Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 13 additions & 1 deletion .github/workflows/ci.yml
Original file line number Diff line number Diff line change
Expand Up @@ -71,6 +71,8 @@ jobs:
dvidia-reference-task --help
dvidia-robot-sim --help
dvidia-robot-sim list
dvidia-assets --help
dvidia-assets list
dvidia-piper-grip --help
dvidia-piper-contact --help
dvidia-piper-recovery --help
Expand All @@ -97,8 +99,9 @@ jobs:
run: |
npm ci --ignore-scripts --prefix viewer
npm run --prefix viewer build
npm run --prefix viewer build:assets
node tests/browser/reference_replay_loader.cjs
git diff --exit-code -- src/dvidia_training/reference_replay.bundle.js src/dvidia_training/reference_replay.third-party.txt
git diff --exit-code -- src/dvidia_training/reference_replay.bundle.js src/dvidia_training/reference_replay.third-party.txt src/dvidia_training/asset_library.bundle.js
- name: Provision pinned reference models
run: |
python -m pip install ".[arm]"
Expand Down Expand Up @@ -142,6 +145,14 @@ jobs:
env:
DVIDIA_ROBOT_PROFILE_ASSETS: runs/robot-profile-assets-ci
run: python -m unittest tests.test_robot_profiles tests.test_piper_task tests.test_piper_task_benchmark tests.test_piper_grip_guard tests.test_piper_grip_benchmark tests.test_piper_contact_memory tests.test_piper_contact_benchmark tests.test_piper_contact_recovery tests.test_piper_recovery_benchmark -v
- name: Test reusable shoe assets and scene previews
env:
DVIDIA_ASSET_PREVIEW_TEST_ASSETS: runs/robot-profile-assets-ci
DVIDIA_ASSET_UR5E_ROOT: runs/reference-assets-ci
DVIDIA_ASSET_LIBRARY_ROOTS: '{"ur5e":"runs/reference-assets-ci","piper":"runs/robot-profile-assets-ci","so101":"runs/robot-profile-assets-ci"}'
DVIDIA_REFERENCE_ASSETS: runs/reference-assets-ci
DVIDIA_ROBOT_PROFILE_ASSETS: runs/robot-profile-assets-ci
run: python -m unittest tests.test_asset_library tests.test_asset_benchmark tests.test_asset_preview -v
- name: Retain PiPER contact task and matched controls
run: |
dvidia-piper-task benchmark runs/robot-profile-assets-ci --output runs/piper-task-ci --nominal 1 --stress 0
Expand All @@ -154,6 +165,7 @@ jobs:
run: |
python -m pip install -r publishing/huggingface/requirements.txt
python -m unittest discover -s publishing/huggingface -p 'test_publish_robot_sim.py' -v
python -m unittest discover -s publishing/huggingface -p 'test_prepare_asset_library.py' -v
python -m unittest discover -s publishing/huggingface/robot-simulation-lab -p 'test_*.py' -v
- name: Preserve additional profile evidence
if: always()
Expand Down
12 changes: 7 additions & 5 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -23,11 +23,13 @@ unloaded mechanics; learned policies and physical qualification remain open. The
[Robot Simulation Lab](https://huggingface.co/spaces/Dvidia/robot-simulation-lab)
publishes recorded evidence and local run instructions.

The [asset-library roadmap](docs/robot-simulation-library.md#reusable-asset-library-planned-product-layer)
extends this toward reusable objects, environments, grippers and sensor profiles,
starting with a shoe-organization collection. Packages will distinguish measured,
estimated and untested physical properties and link to compatible skill benchmarks.
The object catalog and installation interface are planned work.
The [shoe asset library v0.1](docs/asset-library.md) adds six original rigid shoe
proxies, versioned scene descriptors and a `dvidia-assets` loader for UR5e/Robotiq,
PiPER and SO-101. Local 3D previews expose the native geometry; CPU benchmarks
check gravity and support contact with matched controls. Properties are labelled
as synthetic assumptions. Shoe grasping, deformation and hardware transfer still
need their own evidence. The [broader asset roadmap](docs/robot-simulation-library.md#reusable-asset-library-planned-product-layer)
extends this foundation toward measured objects, fixtures, grippers and sensors.

The [PiPER grip-recovery candidate](docs/piper-grip-recovery.md) adds a short
test lift before carrying, one supported recovery attempt, and a paired
Expand Down
54 changes: 54 additions & 0 deletions benchmarks/shoe-asset-library-v01-20261012/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,54 @@
# Shoe asset library v0.1 — complete native evidence

Frozen benchmark source: [`441ee2871620104246ab9caeb89825f12a14b889`](https://github.com/Dvidia-Inference/dvidia-training/tree/441ee2871620104246ab9caeb89825f12a14b889).

- Six original MIT rigid shoe proxies: trainer, ankle-boot and slipper left/right pairs.
- UR5e/Robotiq, PiPER and SO-101 native scenes, portable descriptors and local 3D previews.
- 54/54 numerically valid attempts. Supported scenes: **27/27**. Disabled-table support: **0/27**; all 27 fall as required. All 54 gravity checks pass.
- No warnings or loaded non-table shoe contacts in the 20 ms recorded samples. This sampling does not rule out shorter transient contacts.
- Complete denominator: 3 robots ×3 shoe pairs ×3 seeded layouts ×2 support conditions; 108 repeated synthetic shoe instances, not independent physical specimens.

[Browse the library](library/index.html) · [Full benchmark](library/benchmark/report.html) · [Protocol](library/benchmark/protocol.json) · [Every sampled trajectory and native step duration](library/benchmark/attempts.json) · [File seal](library/library-integrity.json)

## Frozen execution

A wheel built from the exact Git source archive was installed into a fresh Python environment. Fourteen installed runtime/catalog/viewer/lock files were byte-compared with the archive before execution. `sandbox-exec` denied network access throughout the benchmark and three exports; a denied socket-bind self-check was required before attempt zero. `OPENBLAS_NUM_THREADS=1` and `OMP_NUM_THREADS=1` were set.

[Installed runtime](installed-runtime.json) · [Execution receipt](execution.json) · [Complete execution log](execution-log.txt) · [Reusable execution script](run-frozen.py) · [25 native/contract tests](native-tests.txt).

The source declares seed 612047 and the 54-attempt protocol before execution; the emitted protocol is written before the first attempt. Development smoke used seed 109731 separately. No outcome-triggered reruns or omitted attempts were used for this result.

## Runtime boundaries

Apple M5, 10 logical CPUs, 24 GiB RAM, macOS arm64; Python 3.12.9, MuJoCo 3.15.0, NumPy