessible entry profile, with five arm axes plus a gripper. Its pinned CPU simulation is now runnable; its object-task and physical adapter qualification remain open. The UR5e/Robotiq reference is separately characterized. Gripper opening, finger geometry, slip, workspace and usable payload drive the first shoe hardware choice. Device support advances separately from learned skill and physical qualification. ([DVIDIA hardware support map](https://github.com/Dvidia-Inference/dvidia-training/blob/928fed79eca09a1a5081aae7720ab78c2e5b817c/docs/hardware-roadmap.md)) ## Match observable mechanics before calling it fidelity @@ -76,7 +76,7 @@ Use fixed issued-command replays to expose timestep sensitivity, then measure co LeRobot offers hardware/teleoperation interfaces, cameras and dataset tooling without requiring a supplied policy architecture. Its official custom-policy interface supports independently packaged models with training, reset and action-selection methods. DVIDIA can reuse that infrastructure while developing its own objectives, model and skill artifacts. **No DVIDIA/LeRobot integration or robot action policy is built by this reference release.** The earlier compact transition predictor forecasts motion conditional on a command; it does not choose the command. A proposed first seam is a validated local dataset reader and strict observation/action adapter, followed by an original DVIDIA trainer. ([Custom hardware guide](https://huggingface.co/docs/lerobot/main/integrate_hardware), [Custom policy guide](https://huggingface.co/docs/lerobot/main/bring_your_own_policies)) -Hardware selection remains explicit. SO-101's six motor channels are shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll and gripper: **five arm axes plus a gripper, not six arm axes plus a gripper**. Its body positions can use degrees or normalized values, and the jaw uses a normalized command. A separate named profile must specify ordering, units, calibration, action representation and requested versus actually sent targets; it cannot accept DVIDIA's current six-axis/radian artifact unchanged. ([SO-101 guide](https://huggingface.co/docs/lerobot/main/so101), [Follower implementation](https://github.com/huggingface/lerobot/blob/main/src/lerobot/robots/so_follower/so_follower.py), [Action representations](https://huggingface.co/docs/lerobot/main/action_representations)) +Hardware selection remains explicit. SO-101's six motor channels are shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll and gripper: **five arm axes plus a gripper, not six arm axes plus a gripper**. LeRobot's physical follower interface can use degrees or normalized body positions, and its jaw uses a normalized command. A physical adapter must specify ordering, units, calibration, action representation and requested versus actually sent targets; it cannot accept DVIDIA's six-axis reference artifact or the native SO-101 profile's radian targets unchanged. ([SO-101 guide](https://huggingface.co/docs/lerobot/main/so101), [Follower implementation](https://github.com/huggingface/lerobot/blob/main/src/lerobot/robots/so_follower/so_follower.py), [Action representations](https://huggingface.co/docs/lerobot/main/action_representations)) LeRobotDataset v3 stores state/action rows, camera video and episode/schema metadata, and recordings can stay local. However, the recording guide warns that frame-index/FPS timestamps can make motion appear faster when acquisition misses the requested cadence. DVIDIA needs a sidecar with **actual monotonic acquisition, arrival, command and feedback clocks**, camera calibration, missed samples, controller settings and physical source groups. Nominal timestamps alone cannot fit actuator delay or validate physical speed. Pin and test the chosen LeRobot version and review datasets/assets separately from its Apache-2.0 software license. ([Dataset format](https://huggingface.co/docs/lerobot/main/lerobot-dataset-v3), [Recording and cadence guide](https://huggingface.co/docs/lerobot/main/il_robots), [Software license](https://github.com/huggingface/lerobot/blob/main/LICENSE)) @@ -106,6 +106,23 @@ Adjacent-timestep comparisons align 201 elapsed boundaries per comparison, inclu A separate authored sphere-overlap diagnostic produces one arm contact with active collision masks and zero with those masks disabled. It checks that the selected proxy participates in collision detection; it is not a moving collision-response or avoidance test. The software checkpoint also passed 488 local tests, including native contract checks. CPU-only headless characterization is demonstrated for this unloaded reference workload; shoe contact, complete training throughput, minimum hardware, physical accuracy and the composed two-arm system remain unmeasured. ([Collision receipt](https://research.dvidia.org/downloads/topics/physical-grounding/runs/2026-10-11-hardware-reference/report.json), [Reference instructions](https://github.com/Dvidia-Inference/dvidia-training/blob/c95a4385dae4b654b01ef75f56cbcc144fde2f5c/docs/hardware-reference.md)) +## PiPER and SO-101 now have runnable CPU profiles + +The library now includes pinned **PiPER (six arm axes plus one gripper command)** and **SO-101 (five arm axes plus one gripper command)** profiles. Both expose explicit model fetch, offline asset verification, native audit and bounded numerical benchmarking. The packaged models preserve upstream geometry and mechanics; DVIDIA supplies the profile tooling and evidence contracts. PiPER's slider command is single-finger travel in metres with a coupled opposite finger. SO-101's moving jaw uses a hinge target in radians. Neither is a calibrated physical gap or force command, and SO-101 must not be labeled as a six-axis arm. ([Local simulation guide](https://github.com/Dvidia-Inference/dvidia-training/blob/928fed79eca09a1a5081aae7720ab78c2e5b817c/docs/robot-simulation-library.md), [Pinned model contracts](https://github.com/Dvidia-Inference/dvidia-training/blob/928fed79eca09a1a5081aae7720ab78c2e5b817c/src/dvidia_training/robot_profiles.lock.json)) + +A serial run with process-level network denial retained six attempts per profile: hold, small arm sweep and gripper cycle, each repeated twice for two simulated seconds at a 1 ms physics step and 20 ms command boundary. These repeats check numerical mechanics; they are not independent task scenes or a learned-policy evaluation. Both profiles produced **6/6 numerically valid attempts**. Focused tests separately confirm that commanded arm and jaw cases change the native state compared with holding the starting commands. + +| Profile | Simulated seconds / complete-attempt wall second | Per-attempt 20 ms computation p95 | Computation windows above 20 ms | +| --- | ---: | ---: | ---: | +| PiPER | 4.38–14.62× | 0.446–0.500 ms | 0/600 | +| SO-101 | 14.73–20.17× | 0.396–0.744 ms | 0/600 | + +All attempts remain in the [Robot Simulation Lab](https://huggingface.co/spaces/Dvidia/robot-simulation-lab), including PiPER's slower first attempt. Wall time includes asset checks, compilation and recorded state construction; output serialization, browser rendering, physical sensing/transport and learned inference are outside that measurement. Runtime was macOS/Darwin arm64 with Python 3.12.9, MuJoCo 3.15.0 and NumPy 2.5.3. These CPU runs used no discrete-GPU workload; host load was not controlled, and universal minimum memory or real-time hardware behavior is not established. The exact bundles include source/runtime identities, native controls and effort units, gravity settings, every recorded trajectory and original licenses. ([PiPER receipt](https://huggingface.co/spaces/Dvidia/robot-simulation-lab/resolve/main/profiles/piper/report.json), [SO-101 receipt](https://huggingface.co/spaces/Dvidia/robot-simulation-lab/resolve/main/profiles/so101/report.json)) + +PiPER simulation assets are MIT; SO-101 assets are Apache-2.0. The latter has published build/design files and LeRobot guidance, while an open PiPER SDK does not by itself establish open manufacturing designs. The models' original notices are retained, and their attribution is separate from DVIDIA software. ([SO-101 design](https://github.com/TheRobotStudio/SO-ARM100), [LeRobot guide](https://huggingface.co/docs/lerobot/so101), [PiPER SDK](https://github.com/agilexrobotics/piper_sdk)) + +These additions are ready for local unloaded simulation. Each still needs its own object-task environment, grasp/release controls, calibrated observations and fresh closed-loop policy benchmark. YAM, OpenArm, ARX X5 and PAROL6 remain further integration targets; bimanual and dexterous manipulation retain separate contracts. The hosted lab displays the UR5e contact replay separately from these numerical receipts, so additional model support cannot be mistaken for inherited grasp competence. + ## Release an evidence-bearing skill, not an appearance match The installable unit should carry the policy, skill goal, observations/actions and units, named embodiment ordering, adapter, dependency hashes, calibration requirements, supported conditions, refusal/recovery behavior and benchmark receipts. Imported asset structure, characterized simulation and measured hardware agreement advance separately. A reference-only or simulation-qualified package remains useful for research while declaring physical execution unqualified. This follows the current DVIDIA roadmap's separation of collection, learning, simulation and physical qualification. ([Execution map](https://github.com/Dvidia-Inference/dvidia-training/blob/9172c37baeed5925e7bea3af2c92b0bf37a63c24/docs/roadmap.md)) diff --git a/docs/downloads/reports/Native grasp task benchmark.md b/docs/downloads/reports/Native grasp task benchmark.md index 75e07e8..c1b1c5b 100644 --- a/docs/downloads/reports/Native grasp task benchmark.md +++ b/docs/downloads/reports/Native grasp task benchmark.md @@ -82,3 +82,5 @@ The product opportunity is the loop connecting reviewed skill-space evidence, bo Run the benchmark using the [installation and evaluation guide](https://github.com/Dvidia-Inference/dvidia-training/blob/ac920cbf70147a7edeb0c706a9b71047f318e12f/docs/native-task-benchmark.md). The implementation and review are available in [GitHub PR #11](https://github.com/Dvidia-Inference/dvidia-training/pull/11). Published by **DVIDIA** · [hello@dvidia.org](mailto:hello@dvidia.org) · [@iammrriver on X](https://x.com/iammrriver) + +The [public Robot Simulation Lab](https://huggingface.co/spaces/Dvidia/robot-simulation-lab) also provides separate unloaded PiPER and SO-101 receipts and installation instructions. [Browser acceptance evidence](https://research.dvidia.org/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/acceptance.json) checks the recorded task replay on desktop and narrow screens. diff --git a/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/320x740-carry.png b/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/320x740-carry.png new file mode 100644 index 0000000..0b75cba Binary files /dev/null and b/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/320x740-carry.png differ diff --git a/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/README.md b/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/README.md new file mode 100644 index 0000000..8a5cd21 --- /dev/null +++ b/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/README.md @@ -0,0 +1,3 @@ +# Recorded task replay browser checks + +These checks inspect the frozen native replay; they do not rerun physics. All nine displayed attempts and their source states are checked. The complete source denominator remains eighteen attempts. Screenshots show the initial block, elevated carry and released placement. The 320px layout retains readable controls. No browser errors or external requests were observed. diff --git a/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/acceptance.json b/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/acceptance.json new file mode 100644 index 0000000..ae5ecbb --- /dev/null +++ b/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/acceptance.json @@ -0,0 +1,17 @@ +{ + "scope": "Browser recorded-task replay acceptance; no learned-policy or physical qualification", + "demonstrated_attempt_id": "nominal-00--placement", + "recorded_success": true, + "carry_frame": 291, + "carry_seconds": 5.820000000000278, + "rendered_attempts": 9, + "full_benchmark_attempted": 18, + "checks": [ + "Native block, table and target geometry; six joint readouts; authored-controller scope and full source denominator", + "Every selected outcome matches its sealed source; every block pose matches the recorded native state; all controls replay to their terminal row", + "Initial, observed carry and terminal target views captured from actual state; completed source labels require measured lifted transfer and final target tolerance", + "Recorded playback, collision mode and idle rendering; 390 px and 320 px layouts keep 44 px controls without horizontal overflow" + ], + "errors": [], + "remote": [] +} diff --git a/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/carry-view.png b/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/carry-view.png new file mode 100644 index 0000000..a0c4631 Binary files /dev/null and b/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/carry-view.png differ diff --git a/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/final-view.png b/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/final-view.png new file mode 100644 index 0000000..4ad1209 Binary files /dev/null and b/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/final-view.png differ diff --git a/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/initial-view.png b/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/initial-view.png new file mode 100644 index 0000000..c9933bd Binary files /dev/null and b/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/initial-view.png differ diff --git a/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/source.json b/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/source.json new file mode 100644 index 0000000..37a2082 --- /dev/null +++ b/docs/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/source.json @@ -0,0 +1,7 @@ +{ + "scope": "Browser replay acceptance, not fresh physics or hardware validation", + "tested_replay_integrity_sha256": "f4a334c177eab76942ee0ebb282138340040fc22be3f6377713d332fc572ba21", + "harness_source_sha256": "0d41eba08d7717d9b40e446ffa59ee017beeb1c20d9e40d4b73d62d82d61f729", + "checks": "All selected body positions, recorded outcomes, frame counts, paused rendering, desktop and mobile layout; no browser errors or external requests", + "screenshots": "Initial, measured carry, final placement and 320px mobile view; independently visually reviewed" +} diff --git a/docs/feed.xml b/docs/feed.xml index 86588dd..9dbd18e 100644 --- a/docs/feed.xml +++ b/docs/feed.xml @@ -1,2 +1,2 @@ -DVIDIA Researchhttps://research.dvidia.org/Research footprints on demonstrations, calibrated simulation and robot skills.enNative grasp task benchmarkhttps://research.dvidia.org/papers/native-grasp-task-benchmark/https://research.dvidia.org/papers/native-grasp-task-benchmark/Sun, 11 Oct 2026 12:00:00 GMT6/6 simulated placements · controls 0/12 · real-time and hardware unqualified. A pinned six-axis arm and coupled gripper complete 6/6 frozen synthetic rigid-block placements, while matched recorded-arm/open-jaw and hold controls complete 0/12. Native contact must produce grasp, sustained lift, retained carry, release and settled placement with tool retreat. All 18 attempts and timing traces remain in the receipt; one placement takes 29.51 wall seconds for 20 simulated seconds and misses 251 of 1,000 computation windows, so real-time operation is not established. An offline 3D mesh replay exposes the recorded motion. The authored privileged-state baseline provides a benchmark for a future compact DVIDIA action policy, not a learned footage-to-action model, validated shoe simulation or physical robot qualification.Hardware matched robot simulationhttps://research.dvidia.org/papers/hardware-matched-robot-simulation/https://research.dvidia.org/papers/hardware-matched-robot-simulation/Sun, 11 Oct 2026 12:00:00 GMT18/18 numerical runs valid · physical fidelity unmeasured. A separate UR5e/Robotiq reference preserves pinned geometry, inertias, moving-link collision proxies and coupled gripper structure: fourteen modeled coordinates and seven commands. Eighteen deterministic unloaded motion/closure runs are numerically valid; at a 1 ms step, CPU physics throughput measures 40.80–47.29 simulated seconds per measured stepping wall second on one host, excluding loading, sensing, rendering, policy and real actuation. Adjacent timestep comparisons and a matched static overlap expose numerical sensitivity and collision participation. A pure-Python paired trace scorer measures joint, tool and jaw mismatch without automatic qualification. Left/right five-finger references are separately audited; LeRobot collection with an original DVIDIA policy is proposed. Serial calibration, physical response, shoe task learning and bimanual execution remain unmeasured. An offline 3D mesh replay now exposes the recorded arm/gripper poses without rerunning physics. YAM, OpenArm, PiPER, ARX X5 and PAROL6 are upcoming profiles; their DVIDIA integration and skill qualification remain unimplemented.DVIDIA Observation Lab pilothttps://research.dvidia.org/papers/dvidia-observation-lab/https://research.dvidia.org/papers/dvidia-observation-lab/Sun, 11 Oct 2026 12:00:00 GMTLocal review release · motion and actuator source pilots · hardware unqualified. DVIDIA Observation Lab supplies local object/palm proposals, reviewed timeline revisions and separate causal image-motion forecasts; real annotation and motion accuracy remain unmeasured. A new CPU pilot learns next-20-ms joint/jaw deltas from current native state and issued commands. Twelve seeded groups retain 36 recordings with 8/2/2 train/development/test grouping; 4,553 test transitions are not independent robot trials. The 6,438-byte action model computes at 0.207 ms p95 on one host and improves RMSE, while jaw MAE/worst error regress against persistence and an initial opening is substantially overpredicted. These are privileged simulator measurements and model-only timings. A proposed shoe capture plan has no footage or executed shoe benchmark. The v0.1 release and hosted synthetic review demo remain separate; no shoe policy, physical adapter, validated shoe physics or hardware qualification is established.Robotics research pipelinehttps://research.dvidia.org/papers/robotics-research-pipeline/https://research.dvidia.org/papers/robotics-research-pipeline/Fri, 09 Oct 2026 12:00:00 GMTArchitecture proposal · current simulation alpha · hardware unqualified. An updated architecture connects immutable Skillspace recordings, reviewed task-label proposals and synchronized measured telemetry to grouped datasets, separate visual and motor learners, and a proposed local state supervisor. The implemented CPU baseline retains authored phases, gripper decisions and recovery around its learned joint-delta head. New annotation and supervisor pilots, fresh frozen confirmation, physics calibration and compatible skill-installation gates define the next milestones. Automatic labeling, learned supervision, human-video action recovery and physical arm qualification remain unestablished; this publication adds no new training or hardware result.Skillspace footage training pipelinehttps://research.dvidia.org/papers/skillspace-footage-training/https://research.dvidia.org/papers/skillspace-footage-training/Thu, 08 Oct 2026 12:00:00 GMTPublic CPU package · one native scene qualified · simulation-only. The public Dvidia-Inference/dvidia-training repository provides a standalone CPU pipeline that audits Skillspace media, freezes grouped splits and exports inspectable artifacts. Ten synthetic six-second clips train a visual predictor with a 0.5106-second median pipeline time across three repeats on one macOS host. Aligned native simulator telemetry supplies 4,029 training labels for a learned joint-delta head, evaluated on one held-out recording. Its capsule passes native placement qualification in that recording's exact scene; identical joint targets with open jaws fail. Installation and hardware guides accompany the package. Human-video action recovery, shoe organization, demonstration sufficiency, independent scene generalization and physical robot competence remain unestablished.Skill capsule acquisition and optimizationhttps://research.dvidia.org/papers/skill-capsule-acquisition/https://research.dvidia.org/papers/skill-capsule-acquisition/Wed, 07 Oct 2026 12:00:00 GMTSimulation-only · teacher/student 7/8 · no overall speedup. A compact learned CPU movement head replaces joint-target computation inside an authored placement supervisor. Teacher and student both pass 7/8 in a new frozen evaluation, while episode wall time increases from 6.273 to 6.360 seconds. A 39,845-byte JSON capsule imports into the local lab, checks its exact runtime and qualifies the current scene through native placement and a recorded-arm/open-jaw control. Scene changes require a fresh test. One known scene and repeat pass native network denial; perception, complete task learning, other embodiments and hardware remain unqualified.Skillspace arm benchmarks and evolutionhttps://research.dvidia.org/papers/skillspace-arm-benchmarks/https://research.dvidia.org/papers/skillspace-arm-benchmarks/Wed, 07 Oct 2026 12:00:00 GMTMixed held-out result · 30/32 baseline, 29/32 updated. Two controller revisions improve development completion from 14/20 to 17/20, but the frozen paired test declines from 30/32 to 29/32: both pass all 24 normal-strength cases while actuator-stress completion falls from 6/8 to 5/8. Instrumented CPU throughput also declines. We publish the regression, unchanged contact and success rules, exact controller snapshots, richer physical inputs, runtime-bound installation and a proposed staged Skillspace contract. Contact forces and penetration remain uncalibrated; learned perception and physical qualification are open.Installing a Skillspace on a simulated armhttps://research.dvidia.org/papers/skillspace-arm-installation/https://research.dvidia.org/papers/skillspace-arm-installation/Wed, 07 Oct 2026 12:00:00 GMTPublic skill link to authored adapter · 6/6 scene placements. The first link-to-arm product path installs an existing DVIDIA placement envelope, binds a local authored controller to one simulated arm, and reuses it in independently grounded scenes. Native jaw contacts carry a free box; six layouts complete, while no-motion and identical joint commands with open jaws each complete zero of six. Installed execution passes process-level offline checks. Task metadata contains no learned policy; perception, cup geometry, full arm collisions and physical qualification remain open.Offline cable training environmenthttps://research.dvidia.org/papers/offline-cable-training-environment/https://research.dvidia.org/papers/offline-cable-training-environment/Wed, 07 Oct 2026 12:00:00 GMTCPU simulation-only · Controller search and frozen evaluation executed. The first runnable product loop combines an anchored cable environment, local controller-parameter search, frozen evaluation, offline replay and a manifest-backed skill pack. The selected controller and a fixed feedback baseline both completed 10/10 held-out simulated episodes; zero-force and release controls completed 0/10. A native macOS network-denial run completed 3/3 additional episodes. Ideal endpoint attachment and privileged state bound the result; physical rope calibration, gripping, sensors and GPU throughput remain untested.Fast accurate robot training simulationhttps://research.dvidia.org/papers/fast-accurate-robot-training-simulation/https://research.dvidia.org/papers/fast-accurate-robot-training-simulation/Wed, 07 Oct 2026 12:00:00 GMT54 CPU fixture trials · GPU and physical transfer untested. A design for an offline, self-hosted training environment that turns demonstrations into calibrated practice and qualified robot skills. We examine contact, deformation, sensor feedback and the performance–accuracy tradeoff, then report 54 deterministic native MuJoCo CPU fixture trials. Fine endpoint agreement did not establish accurate contact transients, and coarse timesteps missed collisions. Minimum GPU memory, rope accuracy and physical transfer remain unmeasured.Grok robotics research on Xhttps://research.dvidia.org/papers/grok-robotics-research-on-x/https://research.dvidia.org/papers/grok-robotics-research-on-x/Wed, 07 Oct 2026 12:00:00 GMTPrimary sources checked · Experiments proposed. Three Grok 4.7 research passes surface observation-to-action, synthetic-practice and reusable-repair ideas. Independent primary-source checks compare those findings with our thesis: observation supplies goals and state, language retrieves a skill, spatial expectations guide attention, and executable feedback with evaluation establishes competence. X posts remain discovery leads. No universal demonstration count or arbitrary-robot transfer follows from the reviewed results.Spatially grounded downloadable robot skillshttps://research.dvidia.org/papers/spatially-grounded-downloadable-robot-skills/https://research.dvidia.org/papers/spatially-grounded-downloadable-robot-skills/Wed, 07 Oct 2026 12:00:00 GMTSource review · DVIDIA experiments not run. A synthesis of learning from human demonstrations, language-conditioned spatial attention, simulation expansion and robot-specific skill delivery. Spatial priors can accelerate inspection but do not determine contact state or control. The proposed downloadable skill packages a controller, initiation and termination conditions, compatible sensing and hardware, recovery behavior and measured evidence. Separate experiments would test attention, data coverage, feedback and transfer.Clef and grounded video observationshttps://research.dvidia.org/papers/clef-and-grounded-video-observations/https://research.dvidia.org/papers/clef-and-grounded-video-observations/Tue, 06 Oct 2026 12:00:00 GMTFive licensed excerpts · Accuracy evaluation outstanding. A preregistered image-API feasibility run processed five licensed three-second POV excerpts using four timestamped frames, a Gemma caption and a Clef-flash check. All five completed at a price-derived inference estimate of $0.00140634. Post-run inspection found an object error accepted by the checker and omitted state changes. API and schema compatibility were demonstrated; annotation accuracy and whole-task success were not.From Skillspace to robot skillhttps://research.dvidia.org/papers/from-skillspace-to-robot-skill/https://research.dvidia.org/papers/from-skillspace-to-robot-skill/Tue, 06 Oct 2026 12:00:00 GMTGraduation and robot training not run. An Argus review and a proposal to connect original recordings, reviewed episodes, immutable datasets, training runs, held-out evaluation and robot-tested releases. The review separates video annotation from action supervision and defines compatibility beyond a robot’s degree-of-freedom count. A proposed 40-episode annotation pilot and evidence-based graduation criteria remain unrun.Turn physical footage into testable evidencehttps://research.dvidia.org/papers/physical-skill-research-agenda/https://research.dvidia.org/papers/physical-skill-research-agenda/Tue, 06 Oct 2026 12:00:00 GMTFour baseline protocols proposed. A research agenda covering visible capture evidence, relations over time, evaluation provenance and physical grounding. Each question has a bounded protocol and a failure condition. The aim is to measure whether structure and physical signals improve a defined task before scaling collection. Strategic opinions, author-reported results and our untested proposals are kept distinct. +DVIDIA Researchhttps://research.dvidia.org/Research footprints on demonstrations, calibrated simulation and robot skills.enNative grasp task benchmarkhttps://research.dvidia.org/papers/native-grasp-task-benchmark/https://research.dvidia.org/papers/native-grasp-task-benchmark/Sun, 11 Oct 2026 12:00:00 GMT6/6 simulated placements · controls 0/12 · real-time and hardware unqualified. A pinned six-axis arm and coupled gripper complete 6/6 frozen synthetic rigid-block placements, while matched recorded-arm/open-jaw and hold controls complete 0/12. Native contact must produce grasp, sustained lift, retained carry, release and settled placement with tool retreat. All 18 attempts and timing traces remain in the receipt; one placement takes 29.51 wall seconds for 20 simulated seconds and misses 251 of 1,000 computation windows, so real-time operation is not established. An offline 3D mesh replay exposes the recorded motion. The authored privileged-state baseline provides a benchmark for a future compact DVIDIA action policy, not a learned footage-to-action model, validated shoe simulation or physical robot qualification.Hardware matched robot simulationhttps://research.dvidia.org/papers/hardware-matched-robot-simulation/https://research.dvidia.org/papers/hardware-matched-robot-simulation/Sun, 11 Oct 2026 12:00:00 GMT18/18 numerical runs valid · physical fidelity unmeasured. A separate UR5e/Robotiq reference preserves pinned geometry, inertias, moving-link collision proxies and coupled gripper structure: fourteen modeled coordinates and seven commands. Eighteen deterministic unloaded motion/closure runs are numerically valid; at a 1 ms step, CPU physics throughput measures 40.80–47.29 simulated seconds per measured stepping wall second on one host, excluding loading, sensing, rendering, policy and real actuation. Adjacent timestep comparisons and a matched static overlap expose numerical sensitivity and collision participation. A pure-Python paired trace scorer measures joint, tool and jaw mismatch without automatic qualification. Left/right five-finger references are separately audited; LeRobot collection with an original DVIDIA policy is proposed. Serial calibration, physical response, shoe task learning and bimanual execution remain unmeasured. An offline 3D mesh replay now exposes the recorded arm/gripper poses without rerunning physics. PiPER and SO-101 now have pinned CPU profiles with 6/6 valid unloaded numerical attempts each and explicit native command units. Their object-task and physical qualification remain open. YAM, OpenArm, ARX X5 and PAROL6 remain upcoming.DVIDIA Observation Lab pilothttps://research.dvidia.org/papers/dvidia-observation-lab/https://research.dvidia.org/papers/dvidia-observation-lab/Sun, 11 Oct 2026 12:00:00 GMTLocal review release · motion and actuator source pilots · hardware unqualified. DVIDIA Observation Lab supplies local object/palm proposals, reviewed timeline revisions and separate causal image-motion forecasts; real annotation and motion accuracy remain unmeasured. A new CPU pilot learns next-20-ms joint/jaw deltas from current native state and issued commands. Twelve seeded groups retain 36 recordings with 8/2/2 train/development/test grouping; 4,553 test transitions are not independent robot trials. The 6,438-byte action model computes at 0.207 ms p95 on one host and improves RMSE, while jaw MAE/worst error regress against persistence and an initial opening is substantially overpredicted. These are privileged simulator measurements and model-only timings. A proposed shoe capture plan has no footage or executed shoe benchmark. The v0.1 release and hosted synthetic review demo remain separate; no shoe policy, physical adapter, validated shoe physics or hardware qualification is established.Robotics research pipelinehttps://research.dvidia.org/papers/robotics-research-pipeline/https://research.dvidia.org/papers/robotics-research-pipeline/Fri, 09 Oct 2026 12:00:00 GMTArchitecture proposal · current simulation alpha · hardware unqualified. An updated architecture connects immutable Skillspace recordings, reviewed task-label proposals and synchronized measured telemetry to grouped datasets, separate visual and motor learners, and a proposed local state supervisor. The implemented CPU baseline retains authored phases, gripper decisions and recovery around its learned joint-delta head. New annotation and supervisor pilots, fresh frozen confirmation, physics calibration and compatible skill-installation gates define the next milestones. Automatic labeling, learned supervision, human-video action recovery and physical arm qualification remain unestablished; this publication adds no new training or hardware result.Skillspace footage training pipelinehttps://research.dvidia.org/papers/skillspace-footage-training/https://research.dvidia.org/papers/skillspace-footage-training/Thu, 08 Oct 2026 12:00:00 GMTPublic CPU package · one native scene qualified · simulation-only. The public Dvidia-Inference/dvidia-training repository provides a standalone CPU pipeline that audits Skillspace media, freezes grouped splits and exports inspectable artifacts. Ten synthetic six-second clips train a visual predictor with a 0.5106-second median pipeline time across three repeats on one macOS host. Aligned native simulator telemetry supplies 4,029 training labels for a learned joint-delta head, evaluated on one held-out recording. Its capsule passes native placement qualification in that recording's exact scene; identical joint targets with open jaws fail. Installation and hardware guides accompany the package. Human-video action recovery, shoe organization, demonstration sufficiency, independent scene generalization and physical robot competence remain unestablished.Skill capsule acquisition and optimizationhttps://research.dvidia.org/papers/skill-capsule-acquisition/https://research.dvidia.org/papers/skill-capsule-acquisition/Wed, 07 Oct 2026 12:00:00 GMTSimulation-only · teacher/student 7/8 · no overall speedup. A compact learned CPU movement head replaces joint-target computation inside an authored placement supervisor. Teacher and student both pass 7/8 in a new frozen evaluation, while episode wall time increases from 6.273 to 6.360 seconds. A 39,845-byte JSON capsule imports into the local lab, checks its exact runtime and qualifies the current scene through native placement and a recorded-arm/open-jaw control. Scene changes require a fresh test. One known scene and repeat pass native network denial; perception, complete task learning, other embodiments and hardware remain unqualified.Skillspace arm benchmarks and evolutionhttps://research.dvidia.org/papers/skillspace-arm-benchmarks/https://research.dvidia.org/papers/skillspace-arm-benchmarks/Wed, 07 Oct 2026 12:00:00 GMTMixed held-out result · 30/32 baseline, 29/32 updated. Two controller revisions improve development completion from 14/20 to 17/20, but the frozen paired test declines from 30/32 to 29/32: both pass all 24 normal-strength cases while actuator-stress completion falls from 6/8 to 5/8. Instrumented CPU throughput also declines. We publish the regression, unchanged contact and success rules, exact controller snapshots, richer physical inputs, runtime-bound installation and a proposed staged Skillspace contract. Contact forces and penetration remain uncalibrated; learned perception and physical qualification are open.Installing a Skillspace on a simulated armhttps://research.dvidia.org/papers/skillspace-arm-installation/https://research.dvidia.org/papers/skillspace-arm-installation/Wed, 07 Oct 2026 12:00:00 GMTPublic skill link to authored adapter · 6/6 scene placements. The first link-to-arm product path installs an existing DVIDIA placement envelope, binds a local authored controller to one simulated arm, and reuses it in independently grounded scenes. Native jaw contacts carry a free box; six layouts complete, while no-motion and identical joint commands with open jaws each complete zero of six. Installed execution passes process-level offline checks. Task metadata contains no learned policy; perception, cup geometry, full arm collisions and physical qualification remain open.Offline cable training environmenthttps://research.dvidia.org/papers/offline-cable-training-environment/https://research.dvidia.org/papers/offline-cable-training-environment/Wed, 07 Oct 2026 12:00:00 GMTCPU simulation-only · Controller search and frozen evaluation executed. The first runnable product loop combines an anchored cable environment, local controller-parameter search, frozen evaluation, offline replay and a manifest-backed skill pack. The selected controller and a fixed feedback baseline both completed 10/10 held-out simulated episodes; zero-force and release controls completed 0/10. A native macOS network-denial run completed 3/3 additional episodes. Ideal endpoint attachment and privileged state bound the result; physical rope calibration, gripping, sensors and GPU throughput remain untested.Fast accurate robot training simulationhttps://research.dvidia.org/papers/fast-accurate-robot-training-simulation/https://research.dvidia.org/papers/fast-accurate-robot-training-simulation/Wed, 07 Oct 2026 12:00:00 GMT54 CPU fixture trials · GPU and physical transfer untested. A design for an offline, self-hosted training environment that turns demonstrations into calibrated practice and qualified robot skills. We examine contact, deformation, sensor feedback and the performance–accuracy tradeoff, then report 54 deterministic native MuJoCo CPU fixture trials. Fine endpoint agreement did not establish accurate contact transients, and coarse timesteps missed collisions. Minimum GPU memory, rope accuracy and physical transfer remain unmeasured.Grok robotics research on Xhttps://research.dvidia.org/papers/grok-robotics-research-on-x/https://research.dvidia.org/papers/grok-robotics-research-on-x/Wed, 07 Oct 2026 12:00:00 GMTPrimary sources checked · Experiments proposed. Three Grok 4.7 research passes surface observation-to-action, synthetic-practice and reusable-repair ideas. Independent primary-source checks compare those findings with our thesis: observation supplies goals and state, language retrieves a skill, spatial expectations guide attention, and executable feedback with evaluation establishes competence. X posts remain discovery leads. No universal demonstration count or arbitrary-robot transfer follows from the reviewed results.Spatially grounded downloadable robot skillshttps://research.dvidia.org/papers/spatially-grounded-downloadable-robot-skills/https://research.dvidia.org/papers/spatially-grounded-downloadable-robot-skills/Wed, 07 Oct 2026 12:00:00 GMTSource review · DVIDIA experiments not run. A synthesis of learning from human demonstrations, language-conditioned spatial attention, simulation expansion and robot-specific skill delivery. Spatial priors can accelerate inspection but do not determine contact state or control. The proposed downloadable skill packages a controller, initiation and termination conditions, compatible sensing and hardware, recovery behavior and measured evidence. Separate experiments would test attention, data coverage, feedback and transfer.Clef and grounded video observationshttps://research.dvidia.org/papers/clef-and-grounded-video-observations/https://research.dvidia.org/papers/clef-and-grounded-video-observations/Tue, 06 Oct 2026 12:00:00 GMTFive licensed excerpts · Accuracy evaluation outstanding. A preregistered image-API feasibility run processed five licensed three-second POV excerpts using four timestamped frames, a Gemma caption and a Clef-flash check. All five completed at a price-derived inference estimate of $0.00140634. Post-run inspection found an object error accepted by the checker and omitted state changes. API and schema compatibility were demonstrated; annotation accuracy and whole-task success were not.From Skillspace to robot skillhttps://research.dvidia.org/papers/from-skillspace-to-robot-skill/https://research.dvidia.org/papers/from-skillspace-to-robot-skill/Tue, 06 Oct 2026 12:00:00 GMTGraduation and robot training not run. An Argus review and a proposal to connect original recordings, reviewed episodes, immutable datasets, training runs, held-out evaluation and robot-tested releases. The review separates video annotation from action supervision and defines compatibility beyond a robot’s degree-of-freedom count. A proposed 40-episode annotation pilot and evidence-based graduation criteria remain unrun.Turn physical footage into testable evidencehttps://research.dvidia.org/papers/physical-skill-research-agenda/https://research.dvidia.org/papers/physical-skill-research-agenda/Tue, 06 Oct 2026 12:00:00 GMTFour baseline protocols proposed. A research agenda covering visible capture evidence, relations over time, evaluation provenance and physical grounding. Each question has a bounded protocol and a failure condition. The aim is to measure whether structure and physical signals improve a defined task before scaling collection. Strategic opinions, author-reported results and our untested proposals are kept distinct. diff --git a/docs/index.html b/docs/index.html index b1d736f..e3b7eca 100644 --- a/docs/index.html +++ b/docs/index.html @@ -1,4 +1,4 @@ Research journal · DVIDIA Research -

An open research notebook

From observation
to physical skill.

We study how demonstrations become usable evidence, how simulation becomes useful practice, and when a robot skill earns a release.

Published by DVIDIA · hello@dvidia.org

Publications

15 papers and working notes

· Native contact task implementation and authored baseline

Native grasp task benchmark

A pinned six-axis arm and coupled gripper complete 6/6 frozen synthetic rigid-block placements, while matched recorded-arm/open-jaw and hold controls complete 0/12. Native contact must produce grasp, sustained lift, retained carry, release and settled placement with tool retreat. All 18 attempts and timing traces remain in the receipt; one placement takes 29.51 wall seconds for 20 simulated seconds and misses 251 of 1,000 computation windows, so real-time operation is not established. An offline 3D mesh replay exposes the recorded motion. The authored privileged-state baseline provides a benchmark for a future compact DVIDIA action policy, not a learned footage-to-action model, validated shoe simulation or physical robot qualification.

6/6 simulated placements · controls 0/12 · real-time and hardware unqualified

· Pinned mechanics implementation and CPU numerical benchmark

Hardware matched robot simulation

A separate UR5e/Robotiq reference preserves pinned geometry, inertias, moving-link collision proxies and coupled gripper structure: fourteen modeled coordinates and seven commands. Eighteen deterministic unloaded motion/closure runs are numerically valid; at a 1 ms step, CPU physics throughput measures 40.80–47.29 simulated seconds per measured stepping wall second on one host, excluding loading, sensing, rendering, policy and real actuation. Adjacent timestep comparisons and a matched static overlap expose numerical sensitivity and collision participation. A pure-Python paired trace scorer measures joint, tool and jaw mismatch without automatic qualification. Left/right five-finger references are separately audited; LeRobot collection with an original DVIDIA policy is proposed. Serial calibration, physical response, shoe task learning and bimanual execution remain unmeasured. An offline 3D mesh replay now exposes the recorded arm/gripper poses without rerunning physics. YAM, OpenArm, PiPER, ARX X5 and PAROL6 are upcoming profiles; their DVIDIA integration and skill qualification remain unimplemented.

18/18 numerical runs valid · physical fidelity unmeasured

· Local observation workflow and engineering feasibility pilot

DVIDIA Observation Lab pilot

DVIDIA Observation Lab supplies local object/palm proposals, reviewed timeline revisions and separate causal image-motion forecasts; real annotation and motion accuracy remain unmeasured. A new CPU pilot learns next-20-ms joint/jaw deltas from current native state and issued commands. Twelve seeded groups retain 36 recordings with 8/2/2 train/development/test grouping; 4,553 test transitions are not independent robot trials. The 6,438-byte action model computes at 0.207 ms p95 on one host and improves RMSE, while jaw MAE/worst error regress against persistence and an initial opening is substantially overpredicted. These are privileged simulator measurements and model-only timings. A proposed shoe capture plan has no footage or executed shoe benchmark. The v0.1 release and hosted synthetic review demo remain separate; no shoe policy, physical adapter, validated shoe physics or hardware qualification is established.

Local review release · motion and actuator source pilots · hardware unqualified

· Architecture proposal and staged validation roadmap

Robotics research pipeline

An updated architecture connects immutable Skillspace recordings, reviewed task-label proposals and synchronized measured telemetry to grouped datasets, separate visual and motor learners, and a proposed local state supervisor. The implemented CPU baseline retains authored phases, gripper decisions and recovery around its learned joint-delta head. New annotation and supervisor pilots, fresh frozen confirmation, physics calibration and compatible skill-installation gates define the next milestones. Automatic labeling, learned supervision, human-video action recovery and physical arm qualification remain unestablished; this publication adds no new training or hardware result.

Architecture proposal · current simulation alpha · hardware unqualified

· Product implementation and measured synthetic pilot

Skillspace footage training pipeline

The public Dvidia-Inference/dvidia-training repository provides a standalone CPU pipeline that audits Skillspace media, freezes grouped splits and exports inspectable artifacts. Ten synthetic six-second clips train a visual predictor with a 0.5106-second median pipeline time across three repeats on one macOS host. Aligned native simulator telemetry supplies 4,029 training labels for a learned joint-delta head, evaluated on one held-out recording. Its capsule passes native placement qualification in that recording's exact scene; identical joint targets with open jaws fail. Installation and hardware guides accompany the package. Human-video action recovery, shoe organization, demonstration sufficiency, independent scene generalization and physical robot competence remain unestablished.

Public CPU package · one native scene qualified · simulation-only

· Learned movement artifact and installation prototype

Skill capsule acquisition and optimization

A compact learned CPU movement head replaces joint-target computation inside an authored placement supervisor. Teacher and student both pass 7/8 in a new frozen evaluation, while episode wall time increases from 6.273 to 6.360 seconds. A 39,845-byte JSON capsule imports into the local lab, checks its exact runtime and qualifies the current scene through native placement and a recorded-arm/open-jaw control. Scene changes require a fresh test. One known scene and repeat pass native network denial; perception, complete task learning, other embodiments and hardware remain unqualified.

Simulation-only · teacher/student 7/8 · no overall speedup

· Paired simulation benchmark and product assessment

Skillspace arm benchmarks and evolution

Two controller revisions improve development completion from 14/20 to 17/20, but the frozen paired test declines from 30/32 to 29/32: both pass all 24 normal-strength cases while actuator-stress completion falls from 6/8 to 5/8. Instrumented CPU throughput also declines. We publish the regression, unchanged contact and success rules, exact controller snapshots, richer physical inputs, runtime-bound installation and a proposed staged Skillspace contract. Contact forces and penetration remain uncalibrated; learned perception and physical qualification are open.

Mixed held-out result · 30/32 baseline, 29/32 updated

· Product prototype and native contact qualification

Installing a Skillspace on a simulated arm

The first link-to-arm product path installs an existing DVIDIA placement envelope, binds a local authored controller to one simulated arm, and reuses it in independently grounded scenes. Native jaw contacts carry a free box; six layouts complete, while no-motion and identical joint commands with open jaws each complete zero of six. Installed execution passes process-level offline checks. Task metadata contains no learned policy; perception, cup geometry, full arm collisions and physical qualification remain open.

Public skill link to authored adapter · 6/6 scene placements

· Product prototype and local qualification

Offline cable training environment

The first runnable product loop combines an anchored cable environment, local controller-parameter search, frozen evaluation, offline replay and a manifest-backed skill pack. The selected controller and a fixed feedback baseline both completed 10/10 held-out simulated episodes; zero-force and release controls completed 0/10. A native macOS network-denial run completed 3/3 additional episodes. Ideal endpoint attachment and privileged state bound the result; physical rope calibration, gripping, sensors and GPU throughput remain untested.

CPU simulation-only · Controller search and frozen evaluation executed

· Decision report and local experiment

Fast accurate robot training simulation

A design for an offline, self-hosted training environment that turns demonstrations into calibrated practice and qualified robot skills. We examine contact, deformation, sensor feedback and the performance–accuracy tradeoff, then report 54 deterministic native MuJoCo CPU fixture trials. Fine endpoint agreement did not establish accurate contact transients, and coarse timesteps missed collisions. Minimum GPU memory, rope accuracy and physical transfer remain unmeasured.

54 CPU fixture trials · GPU and physical transfer untested

· Source review and thesis comparison

Grok robotics research on X

Three Grok 4.7 research passes surface observation-to-action, synthetic-practice and reusable-repair ideas. Independent primary-source checks compare those findings with our thesis: observation supplies goals and state, language retrieves a skill, spatial expectations guide attention, and executable feedback with evaluation establishes competence. X posts remain discovery leads. No universal demonstration count or arbitrary-robot transfer follows from the reviewed results.

Primary sources checked · Experiments proposed

· Research synthesis and proposed experiments

Spatially grounded downloadable robot skills

A synthesis of learning from human demonstrations, language-conditioned spatial attention, simulation expansion and robot-specific skill delivery. Spatial priors can accelerate inspection but do not determine contact state or control. The proposed downloadable skill packages a controller, initiation and termination conditions, compatible sensing and hardware, recovery behavior and measured evidence. Separate experiments would test attention, data coverage, feedback and transfer.

Source review · DVIDIA experiments not run

· Engineering feasibility pilot

Clef and grounded video observations

A preregistered image-API feasibility run processed five licensed three-second POV excerpts using four timestamped frames, a Gemma caption and a Clef-flash check. All five completed at a price-derived inference estimate of $0.00140634. Post-run inspection found an object error accepted by the checker and omitted state changes. API and schema compatibility were demonstrated; annotation accuracy and whole-task success were not.

Five licensed excerpts · Accuracy evaluation outstanding

· Source review and implementation proposal

From Skillspace to robot skill

An Argus review and a proposal to connect original recordings, reviewed episodes, immutable datasets, training runs, held-out evaluation and robot-tested releases. The review separates video annotation from action supervision and defines compatibility beyond a robot’s degree-of-freedom count. A proposed 40-episode annotation pilot and evidence-based graduation criteria remain unrun.

Graduation and robot training not run

· Research agenda

Turn physical footage into testable evidence

A research agenda covering visible capture evidence, relations over time, evaluation provenance and physical grounding. Each question has a bounded protocol and a failure condition. The aim is to measure whether structure and physical signals improve a defined task before scaling collection. Strategic opinions, author-reported results and our untested proposals are kept distinct.

Four baseline protocols proposed

Research footprints

  1. Give the reference arm a contact task

    A free rigid block must be grasped, lifted, carried, released and left settled at its destination. Frozen scenarios compare the authored controller with identical arm commands and open jaws, and with a hold control. Full traces and 3D replay preserve the evidence for the next compact action-learning experiment. Shoe physics, a learned task policy and hardware transfer remain open.

    Read the native task benchmark · Replay grasp and placement
  2. Build a hardware reference we can measure

    A pinned six-axis arm and coupled gripper expose fourteen coordinates and seven commands. All 18 unloaded numerical runs are valid; 1 ms CPU physics measures 40.80–47.29× elapsed-time throughput with sensing, rendering and policy excluded. A paired trace scorer makes physical mismatch measurable. Five-finger references and LeRobot collection have separate contracts; hardware fidelity and shoe competence remain unmeasured.

    Read the mechanical contract and upcoming arms · Open the 3D replay
  3. Learn a compact command-to-motion predictor

    A 6,438-byte CPU model predicts the next 20 ms of joint/jaw movement from native state and issued commands. Twelve seeded groups retain 36 recordings; 4,553 test transitions are not independent trials. Warm computation measures 0.207 ms p95, but jaw mean absolute/worst errors regress against persistence and the first opening is overpredicted. Shoe collection, hardware sensing and validated shoe physics remain open.

    Read the measured pilot and shoe collection plan
  4. Observe image motion and forecast its path

    A manually seeded CPU source alpha records pixel speed and causal forecasts. Six authored cases process 720 frames in 1.454 seconds; occlusion and low-texture losses remain visible, and a 500 ms reversal forecast misses by nearly 90 pixels. The v0.1 release/demo remain separate; real-scene accuracy, metric speed and live robot control are unqualified.

    Read the source alpha and measured limitations
  5. Publish the robotics research pipeline

    A diagram connects footage review, measured robot state, local training, frozen simulation evaluation and compatible skill installation. Automatic annotation and a learned task supervisor are proposed; their companion validation studies and physical robot qualification remain unrun.

    Read the architecture and validation gates
  6. A public skill link reaches a simulated arm

    One-click local installation binds a DVIDIA placement envelope to an authored contact controller. Six scene layouts pass; identical joint commands with open jaws and no-motion controls each pass 0/6. Physical transfer and learned policies remain open.

    Read the installation contract and arm replay
  7. The offline cable environment is runnable

    A local controller-training loop and simulation-only skill pack now run on CPU. Trained and baseline feedback each pass 10/10 simulated seeds; zero force and early release each pass 0/10. Real gripping, calibration and GPU qualification remain open.

    Read the first product build and replay
  8. Contact accuracy meets runtime

    54 deterministic CPU fixture trials recorded. Endpoint agreement coexists with unqualified force transients; coarse wall tests expose missed contact. No GPU or physical-transfer claim.

    Read the measurements
  9. Compare the observation-to-skill thesis with prior work

    Spatial attention, demonstration coverage, executable simulation and reusable repairs are reviewed separately. Three parallel Grok 4.7 searches support discovery; primary sources support the claims.

    Read the comparison
  10. Five licensed POV excerpts processed

    Ten image calls completed. A checker accepted an object error and captions omitted visible state changes. The separate accuracy pilot remains unrun.

    Read the feasibility run
  11. Define the evidence and release questions

    Four baseline protocols and a Skillspace graduation proposal establish the questions to test. A proposed protocol is distinct from an executed experiment.

    Read the agenda

Open questions

Follow new publications through RSS or inspect the source and revisions on GitHub.

+

An open research notebook

From observation
to physical skill.

We study how demonstrations become usable evidence, how simulation becomes useful practice, and when a robot skill earns a release.

Published by DVIDIA · hello@dvidia.org

Publications

15 papers and working notes

· Native contact task implementation and authored baseline

Native grasp task benchmark

A pinned six-axis arm and coupled gripper complete 6/6 frozen synthetic rigid-block placements, while matched recorded-arm/open-jaw and hold controls complete 0/12. Native contact must produce grasp, sustained lift, retained carry, release and settled placement with tool retreat. All 18 attempts and timing traces remain in the receipt; one placement takes 29.51 wall seconds for 20 simulated seconds and misses 251 of 1,000 computation windows, so real-time operation is not established. An offline 3D mesh replay exposes the recorded motion. The authored privileged-state baseline provides a benchmark for a future compact DVIDIA action policy, not a learned footage-to-action model, validated shoe simulation or physical robot qualification.

6/6 simulated placements · controls 0/12 · real-time and hardware unqualified

· Pinned mechanics implementation and CPU numerical benchmark

Hardware matched robot simulation

A separate UR5e/Robotiq reference preserves pinned geometry, inertias, moving-link collision proxies and coupled gripper structure: fourteen modeled coordinates and seven commands. Eighteen deterministic unloaded motion/closure runs are numerically valid; at a 1 ms step, CPU physics throughput measures 40.80–47.29 simulated seconds per measured stepping wall second on one host, excluding loading, sensing, rendering, policy and real actuation. Adjacent timestep comparisons and a matched static overlap expose numerical sensitivity and collision participation. A pure-Python paired trace scorer measures joint, tool and jaw mismatch without automatic qualification. Left/right five-finger references are separately audited; LeRobot collection with an original DVIDIA policy is proposed. Serial calibration, physical response, shoe task learning and bimanual execution remain unmeasured. An offline 3D mesh replay now exposes the recorded arm/gripper poses without rerunning physics. PiPER and SO-101 now have pinned CPU profiles with 6/6 valid unloaded numerical attempts each and explicit native command units. Their object-task and physical qualification remain open. YAM, OpenArm, ARX X5 and PAROL6 remain upcoming.

18/18 numerical runs valid · physical fidelity unmeasured

· Local observation workflow and engineering feasibility pilot

DVIDIA Observation Lab pilot

DVIDIA Observation Lab supplies local object/palm proposals, reviewed timeline revisions and separate causal image-motion forecasts; real annotation and motion accuracy remain unmeasured. A new CPU pilot learns next-20-ms joint/jaw deltas from current native state and issued commands. Twelve seeded groups retain 36 recordings with 8/2/2 train/development/test grouping; 4,553 test transitions are not independent robot trials. The 6,438-byte action model computes at 0.207 ms p95 on one host and improves RMSE, while jaw MAE/worst error regress against persistence and an initial opening is substantially overpredicted. These are privileged simulator measurements and model-only timings. A proposed shoe capture plan has no footage or executed shoe benchmark. The v0.1 release and hosted synthetic review demo remain separate; no shoe policy, physical adapter, validated shoe physics or hardware qualification is established.

Local review release · motion and actuator source pilots · hardware unqualified

· Architecture proposal and staged validation roadmap

Robotics research pipeline

An updated architecture connects immutable Skillspace recordings, reviewed task-label proposals and synchronized measured telemetry to grouped datasets, separate visual and motor learners, and a proposed local state supervisor. The implemented CPU baseline retains authored phases, gripper decisions and recovery around its learned joint-delta head. New annotation and supervisor pilots, fresh frozen confirmation, physics calibration and compatible skill-installation gates define the next milestones. Automatic labeling, learned supervision, human-video action recovery and physical arm qualification remain unestablished; this publication adds no new training or hardware result.

Architecture proposal · current simulation alpha · hardware unqualified

· Product implementation and measured synthetic pilot

Skillspace footage training pipeline

The public Dvidia-Inference/dvidia-training repository provides a standalone CPU pipeline that audits Skillspace media, freezes grouped splits and exports inspectable artifacts. Ten synthetic six-second clips train a visual predictor with a 0.5106-second median pipeline time across three repeats on one macOS host. Aligned native simulator telemetry supplies 4,029 training labels for a learned joint-delta head, evaluated on one held-out recording. Its capsule passes native placement qualification in that recording's exact scene; identical joint targets with open jaws fail. Installation and hardware guides accompany the package. Human-video action recovery, shoe organization, demonstration sufficiency, independent scene generalization and physical robot competence remain unestablished.

Public CPU package · one native scene qualified · simulation-only

· Learned movement artifact and installation prototype

Skill capsule acquisition and optimization

A compact learned CPU movement head replaces joint-target computation inside an authored placement supervisor. Teacher and student both pass 7/8 in a new frozen evaluation, while episode wall time increases from 6.273 to 6.360 seconds. A 39,845-byte JSON capsule imports into the local lab, checks its exact runtime and qualifies the current scene through native placement and a recorded-arm/open-jaw control. Scene changes require a fresh test. One known scene and repeat pass native network denial; perception, complete task learning, other embodiments and hardware remain unqualified.

Simulation-only · teacher/student 7/8 · no overall speedup

· Paired simulation benchmark and product assessment

Skillspace arm benchmarks and evolution

Two controller revisions improve development completion from 14/20 to 17/20, but the frozen paired test declines from 30/32 to 29/32: both pass all 24 normal-strength cases while actuator-stress completion falls from 6/8 to 5/8. Instrumented CPU throughput also declines. We publish the regression, unchanged contact and success rules, exact controller snapshots, richer physical inputs, runtime-bound installation and a proposed staged Skillspace contract. Contact forces and penetration remain uncalibrated; learned perception and physical qualification are open.

Mixed held-out result · 30/32 baseline, 29/32 updated

· Product prototype and native contact qualification

Installing a Skillspace on a simulated arm

The first link-to-arm product path installs an existing DVIDIA placement envelope, binds a local authored controller to one simulated arm, and reuses it in independently grounded scenes. Native jaw contacts carry a free box; six layouts complete, while no-motion and identical joint commands with open jaws each complete zero of six. Installed execution passes process-level offline checks. Task metadata contains no learned policy; perception, cup geometry, full arm collisions and physical qualification remain open.

Public skill link to authored adapter · 6/6 scene placements

· Product prototype and local qualification

Offline cable training environment

The first runnable product loop combines an anchored cable environment, local controller-parameter search, frozen evaluation, offline replay and a manifest-backed skill pack. The selected controller and a fixed feedback baseline both completed 10/10 held-out simulated episodes; zero-force and release controls completed 0/10. A native macOS network-denial run completed 3/3 additional episodes. Ideal endpoint attachment and privileged state bound the result; physical rope calibration, gripping, sensors and GPU throughput remain untested.

CPU simulation-only · Controller search and frozen evaluation executed

· Decision report and local experiment

Fast accurate robot training simulation

A design for an offline, self-hosted training environment that turns demonstrations into calibrated practice and qualified robot skills. We examine contact, deformation, sensor feedback and the performance–accuracy tradeoff, then report 54 deterministic native MuJoCo CPU fixture trials. Fine endpoint agreement did not establish accurate contact transients, and coarse timesteps missed collisions. Minimum GPU memory, rope accuracy and physical transfer remain unmeasured.

54 CPU fixture trials · GPU and physical transfer untested

· Source review and thesis comparison

Grok robotics research on X

Three Grok 4.7 research passes surface observation-to-action, synthetic-practice and reusable-repair ideas. Independent primary-source checks compare those findings with our thesis: observation supplies goals and state, language retrieves a skill, spatial expectations guide attention, and executable feedback with evaluation establishes competence. X posts remain discovery leads. No universal demonstration count or arbitrary-robot transfer follows from the reviewed results.

Primary sources checked · Experiments proposed

· Research synthesis and proposed experiments

Spatially grounded downloadable robot skills

A synthesis of learning from human demonstrations, language-conditioned spatial attention, simulation expansion and robot-specific skill delivery. Spatial priors can accelerate inspection but do not determine contact state or control. The proposed downloadable skill packages a controller, initiation and termination conditions, compatible sensing and hardware, recovery behavior and measured evidence. Separate experiments would test attention, data coverage, feedback and transfer.

Source review · DVIDIA experiments not run

· Engineering feasibility pilot

Clef and grounded video observations

A preregistered image-API feasibility run processed five licensed three-second POV excerpts using four timestamped frames, a Gemma caption and a Clef-flash check. All five completed at a price-derived inference estimate of $0.00140634. Post-run inspection found an object error accepted by the checker and omitted state changes. API and schema compatibility were demonstrated; annotation accuracy and whole-task success were not.

Five licensed excerpts · Accuracy evaluation outstanding

· Source review and implementation proposal

From Skillspace to robot skill

An Argus review and a proposal to connect original recordings, reviewed episodes, immutable datasets, training runs, held-out evaluation and robot-tested releases. The review separates video annotation from action supervision and defines compatibility beyond a robot’s degree-of-freedom count. A proposed 40-episode annotation pilot and evidence-based graduation criteria remain unrun.

Graduation and robot training not run

· Research agenda

Turn physical footage into testable evidence

A research agenda covering visible capture evidence, relations over time, evaluation provenance and physical grounding. Each question has a bounded protocol and a failure condition. The aim is to measure whether structure and physical signals improve a defined task before scaling collection. Strategic opinions, author-reported results and our untested proposals are kept distinct.

Four baseline protocols proposed

Research footprints

  1. Run more arms on a local CPU

    PiPER and SO-101 join the library with pinned models, explicit arm/gripper units, preserved licenses and complete numerical receipts. Each completes six unloaded motion checks offline. Their task learning and physical qualification remain separate from the UR5e grasp demonstration.

    Read the profile measurements · Open the Robot Simulation Lab
  2. Give the reference arm a contact task

    A free rigid block must be grasped, lifted, carried, released and left settled at its destination. Frozen scenarios compare the authored controller with identical arm commands and open jaws, and with a hold control. Full traces and 3D replay preserve the evidence for the next compact action-learning experiment. Shoe physics, a learned task policy and hardware transfer remain open.

    Read the native task benchmark · Replay grasp and placement
  3. Build a hardware reference we can measure

    A pinned six-axis arm and coupled gripper expose fourteen coordinates and seven commands. All 18 unloaded numerical runs are valid; 1 ms CPU physics measures 40.80–47.29× elapsed-time throughput with sensing, rendering and policy excluded. A paired trace scorer makes physical mismatch measurable. Five-finger references and LeRobot collection have separate contracts; hardware fidelity and shoe competence remain unmeasured.

    Read the mechanical contract and upcoming arms · Open the 3D replay
  4. Learn a compact command-to-motion predictor

    A 6,438-byte CPU model predicts the next 20 ms of joint/jaw movement from native state and issued commands. Twelve seeded groups retain 36 recordings; 4,553 test transitions are not independent trials. Warm computation measures 0.207 ms p95, but jaw mean absolute/worst errors regress against persistence and the first opening is overpredicted. Shoe collection, hardware sensing and validated shoe physics remain open.

    Read the measured pilot and shoe collection plan
  5. Observe image motion and forecast its path

    A manually seeded CPU source alpha records pixel speed and causal forecasts. Six authored cases process 720 frames in 1.454 seconds; occlusion and low-texture losses remain visible, and a 500 ms reversal forecast misses by nearly 90 pixels. The v0.1 release/demo remain separate; real-scene accuracy, metric speed and live robot control are unqualified.

    Read the source alpha and measured limitations
  6. Publish the robotics research pipeline

    A diagram connects footage review, measured robot state, local training, frozen simulation evaluation and compatible skill installation. Automatic annotation and a learned task supervisor are proposed; their companion validation studies and physical robot qualification remain unrun.

    Read the architecture and validation gates
  7. A public skill link reaches a simulated arm

    One-click local installation binds a DVIDIA placement envelope to an authored contact controller. Six scene layouts pass; identical joint commands with open jaws and no-motion controls each pass 0/6. Physical transfer and learned policies remain open.

    Read the installation contract and arm replay
  8. The offline cable environment is runnable

    A local controller-training loop and simulation-only skill pack now run on CPU. Trained and baseline feedback each pass 10/10 simulated seeds; zero force and early release each pass 0/10. Real gripping, calibration and GPU qualification remain open.

    Read the first product build and replay
  9. Contact accuracy meets runtime

    54 deterministic CPU fixture trials recorded. Endpoint agreement coexists with unqualified force transients; coarse wall tests expose missed contact. No GPU or physical-transfer claim.

    Read the measurements
  10. Compare the observation-to-skill thesis with prior work

    Spatial attention, demonstration coverage, executable simulation and reusable repairs are reviewed separately. Three parallel Grok 4.7 searches support discovery; primary sources support the claims.

    Read the comparison
  11. Five licensed POV excerpts processed

    Ten image calls completed. A checker accepted an object error and captions omitted visible state changes. The separate accuracy pilot remains unrun.

    Read the feasibility run
  12. Define the evidence and release questions

    Four baseline protocols and a Skillspace graduation proposal establish the questions to test. A proposed protocol is distinct from an executed experiment.

    Read the agenda

Open questions

Follow new publications through RSS or inspect the source and revisions on GitHub.

diff --git a/docs/papers/hardware-matched-robot-simulation/index.html b/docs/papers/hardware-matched-robot-simulation/index.html index f8fb839..f55e7cd 100644 --- a/docs/papers/hardware-matched-robot-simulation/index.html +++ b/docs/papers/hardware-matched-robot-simulation/index.html @@ -1,7 +1,7 @@ -Hardware matched robot simulation · DVIDIA Research -

Pinned mechanics implementation and CPU numerical benchmark

Hardware matched robot simulation

· DVIDIA

18/18 numerical runs valid · physical fidelity unmeasured

Source & history · Download Markdown

Abstract

A separate UR5e/Robotiq reference preserves pinned geometry, inertias, moving-link collision proxies and coupled gripper structure: fourteen modeled coordinates and seven commands. Eighteen deterministic unloaded motion/closure runs are numerically valid; at a 1 ms step, CPU physics throughput measures 40.80–47.29 simulated seconds per measured stepping wall second on one host, excluding loading, sensing, rendering, policy and real actuation. Adjacent timestep comparisons and a matched static overlap expose numerical sensitivity and collision participation. A pure-Python paired trace scorer measures joint, tool and jaw mismatch without automatic qualification. Left/right five-finger references are separately audited; LeRobot collection with an original DVIDIA policy is proposed. Serial calibration, physical response, shoe task learning and bimanual execution remain unmeasured. An offline 3D mesh replay now exposes the recorded arm/gripper poses without rerunning physics. YAM, OpenArm, PiPER, ARX X5 and PAROL6 are upcoming profiles; their DVIDIA integration and skill qualification remain unimplemented.

In this paper

DVIDIA research · 11 October 2026 · Compiled reference assembly and unloaded CPU diagnostics; physical transfer unmeasured

+Hardware matched robot simulation · DVIDIA Research +

Pinned mechanics implementation and CPU numerical benchmark

Hardware matched robot simulation

· DVIDIA

18/18 numerical runs valid · physical fidelity unmeasured

Source & history · Download Markdown

Abstract

A separate UR5e/Robotiq reference preserves pinned geometry, inertias, moving-link collision proxies and coupled gripper structure: fourteen modeled coordinates and seven commands. Eighteen deterministic unloaded motion/closure runs are numerically valid; at a 1 ms step, CPU physics throughput measures 40.80–47.29 simulated seconds per measured stepping wall second on one host, excluding loading, sensing, rendering, policy and real actuation. Adjacent timestep comparisons and a matched static overlap expose numerical sensitivity and collision participation. A pure-Python paired trace scorer measures joint, tool and jaw mismatch without automatic qualification. Left/right five-finger references are separately audited; LeRobot collection with an original DVIDIA policy is proposed. Serial calibration, physical response, shoe task learning and bimanual execution remain unmeasured. An offline 3D mesh replay now exposes the recorded arm/gripper poses without rerunning physics. PiPER and SO-101 now have pinned CPU profiles with 6/6 valid unloaded numerical attempts each and explicit native command units. Their object-task and physical qualification remain open. YAM, OpenArm, ARX X5 and PAROL6 remain upcoming.

In this paper

DVIDIA research · 11 October 2026 · Compiled reference assembly and unloaded CPU diagnostics; physical transfer unmeasured

Build the first simulator around a specific six-axis arm and gripper, then measure the physical response that the learner will encounter. DVIDIA's vision remains a Skillspace that supplies demonstrations, local training and an installable skill. Rendered images support visual learning, but matching a room's appearance and the arm's number of joints does not establish that a learned action will produce the same motion, grip or outcome on hardware. We have compiled a pinned UR5e and Robotiq 2F-85 reference and run eighteen unloaded CPU diagnostics; this does not select the user's eventual purchase or reproduce that robot's controller. Later, two arms with five-finger hands need a different command, sensing and contact contract. Imported structure, numerical characterization and held-out physical agreement remain separate evidence states. Useful speed is the cost of acquiring a qualified capability, measured alongside errors and failures.

Six arm joints do not specify the whole robot

DVIDIA's existing dvidia-authored-6dof-parallel-jaw-v0 environment is an authored rigid-box contact experiment. It has six arm hinges, two simulated sliding jaws and eight native position actuators, exposed through six joint targets and one total-gap target. Arm links, base and palm have collision masks disabled; the experiment therefore does not evaluate whole-arm avoidance or self-collision. Exact object state, contact counts and pad forces are native simulator observations, not measured camera or fingertip signals. Preserve this profile and its frozen results while adding a separately named hardware reference. (Current arm source)

@@ -15,10 +15,10 @@

Inspect the arm and gri

The native mesh replay at 1.10 seconds in the recorded arm sweep. Eight browser acceptance groups passed across desktop, phone and landscape layouts; this is display validation, not physical-fidelity or rendering-speed evidence. Browser checks.

The exporter verifies the frozen receipt and pinned assets, reconstructs recorded arm/linkage poses with native forward kinematics and checks tool/pad/link-center agreement before writing display transforms. It does not step physics or rerun the benchmark. Mesh vertices and transforms use float32 display storage; exact repeated vertices are deduplicated without changing triangle coordinates. Original numerical files remain unchanged and downloadable under the replay's receipt/ directory. The viewer is bundled locally with retained asset and Three.js notices; it requires browser WebGL2 and reports loading/graphics failures explicitly. No new rendering-speed, task-success or physical-fidelity claim follows from this display. (Export and offline guide, Original frozen report)

The ready-made browser replay can be viewed on another supported browser without MuJoCo. Reconstructing an original numerical receipt requires its compiled-model hash to match; native builds can differ between platforms. When they differ, generate and retain a separate benchmark and replay on the current machine alongside the original results.

-

Upcoming hardware profiles

-

The following support list is a plan, not a purchase decision or implemented compatibility. All five additions remain upcoming. Each needs a pinned simulator, physical adapter and benchmark evidence tied to its exact hardware/controller revision. Arm-axis counts exclude the gripper; a future five-finger hand has a separate actuation and sensing contract.

-
PriorityTargetIntended role and remaining work
1YAM — 6 arm axes + gripperFirst candidate for a full-size shoe grasp-and-place benchmark. MIT control/models include MuJoCo and teleoperation; complete open manufacturing designs are not established. Pin arm/gripper revisions and measure their physical response. Official SDK/models.
2OpenArm — 7 arm axes + gripper per sideLater two-arm coordination. CERN-OHL-S-2.0 hardware and Apache-2.0 simulation; native LeRobot connection currently requires Linux. Keep v1/v2 profiles distinct and assess hand mounts, mass, power and remaining payload. Hardware, MuJoCo, LeRobot.
3AgileX PiPER — 6 arm axes + gripperAlternative six-axis research/community target. Open SDK and MIT Menagerie model; full manufacturing openness is not established. Bind firmware, offsets, control modes and jaw calibration. SDK, model.
4ARX X5 — 6 arm axes + gripperJoint position/velocity/torque research adapter and later multiple-arm control. MIT research SDK and community LeRobot plugin; exact X5 simulation and hardware licensing need qualification. SDK, plugin.
5PAROL6 — 6 arm axes + separate gripperCommunity-build desktop target. GPLv3 mechanics/software and ROS/Python tools; the documented open-loop version has limit-switch sensing, so actual joint feedback requires verification. Complete electronics openness is not established. Project, specifications.
-

SO-101 remains the accessible planned entry profile, with five arm axes plus a gripper. The current UR5e/Robotiq reference is separately characterized. Gripper opening, finger geometry, slip, workspace and usable payload drive the first shoe hardware choice. Device support advances separately from learned skill and physical qualification. (DVIDIA hardware support map)

+

Hardware integration roadmap

+

The following list separates runnable numerical simulation from planned integration. PiPER now has a pinned CPU simulation profile; YAM, OpenArm, ARX X5 and PAROL6 remain upcoming. All five still need their own object-task qualification and physical adapter evidence tied to the exact hardware/controller revision. Arm-axis counts exclude the gripper; a future five-finger hand has a separate actuation and sensing contract.

+
PriorityTargetIntended role and remaining work
1YAM — 6 arm axes + gripperFirst candidate for a full-size shoe grasp-and-place benchmark. MIT control/models include MuJoCo and teleoperation; complete open manufacturing designs are not established. Pin arm/gripper revisions and measure their physical response. Official SDK/models.
2OpenArm — 7 arm axes + gripper per sideLater two-arm coordination. CERN-OHL-S-2.0 hardware and Apache-2.0 simulation; native LeRobot connection currently requires Linux. Keep v1/v2 profiles distinct and assess hand mounts, mass, power and remaining payload. Hardware, MuJoCo, LeRobot.
3AgileX PiPER — 6 arm axes + gripperRunnable pinned CPU profile with unloaded numerical receipts; object-task and physical qualification remain open. Open SDK and MIT Menagerie model; full manufacturing openness is not established. Bind firmware, offsets, control modes and jaw calibration for a physical adapter. SDK, model.
4ARX X5 — 6 arm axes + gripperJoint position/velocity/torque research adapter and later multiple-arm control. MIT research SDK and community LeRobot plugin; exact X5 simulation and hardware licensing need qualification. SDK, plugin.
5PAROL6 — 6 arm axes + separate gripperCommunity-build desktop target. GPLv3 mechanics/software and ROS/Python tools; the documented open-loop version has limit-switch sensing, so actual joint feedback requires verification. Complete electronics openness is not established. Project, specifications.
+

SO-101 is the accessible entry profile, with five arm axes plus a gripper. Its pinned CPU simulation is now runnable; its object-task and physical adapter qualification remain open. The UR5e/Robotiq reference is separately characterized. Gripper opening, finger geometry, slip, workspace and usable payload drive the first shoe hardware choice. Device support advances separately from learned skill and physical qualification. (DVIDIA hardware support map)

Match observable mechanics before calling it fidelity

The first physical calibration should fit base and tool transforms and serial-specific kinematics on measured poses, then score untouched poses throughout the intended tabletop workspace. Universal Robots warns that using generic calibration can introduce end-effector discrepancies of centimeters. Report independently measured tool-position and orientation residuals, reference uncertainty, operating conditions and the calibration revision. Joint encoder agreement alone cannot reveal a wrong base or tool transform. (UR calibration documentation)

Next, record issued commands and actual joint motion during bounded unloaded trajectories, then repeat under declared payloads. Compare onset delay, tracking residuals, overshoot and settling. UR's RTDE interface supplies timestamps, target and actual joint states, TCP data and safety status; its e-series output can reach 500 Hz, while busy controllers can skip packages. That interface rate does not establish a 2 ms application reaction time. The servoj command's gain and lookahead change response, so the reference model's affine servos cannot silently stand in for every controller setting. (RTDE guide, Official servoj manual)

@@ -36,7 +36,7 @@

Offline traini

Use fixed issued-command replays to expose timestep sensitivity, then measure complete task attempts with resets, observations and failures charged to the cost. A physics-only benchmark supports a narrow simulator-speed statement. A physical skill claim additionally needs fresh hardware trials, independently observable success, critical-error and recovery criteria and a declared operating envelope. No hardware minimum or transfer-success target can be justified from rendered-image count alone.

LeRobot can collect evidence while DVIDIA owns the model

LeRobot offers hardware/teleoperation interfaces, cameras and dataset tooling without requiring a supplied policy architecture. Its official custom-policy interface supports independently packaged models with training, reset and action-selection methods. DVIDIA can reuse that infrastructure while developing its own objectives, model and skill artifacts. No DVIDIA/LeRobot integration or robot action policy is built by this reference release. The earlier compact transition predictor forecasts motion conditional on a command; it does not choose the command. A proposed first seam is a validated local dataset reader and strict observation/action adapter, followed by an original DVIDIA trainer. (Custom hardware guide, Custom policy guide)

-

Hardware selection remains explicit. SO-101's six motor channels are shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll and gripper: five arm axes plus a gripper, not six arm axes plus a gripper. Its body positions can use degrees or normalized values, and the jaw uses a normalized command. A separate named profile must specify ordering, units, calibration, action representation and requested versus actually sent targets; it cannot accept DVIDIA's current six-axis/radian artifact unchanged. (SO-101 guide, Follower implementation, Action representations)

+

Hardware selection remains explicit. SO-101's six motor channels are shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll and gripper: five arm axes plus a gripper, not six arm axes plus a gripper. LeRobot's physical follower interface can use degrees or normalized body positions, and its jaw uses a normalized command. A physical adapter must specify ordering, units, calibration, action representation and requested versus actually sent targets; it cannot accept DVIDIA's six-axis reference artifact or the native SO-101 profile's radian targets unchanged. (SO-101 guide, Follower implementation, Action representations)

LeRobotDataset v3 stores state/action rows, camera video and episode/schema metadata, and recordings can stay local. However, the recording guide warns that frame-index/FPS timestamps can make motion appear faster when acquisition misses the requested cadence. DVIDIA needs a sidecar with actual monotonic acquisition, arrival, command and feedback clocks, camera calibration, missed samples, controller settings and physical source groups. Nominal timestamps alone cannot fit actuator delay or validate physical speed. Pin and test the chosen LeRobot version and review datasets/assets separately from its Apache-2.0 software license. (Dataset format, Recording and cadence guide, Software license)

Eighteen unloaded CPU runs characterize this reference

The new native benchmark attempts three synthetic cases—hold-open, base-joint sweep and jaw cycle—at 2, 1 and 0.5 ms timesteps, with two deterministic repeats and four simulated seconds per attempt. All 18 attempts are numerically valid with no captured warnings. Commands update at 20 ms boundaries. The measured platform is macOS 26.5.2 arm64, Python 3.12.9 and MuJoCo 3.15.0; these runs used CPU physics. They contain no task object, rendering, perception, learned policy or hardware actuation. Repeats are repeated numerical/performance diagnostics, not independent physical trials. (Interactive report, Exact receipt)

@@ -45,6 +45,13 @@

Eighteen unloade

Adjacent-timestep comparisons align 201 elapsed boundaries per comparison, including the post-settling start. For repeat index 0, halving the timestep reduces arm-sweep tool-translation and jaw-cycle pad-center differences as shown below. Pad-center separation is not calibrated inner jaw gap. These residuals compare two numerical resolutions of the same model, not the model with reality. Both deterministic repeats are retained in the receipt. (Numerical comparisons)

Case and measured quantityTimesteps comparedRMSEWorst difference
Arm sweep, reference tool translation2 versus 1 ms0.07424 mm0.11504 mm
Arm sweep, reference tool translation1 versus 0.5 ms0.03702 mm0.05738 mm
Jaw cycle, pad-center separation2 versus 1 ms0.03448 mm0.05043 mm
Jaw cycle, pad-center separation1 versus 0.5 ms0.01733 mm0.02538 mm

A separate authored sphere-overlap diagnostic produces one arm contact with active collision masks and zero with those masks disabled. It checks that the selected proxy participates in collision detection; it is not a moving collision-response or avoidance test. The software checkpoint also passed 488 local tests, including native contract checks. CPU-only headless characterization is demonstrated for this unloaded reference workload; shoe contact, complete training throughput, minimum hardware, physical accuracy and the composed two-arm system remain unmeasured. (Collision receipt, Reference instructions)

+

PiPER and SO-101 now have runnable CPU profiles

+

The library now includes pinned PiPER (six arm axes plus one gripper command) and SO-101 (five arm axes plus one gripper command) profiles. Both expose explicit model fetch, offline asset verification, native audit and bounded numerical benchmarking. The packaged models preserve upstream geometry and mechanics; DVIDIA supplies the profile tooling and evidence contracts. PiPER's slider command is single-finger travel in metres with a coupled opposite finger. SO-101's moving jaw uses a hinge target in radians. Neither is a calibrated physical gap or force command, and SO-101 must not be labeled as a six-axis arm. (Local simulation guide, Pinned model contracts)

+

A serial run with process-level network denial retained six attempts per profile: hold, small arm sweep and gripper cycle, each repeated twice for two simulated seconds at a 1 ms physics step and 20 ms command boundary. These repeats check numerical mechanics; they are not independent task scenes or a learned-policy evaluation. Both profiles produced 6/6 numerically valid attempts. Focused tests separately confirm that commanded arm and jaw cases change the native state compared with holding the starting commands.

+
ProfileSimulated seconds / complete-attempt wall secondPer-attempt 20 ms computation p95Computation windows above 20 ms
PiPER4.38–14.62×0.446–0.500 ms0/600
SO-10114.73–20.17×0.396–0.744 ms0/600
+

All attempts remain in the Robot Simulation Lab, including PiPER's slower first attempt. Wall time includes asset checks, compilation and recorded state construction; output serialization, browser rendering, physical sensing/transport and learned inference are outside that measurement. Runtime was macOS/Darwin arm64 with Python 3.12.9, MuJoCo 3.15.0 and NumPy 2.5.3. These CPU runs used no discrete-GPU workload; host load was not controlled, and universal minimum memory or real-time hardware behavior is not established. The exact bundles include source/runtime identities, native controls and effort units, gravity settings, every recorded trajectory and original licenses. (PiPER receipt, SO-101 receipt)

+

PiPER simulation assets are MIT; SO-101 assets are Apache-2.0. The latter has published build/design files and LeRobot guidance, while an open PiPER SDK does not by itself establish open manufacturing designs. The models' original notices are retained, and their attribution is separate from DVIDIA software. (SO-101 design, LeRobot guide, PiPER SDK)

+

These additions are ready for local unloaded simulation. Each still needs its own object-task environment, grasp/release controls, calibrated observations and fresh closed-loop policy benchmark. YAM, OpenArm, ARX X5 and PAROL6 remain further integration targets; bimanual and dexterous manipulation retain separate contracts. The hosted lab displays the UR5e contact replay separately from these numerical receipts, so additional model support cannot be mistaken for inherited grasp competence.

Release an evidence-bearing skill, not an appearance match

The installable unit should carry the policy, skill goal, observations/actions and units, named embodiment ordering, adapter, dependency hashes, calibration requirements, supported conditions, refusal/recovery behavior and benchmark receipts. Imported asset structure, characterized simulation and measured hardware agreement advance separately. A reference-only or simulation-qualified package remains useful for research while declaring physical execution unqualified. This follows the current DVIDIA roadmap's separation of collection, learning, simulation and physical qualification. (Execution map)

The product opportunity is an efficient route from missing evidence to a tested capability. A hardware-matched reference gives that route a concrete target; measured residuals identify what to fix or collect next. When simulation succeeds and hardware fails, the discrepancy becomes a reproducible investigation rather than a reason to gather undirected images. That feedback loop is how Skillspace can accumulate reusable knowledge across skills without pretending that one visual dataset or one embodiment solves every interaction.

diff --git a/docs/papers/native-grasp-task-benchmark/index.html b/docs/papers/native-grasp-task-benchmark/index.html index e32c2bb..faa8ac9 100644 --- a/docs/papers/native-grasp-task-benchmark/index.html +++ b/docs/papers/native-grasp-task-benchmark/index.html @@ -33,4 +33,5 @@

Conclusion

The product opportunity is the loop connecting reviewed skill-space evidence, bounded local learning, frozen task receipts and compatible deployment. Each new skill can inherit that loop while supplying its own task conditions and failures. That is a concrete route toward fast downloadable robot behavior, with uncertainty exposed early enough to improve the data and simulation instead of hiding it in the replay.

Run the benchmark using the installation and evaluation guide. The implementation and review are available in GitHub PR #11.

Published by DVIDIA · hello@dvidia.org · @iammrriver on X

+

The public Robot Simulation Lab also provides separate unloaded PiPER and SO-101 receipts and installation instructions. Browser acceptance evidence checks the recorded task replay on desktop and narrow screens.

diff --git a/docs/publications.json b/docs/publications.json index a308773..9e972d7 100644 --- a/docs/publications.json +++ b/docs/publications.json @@ -31,7 +31,7 @@ "evaluation-provenance", "skill-relations" ], - "abstract": "A separate UR5e/Robotiq reference preserves pinned geometry, inertias, moving-link collision proxies and coupled gripper structure: fourteen modeled coordinates and seven commands. Eighteen deterministic unloaded motion/closure runs are numerically valid; at a 1 ms step, CPU physics throughput measures 40.80–47.29 simulated seconds per measured stepping wall second on one host, excluding loading, sensing, rendering, policy and real actuation. Adjacent timestep comparisons and a matched static overlap expose numerical sensitivity and collision participation. A pure-Python paired trace scorer measures joint, tool and jaw mismatch without automatic qualification. Left/right five-finger references are separately audited; LeRobot collection with an original DVIDIA policy is proposed. Serial calibration, physical response, shoe task learning and bimanual execution remain unmeasured. An offline 3D mesh replay now exposes the recorded arm/gripper poses without rerunning physics. YAM, OpenArm, PiPER, ARX X5 and PAROL6 are upcoming profiles; their DVIDIA integration and skill qualification remain unimplemented." + "abstract": "A separate UR5e/Robotiq reference preserves pinned geometry, inertias, moving-link collision proxies and coupled gripper structure: fourteen modeled coordinates and seven commands. Eighteen deterministic unloaded motion/closure runs are numerically valid; at a 1 ms step, CPU physics throughput measures 40.80–47.29 simulated seconds per measured stepping wall second on one host, excluding loading, sensing, rendering, policy and real actuation. Adjacent timestep comparisons and a matched static overlap expose numerical sensitivity and collision participation. A pure-Python paired trace scorer measures joint, tool and jaw mismatch without automatic qualification. Left/right five-finger references are separately audited; LeRobot collection with an original DVIDIA policy is proposed. Serial calibration, physical response, shoe task learning and bimanual execution remain unmeasured. An offline 3D mesh replay now exposes the recorded arm/gripper poses without rerunning physics. PiPER and SO-101 now have pinned CPU profiles with 6/6 valid unloaded numerical attempts each and explicit native command units. Their object-task and physical qualification remain open. YAM, OpenArm, ARX X5 and PAROL6 remain upcoming." }, { "slug": "dvidia-observation-lab", diff --git a/publications.json b/publications.json index 84737d5..a760910 100644 --- a/publications.json +++ b/publications.json @@ -31,7 +31,7 @@ "evaluation-provenance", "skill-relations" ], - "abstract": "A separate UR5e/Robotiq reference preserves pinned geometry, inertias, moving-link collision proxies and coupled gripper structure: fourteen modeled coordinates and seven commands. Eighteen deterministic unloaded motion/closure runs are numerically valid; at a 1 ms step, CPU physics throughput measures 40.80\u201347.29 simulated seconds per measured stepping wall second on one host, excluding loading, sensing, rendering, policy and real actuation. Adjacent timestep comparisons and a matched static overlap expose numerical sensitivity and collision participation. A pure-Python paired trace scorer measures joint, tool and jaw mismatch without automatic qualification. Left/right five-finger references are separately audited; LeRobot collection with an original DVIDIA policy is proposed. Serial calibration, physical response, shoe task learning and bimanual execution remain unmeasured. An offline 3D mesh replay now exposes the recorded arm/gripper poses without rerunning physics. YAM, OpenArm, PiPER, ARX X5 and PAROL6 are upcoming profiles; their DVIDIA integration and skill qualification remain unimplemented." + "abstract": "A separate UR5e/Robotiq reference preserves pinned geometry, inertias, moving-link collision proxies and coupled gripper structure: fourteen modeled coordinates and seven commands. Eighteen deterministic unloaded motion/closure runs are numerically valid; at a 1 ms step, CPU physics throughput measures 40.80\u201347.29 simulated seconds per measured stepping wall second on one host, excluding loading, sensing, rendering, policy and real actuation. Adjacent timestep comparisons and a matched static overlap expose numerical sensitivity and collision participation. A pure-Python paired trace scorer measures joint, tool and jaw mismatch without automatic qualification. Left/right five-finger references are separately audited; LeRobot collection with an original DVIDIA policy is proposed. Serial calibration, physical response, shoe task learning and bimanual execution remain unmeasured. An offline 3D mesh replay now exposes the recorded arm/gripper poses without rerunning physics. PiPER and SO-101 now have pinned CPU profiles with 6/6 valid unloaded numerical attempts each and explicit native command units. Their object-task and physical qualification remain open. YAM, OpenArm, ARX X5 and PAROL6 remain upcoming." }, { "slug": "dvidia-observation-lab", diff --git a/reports/Hardware matched robot simulation.md b/reports/Hardware matched robot simulation.md index 4631697..a5eba5e 100644 --- a/reports/Hardware matched robot simulation.md +++ b/reports/Hardware matched robot simulation.md @@ -28,19 +28,19 @@ The exporter verifies the frozen receipt and pinned assets, reconstructs recorde The ready-made browser replay can be viewed on another supported browser without MuJoCo. Reconstructing an original numerical receipt requires its compiled-model hash to match; native builds can differ between platforms. When they differ, generate and retain a separate benchmark and replay on the current machine alongside the original results. -## Upcoming hardware profiles +## Hardware integration roadmap -The following support list is a plan, not a purchase decision or implemented compatibility. **All five additions remain upcoming.** Each needs a pinned simulator, physical adapter and benchmark evidence tied to its exact hardware/controller revision. Arm-axis counts exclude the gripper; a future five-finger hand has a separate actuation and sensing contract. +The following list separates runnable numerical simulation from planned integration. **PiPER now has a pinned CPU simulation profile; YAM, OpenArm, ARX X5 and PAROL6 remain upcoming.** All five still need their own object-task qualification and physical adapter evidence tied to the exact hardware/controller revision. Arm-axis counts exclude the gripper; a future five-finger hand has a separate actuation and sensing contract. | Priority | Target | Intended role and remaining work | | --- | --- | --- | | 1 | **YAM — 6 arm axes + gripper** | First candidate for a full-size shoe grasp-and-place benchmark. MIT control/models include MuJoCo and teleoperation; complete open manufacturing designs are not established. Pin arm/gripper revisions and measure their physical response. [Official SDK/models](https://github.com/i2rt-robotics/i2rt). | | 2 | **OpenArm — 7 arm axes + gripper per side** | Later two-arm coordination. CERN-OHL-S-2.0 hardware and Apache-2.0 simulation; native LeRobot connection currently requires Linux. Keep v1/v2 profiles distinct and assess hand mounts, mass, power and remaining payload. [Hardware](https://github.com/enactic/openarm_hardware), [MuJoCo](https://github.com/enactic/openarm_mujoco), [LeRobot](https://huggingface.co/docs/lerobot/main/openarm). | -| 3 | **AgileX PiPER — 6 arm axes + gripper** | Alternative six-axis research/community target. Open SDK and MIT Menagerie model; full manufacturing openness is not established. Bind firmware, offsets, control modes and jaw calibration. [SDK](https://github.com/agilexrobotics/piper_sdk), [model](https://github.com/google-deepmind/mujoco_menagerie/tree/main/agilex_piper). | +| 3 | **AgileX PiPER — 6 arm axes + gripper** | Runnable pinned CPU profile with unloaded numerical receipts; object-task and physical qualification remain open. Open SDK and MIT Menagerie model; full manufacturing openness is not established. Bind firmware, offsets, control modes and jaw calibration for a physical adapter. [SDK](https://github.com/agilexrobotics/piper_sdk), [model](https://github.com/google-deepmind/mujoco_menagerie/tree/main/agilex_piper). | | 4 | **ARX X5 — 6 arm axes + gripper** | Joint position/velocity/torque research adapter and later multiple-arm control. MIT research SDK and community LeRobot plugin; exact X5 simulation and hardware licensing need qualification. [SDK](https://github.com/real-stanford/arx5-sdk), [plugin](https://github.com/villekuosmanen/lerobot-arx5). | | 5 | **PAROL6 — 6 arm axes + separate gripper** | Community-build desktop target. GPLv3 mechanics/software and ROS/Python tools; the documented open-loop version has limit-switch sensing, so actual joint feedback requires verification. Complete electronics openness is not established. [Project](https://github.com/Source-Robotics/PAROL6-Desktop-robot-arm), [specifications](https://source-robotics.github.io/PAROL-docs/page2_2/). | -SO-101 remains the accessible planned entry profile, with five arm axes plus a gripper. The current UR5e/Robotiq reference is separately characterized. Gripper opening, finger geometry, slip, workspace and usable payload drive the first shoe hardware choice. Device support advances separately from learned skill and physical qualification. ([DVIDIA hardware support map](https://github.com/Dvidia-Inference/dvidia-training/blob/93fc549a370cdd448b49380c5beff3b2641c2795/docs/hardware-roadmap.md)) +SO-101 is the accessible entry profile, with five arm axes plus a gripper. Its pinned CPU simulation is now runnable; its object-task and physical adapter qualification remain open. The UR5e/Robotiq reference is separately characterized. Gripper opening, finger geometry, slip, workspace and usable payload drive the first shoe hardware choice. Device support advances separately from learned skill and physical qualification. ([DVIDIA hardware support map](https://github.com/Dvidia-Inference/dvidia-training/blob/928fed79eca09a1a5081aae7720ab78c2e5b817c/docs/hardware-roadmap.md)) ## Match observable mechanics before calling it fidelity @@ -76,7 +76,7 @@ Use fixed issued-command replays to expose timestep sensitivity, then measure co LeRobot offers hardware/teleoperation interfaces, cameras and dataset tooling without requiring a supplied policy architecture. Its official custom-policy interface supports independently packaged models with training, reset and action-selection methods. DVIDIA can reuse that infrastructure while developing its own objectives, model and skill artifacts. **No DVIDIA/LeRobot integration or robot action policy is built by this reference release.** The earlier compact transition predictor forecasts motion conditional on a command; it does not choose the command. A proposed first seam is a validated local dataset reader and strict observation/action adapter, followed by an original DVIDIA trainer. ([Custom hardware guide](https://huggingface.co/docs/lerobot/main/integrate_hardware), [Custom policy guide](https://huggingface.co/docs/lerobot/main/bring_your_own_policies)) -Hardware selection remains explicit. SO-101's six motor channels are shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll and gripper: **five arm axes plus a gripper, not six arm axes plus a gripper**. Its body positions can use degrees or normalized values, and the jaw uses a normalized command. A separate named profile must specify ordering, units, calibration, action representation and requested versus actually sent targets; it cannot accept DVIDIA's current six-axis/radian artifact unchanged. ([SO-101 guide](https://huggingface.co/docs/lerobot/main/so101), [Follower implementation](https://github.com/huggingface/lerobot/blob/main/src/lerobot/robots/so_follower/so_follower.py), [Action representations](https://huggingface.co/docs/lerobot/main/action_representations)) +Hardware selection remains explicit. SO-101's six motor channels are shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll and gripper: **five arm axes plus a gripper, not six arm axes plus a gripper**. LeRobot's physical follower interface can use degrees or normalized body positions, and its jaw uses a normalized command. A physical adapter must specify ordering, units, calibration, action representation and requested versus actually sent targets; it cannot accept DVIDIA's six-axis reference artifact or the native SO-101 profile's radian targets unchanged. ([SO-101 guide](https://huggingface.co/docs/lerobot/main/so101), [Follower implementation](https://github.com/huggingface/lerobot/blob/main/src/lerobot/robots/so_follower/so_follower.py), [Action representations](https://huggingface.co/docs/lerobot/main/action_representations)) LeRobotDataset v3 stores state/action rows, camera video and episode/schema metadata, and recordings can stay local. However, the recording guide warns that frame-index/FPS timestamps can make motion appear faster when acquisition misses the requested cadence. DVIDIA needs a sidecar with **actual monotonic acquisition, arrival, command and feedback clocks**, camera calibration, missed samples, controller settings and physical source groups. Nominal timestamps alone cannot fit actuator delay or validate physical speed. Pin and test the chosen LeRobot version and review datasets/assets separately from its Apache-2.0 software license. ([Dataset format](https://huggingface.co/docs/lerobot/main/lerobot-dataset-v3), [Recording and cadence guide](https://huggingface.co/docs/lerobot/main/il_robots), [Software license](https://github.com/huggingface/lerobot/blob/main/LICENSE)) @@ -106,6 +106,23 @@ Adjacent-timestep comparisons align 201 elapsed boundaries per comparison, inclu A separate authored sphere-overlap diagnostic produces one arm contact with active collision masks and zero with those masks disabled. It checks that the selected proxy participates in collision detection; it is not a moving collision-response or avoidance test. The software checkpoint also passed 488 local tests, including native contract checks. CPU-only headless characterization is demonstrated for this unloaded reference workload; shoe contact, complete training throughput, minimum hardware, physical accuracy and the composed two-arm system remain unmeasured. ([Collision receipt](https://research.dvidia.org/downloads/topics/physical-grounding/runs/2026-10-11-hardware-reference/report.json), [Reference instructions](https://github.com/Dvidia-Inference/dvidia-training/blob/c95a4385dae4b654b01ef75f56cbcc144fde2f5c/docs/hardware-reference.md)) +## PiPER and SO-101 now have runnable CPU profiles + +The library now includes pinned **PiPER (six arm axes plus one gripper command)** and **SO-101 (five arm axes plus one gripper command)** profiles. Both expose explicit model fetch, offline asset verification, native audit and bounded numerical benchmarking. The packaged models preserve upstream geometry and mechanics; DVIDIA supplies the profile tooling and evidence contracts. PiPER's slider command is single-finger travel in metres with a coupled opposite finger. SO-101's moving jaw uses a hinge target in radians. Neither is a calibrated physical gap or force command, and SO-101 must not be labeled as a six-axis arm. ([Local simulation guide](https://github.com/Dvidia-Inference/dvidia-training/blob/928fed79eca09a1a5081aae7720ab78c2e5b817c/docs/robot-simulation-library.md), [Pinned model contracts](https://github.com/Dvidia-Inference/dvidia-training/blob/928fed79eca09a1a5081aae7720ab78c2e5b817c/src/dvidia_training/robot_profiles.lock.json)) + +A serial run with process-level network denial retained six attempts per profile: hold, small arm sweep and gripper cycle, each repeated twice for two simulated seconds at a 1 ms physics step and 20 ms command boundary. These repeats check numerical mechanics; they are not independent task scenes or a learned-policy evaluation. Both profiles produced **6/6 numerically valid attempts**. Focused tests separately confirm that commanded arm and jaw cases change the native state compared with holding the starting commands. + +| Profile | Simulated seconds / complete-attempt wall second | Per-attempt 20 ms computation p95 | Computation windows above 20 ms | +| --- | ---: | ---: | ---: | +| PiPER | 4.38–14.62× | 0.446–0.500 ms | 0/600 | +| SO-101 | 14.73–20.17× | 0.396–0.744 ms | 0/600 | + +All attempts remain in the [Robot Simulation Lab](https://huggingface.co/spaces/Dvidia/robot-simulation-lab), including PiPER's slower first attempt. Wall time includes asset checks, compilation and recorded state construction; output serialization, browser rendering, physical sensing/transport and learned inference are outside that measurement. Runtime was macOS/Darwin arm64 with Python 3.12.9, MuJoCo 3.15.0 and NumPy 2.5.3. These CPU runs used no discrete-GPU workload; host load was not controlled, and universal minimum memory or real-time hardware behavior is not established. The exact bundles include source/runtime identities, native controls and effort units, gravity settings, every recorded trajectory and original licenses. ([PiPER receipt](https://huggingface.co/spaces/Dvidia/robot-simulation-lab/resolve/main/profiles/piper/report.json), [SO-101 receipt](https://huggingface.co/spaces/Dvidia/robot-simulation-lab/resolve/main/profiles/so101/report.json)) + +PiPER simulation assets are MIT; SO-101 assets are Apache-2.0. The latter has published build/design files and LeRobot guidance, while an open PiPER SDK does not by itself establish open manufacturing designs. The models' original notices are retained, and their attribution is separate from DVIDIA software. ([SO-101 design](https://github.com/TheRobotStudio/SO-ARM100), [LeRobot guide](https://huggingface.co/docs/lerobot/so101), [PiPER SDK](https://github.com/agilexrobotics/piper_sdk)) + +These additions are ready for local unloaded simulation. Each still needs its own object-task environment, grasp/release controls, calibrated observations and fresh closed-loop policy benchmark. YAM, OpenArm, ARX X5 and PAROL6 remain further integration targets; bimanual and dexterous manipulation retain separate contracts. The hosted lab displays the UR5e contact replay separately from these numerical receipts, so additional model support cannot be mistaken for inherited grasp competence. + ## Release an evidence-bearing skill, not an appearance match The installable unit should carry the policy, skill goal, observations/actions and units, named embodiment ordering, adapter, dependency hashes, calibration requirements, supported conditions, refusal/recovery behavior and benchmark receipts. Imported asset structure, characterized simulation and measured hardware agreement advance separately. A reference-only or simulation-qualified package remains useful for research while declaring physical execution unqualified. This follows the current DVIDIA roadmap's separation of collection, learning, simulation and physical qualification. ([Execution map](https://github.com/Dvidia-Inference/dvidia-training/blob/9172c37baeed5925e7bea3af2c92b0bf37a63c24/docs/roadmap.md)) diff --git a/reports/Native grasp task benchmark.md b/reports/Native grasp task benchmark.md index 75e07e8..c1b1c5b 100644 --- a/reports/Native grasp task benchmark.md +++ b/reports/Native grasp task benchmark.md @@ -82,3 +82,5 @@ The product opportunity is the loop connecting reviewed skill-space evidence, bo Run the benchmark using the [installation and evaluation guide](https://github.com/Dvidia-Inference/dvidia-training/blob/ac920cbf70147a7edeb0c706a9b71047f318e12f/docs/native-task-benchmark.md). The implementation and review are available in [GitHub PR #11](https://github.com/Dvidia-Inference/dvidia-training/pull/11). Published by **DVIDIA** · [hello@dvidia.org](mailto:hello@dvidia.org) · [@iammrriver on X](https://x.com/iammrriver) + +The [public Robot Simulation Lab](https://huggingface.co/spaces/Dvidia/robot-simulation-lab) also provides separate unloaded PiPER and SO-101 receipts and installation instructions. [Browser acceptance evidence](https://research.dvidia.org/downloads/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/acceptance.json) checks the recorded task replay on desktop and narrow screens. diff --git a/scripts/build-papers.mjs b/scripts/build-papers.mjs index b860535..dc1dc89 100644 --- a/scripts/build-papers.mjs +++ b/scripts/build-papers.mjs @@ -186,7 +186,7 @@ export async function buildPublications() { pages.push({ path: page, date: papers.reduce((latest, paper) => paper.date > latest ? paper.date : latest, topic.updated) }); } const paperRows = catalog.publications.map(p => `

· ${escapeHtml(p.type)}

${escapeHtml(p.title)}

${escapeHtml(p.abstract)}

${escapeHtml(p.status)}

`).join(""); - const content = `

An open research notebook

From observation
to physical skill.

We study how demonstrations become usable evidence, how simulation becomes useful practice, and when a robot skill earns a release.

Published by DVIDIA · hello@dvidia.org

Publications

${catalog.publications.length} papers and working notes

${paperRows}

Research footprints

  1. Give the reference arm a contact task

    A free rigid block must be grasped, lifted, carried, released and left settled at its destination. Frozen scenarios compare the authored controller with identical arm commands and open jaws, and with a hold control. Full traces and 3D replay preserve the evidence for the next compact action-learning experiment. Shoe physics, a learned task policy and hardware transfer remain open.

    Read the native task benchmark · Replay grasp and placement
  2. Build a hardware reference we can measure

    A pinned six-axis arm and coupled gripper expose fourteen coordinates and seven commands. All 18 unloaded numerical runs are valid; 1 ms CPU physics measures 40.80–47.29× elapsed-time throughput with sensing, rendering and policy excluded. A paired trace scorer makes physical mismatch measurable. Five-finger references and LeRobot collection have separate contracts; hardware fidelity and shoe competence remain unmeasured.

    Read the mechanical contract and upcoming arms · Open the 3D replay
  3. Learn a compact command-to-motion predictor

    A 6,438-byte CPU model predicts the next 20 ms of joint/jaw movement from native state and issued commands. Twelve seeded groups retain 36 recordings; 4,553 test transitions are not independent trials. Warm computation measures 0.207 ms p95, but jaw mean absolute/worst errors regress against persistence and the first opening is overpredicted. Shoe collection, hardware sensing and validated shoe physics remain open.

    Read the measured pilot and shoe collection plan
  4. Observe image motion and forecast its path

    A manually seeded CPU source alpha records pixel speed and causal forecasts. Six authored cases process 720 frames in 1.454 seconds; occlusion and low-texture losses remain visible, and a 500 ms reversal forecast misses by nearly 90 pixels. The v0.1 release/demo remain separate; real-scene accuracy, metric speed and live robot control are unqualified.

    Read the source alpha and measured limitations
  5. Publish the robotics research pipeline

    A diagram connects footage review, measured robot state, local training, frozen simulation evaluation and compatible skill installation. Automatic annotation and a learned task supervisor are proposed; their companion validation studies and physical robot qualification remain unrun.

    Read the architecture and validation gates
  6. A public skill link reaches a simulated arm

    One-click local installation binds a DVIDIA placement envelope to an authored contact controller. Six scene layouts pass; identical joint commands with open jaws and no-motion controls each pass 0/6. Physical transfer and learned policies remain open.

    Read the installation contract and arm replay
  7. The offline cable environment is runnable

    A local controller-training loop and simulation-only skill pack now run on CPU. Trained and baseline feedback each pass 10/10 simulated seeds; zero force and early release each pass 0/10. Real gripping, calibration and GPU qualification remain open.

    Read the first product build and replay
  8. Contact accuracy meets runtime

    54 deterministic CPU fixture trials recorded. Endpoint agreement coexists with unqualified force transients; coarse wall tests expose missed contact. No GPU or physical-transfer claim.

    Read the measurements
  9. Compare the observation-to-skill thesis with prior work

    Spatial attention, demonstration coverage, executable simulation and reusable repairs are reviewed separately. Three parallel Grok 4.7 searches support discovery; primary sources support the claims.

    Read the comparison
  10. Five licensed POV excerpts processed

    Ten image calls completed. A checker accepted an object error and captions omitted visible state changes. The separate accuracy pilot remains unrun.

    Read the feasibility run
  11. Define the evidence and release questions

    Four baseline protocols and a Skillspace graduation proposal establish the questions to test. A proposed protocol is distinct from an executed experiment.

    Read the agenda

Open questions

Follow new publications through RSS or inspect the source and revisions on GitHub.

`; + const content = `

An open research notebook

From observation
to physical skill.

We study how demonstrations become usable evidence, how simulation becomes useful practice, and when a robot skill earns a release.

Published by DVIDIA · hello@dvidia.org

Publications

${catalog.publications.length} papers and working notes

${paperRows}

Research footprints

  1. Run more arms on a local CPU

    PiPER and SO-101 join the library with pinned models, explicit arm/gripper units, preserved licenses and complete numerical receipts. Each completes six unloaded motion checks offline. Their task learning and physical qualification remain separate from the UR5e grasp demonstration.

    Read the profile measurements · Open the Robot Simulation Lab
  2. Give the reference arm a contact task

    A free rigid block must be grasped, lifted, carried, released and left settled at its destination. Frozen scenarios compare the authored controller with identical arm commands and open jaws, and with a hold control. Full traces and 3D replay preserve the evidence for the next compact action-learning experiment. Shoe physics, a learned task policy and hardware transfer remain open.

    Read the native task benchmark · Replay grasp and placement
  3. Build a hardware reference we can measure

    A pinned six-axis arm and coupled gripper expose fourteen coordinates and seven commands. All 18 unloaded numerical runs are valid; 1 ms CPU physics measures 40.80–47.29× elapsed-time throughput with sensing, rendering and policy excluded. A paired trace scorer makes physical mismatch measurable. Five-finger references and LeRobot collection have separate contracts; hardware fidelity and shoe competence remain unmeasured.

    Read the mechanical contract and upcoming arms · Open the 3D replay
  4. Learn a compact command-to-motion predictor

    A 6,438-byte CPU model predicts the next 20 ms of joint/jaw movement from native state and issued commands. Twelve seeded groups retain 36 recordings; 4,553 test transitions are not independent trials. Warm computation measures 0.207 ms p95, but jaw mean absolute/worst errors regress against persistence and the first opening is overpredicted. Shoe collection, hardware sensing and validated shoe physics remain open.

    Read the measured pilot and shoe collection plan
  5. Observe image motion and forecast its path

    A manually seeded CPU source alpha records pixel speed and causal forecasts. Six authored cases process 720 frames in 1.454 seconds; occlusion and low-texture losses remain visible, and a 500 ms reversal forecast misses by nearly 90 pixels. The v0.1 release/demo remain separate; real-scene accuracy, metric speed and live robot control are unqualified.

    Read the source alpha and measured limitations
  6. Publish the robotics research pipeline

    A diagram connects footage review, measured robot state, local training, frozen simulation evaluation and compatible skill installation. Automatic annotation and a learned task supervisor are proposed; their companion validation studies and physical robot qualification remain unrun.

    Read the architecture and validation gates
  7. A public skill link reaches a simulated arm

    One-click local installation binds a DVIDIA placement envelope to an authored contact controller. Six scene layouts pass; identical joint commands with open jaws and no-motion controls each pass 0/6. Physical transfer and learned policies remain open.

    Read the installation contract and arm replay
  8. The offline cable environment is runnable

    A local controller-training loop and simulation-only skill pack now run on CPU. Trained and baseline feedback each pass 10/10 simulated seeds; zero force and early release each pass 0/10. Real gripping, calibration and GPU qualification remain open.

    Read the first product build and replay
  9. Contact accuracy meets runtime

    54 deterministic CPU fixture trials recorded. Endpoint agreement coexists with unqualified force transients; coarse wall tests expose missed contact. No GPU or physical-transfer claim.

    Read the measurements
  10. Compare the observation-to-skill thesis with prior work

    Spatial attention, demonstration coverage, executable simulation and reusable repairs are reviewed separately. Three parallel Grok 4.7 searches support discovery; primary sources support the claims.

    Read the comparison
  11. Five licensed POV excerpts processed

    Ten image calls completed. A checker accepted an object error and captions omitted visible state changes. The separate accuracy pilot remains unrun.

    Read the feasibility run
  12. Define the evidence and release questions

    Four baseline protocols and a Skillspace graduation proposal establish the questions to test. A proposed protocol is distinct from an executed experiment.

    Read the agenda

Open questions

Follow new publications through RSS or inspect the source and revisions on GitHub.

`; await writePage("index.html", layout({ page: "index.html", title: "Research journal", description: "An open notebook on demonstrations, physical grounding, calibrated simulation and qualified robot skills.", content, catalog, topics, active: "index" })); pages.unshift({ path: "index.html", date: catalog.publications.reduce((latest, paper) => paper.date > latest ? paper.date : latest, "2026-10-07") }); const rss = `\nDVIDIA Research${ORIGIN}/Research footprints on demonstrations, calibrated simulation and robot skills.en${catalog.publications.map(p => { const url = `${ORIGIN}/papers/${p.slug}/`; return `${escapeHtml(p.title)}${url}${url}${new Date(`${p.date}T12:00:00Z`).toUTCString()}${escapeHtml(`${p.status}. ${p.abstract}`)}`; }).join("")}\n`; diff --git a/scripts/reference-replay.test.mjs b/scripts/reference-replay.test.mjs index ab7e90d..c69891a 100644 --- a/scripts/reference-replay.test.mjs +++ b/scripts/reference-replay.test.mjs @@ -155,7 +155,7 @@ test("3D replay seals downloadable artifacts and retains upstream license notice assert.match(notices, /three/i); assert.match(notices, /MIT/i); }); -test("research exposes recorded 3D replay and keeps all five hardware additions upcoming", async () => { +test("research exposes recorded 3D replay and distinguishes runnable profiles from qualification", async () => { // Read source and render Markdown without rebuilding docs: another test file // owns publication generation, so this check never races its directory writes. const [paper, homepage] = await Promise.all([ @@ -163,8 +163,11 @@ test("research exposes recorded 3D replay and keeps all five hardware additions readFile(resolve(ROOT, "scripts/build-papers.mjs"), "utf8") ]); const rendered = renderMarkdown(paper).html; - for (const target of ["YAM", "OpenArm", "PiPER", "ARX X5", "PAROL6"]) assert.ok(rendered.includes(target), target); - assert.match(paper, /all five additions remain upcoming/i); + for (const target of ["YAM", "OpenArm", "PiPER", "ARX X5", "PAROL6", "SO-101"]) assert.ok(rendered.includes(target), target); + assert.match(paper, /PiPER now has a pinned CPU simulation profile; YAM, OpenArm, ARX X5 and PAROL6 remain upcoming/i); + assert.match(paper, /SO-101[\s\S]*pinned CPU simulation is now runnable; its object-task and physical adapter qualification remain open/i); + assert.match(paper, /All five still need their own object-task qualification and physical adapter evidence/i); + assert.doesNotMatch(paper, /all five additions remain upcoming|SO-101 remains the accessible planned entry profile/i); const replayPath = "downloads/topics/physical-grounding/runs/2026-10-11-hardware-reference/replay/index.html"; assert.ok(rendered.includes(replayPath)); assert.ok(homepage.includes(replayPath)); diff --git a/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/320x740-carry.png b/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/320x740-carry.png new file mode 100644 index 0000000..0b75cba Binary files /dev/null and b/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/320x740-carry.png differ diff --git a/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/README.md b/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/README.md new file mode 100644 index 0000000..8a5cd21 --- /dev/null +++ b/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/README.md @@ -0,0 +1,3 @@ +# Recorded task replay browser checks + +These checks inspect the frozen native replay; they do not rerun physics. All nine displayed attempts and their source states are checked. The complete source denominator remains eighteen attempts. Screenshots show the initial block, elevated carry and released placement. The 320px layout retains readable controls. No browser errors or external requests were observed. diff --git a/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/acceptance.json b/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/acceptance.json new file mode 100644 index 0000000..ae5ecbb --- /dev/null +++ b/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/acceptance.json @@ -0,0 +1,17 @@ +{ + "scope": "Browser recorded-task replay acceptance; no learned-policy or physical qualification", + "demonstrated_attempt_id": "nominal-00--placement", + "recorded_success": true, + "carry_frame": 291, + "carry_seconds": 5.820000000000278, + "rendered_attempts": 9, + "full_benchmark_attempted": 18, + "checks": [ + "Native block, table and target geometry; six joint readouts; authored-controller scope and full source denominator", + "Every selected outcome matches its sealed source; every block pose matches the recorded native state; all controls replay to their terminal row", + "Initial, observed carry and terminal target views captured from actual state; completed source labels require measured lifted transfer and final target tolerance", + "Recorded playback, collision mode and idle rendering; 390 px and 320 px layouts keep 44 px controls without horizontal overflow" + ], + "errors": [], + "remote": [] +} diff --git a/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/carry-view.png b/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/carry-view.png new file mode 100644 index 0000000..a0c4631 Binary files /dev/null and b/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/carry-view.png differ diff --git a/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/final-view.png b/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/final-view.png new file mode 100644 index 0000000..4ad1209 Binary files /dev/null and b/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/final-view.png differ diff --git a/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/initial-view.png b/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/initial-view.png new file mode 100644 index 0000000..c9933bd Binary files /dev/null and b/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/initial-view.png differ diff --git a/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/source.json b/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/source.json new file mode 100644 index 0000000..37a2082 --- /dev/null +++ b/topics/physical-grounding/runs/2026-10-11-native-grasp/replay-browser/source.json @@ -0,0 +1,7 @@ +{ + "scope": "Browser replay acceptance, not fresh physics or hardware validation", + "tested_replay_integrity_sha256": "f4a334c177eab76942ee0ebb282138340040fc22be3f6377713d332fc572ba21", + "harness_source_sha256": "0d41eba08d7717d9b40e446ffa59ee017beeb1c20d9e40d4b73d62d82d61f729", + "checks": "All selected body positions, recorded outcomes, frame counts, paused rendering, desktop and mobile layout; no browser errors or external requests", + "screenshots": "Initial, measured carry, final placement and 320px mobile view; independently visually reviewed" +}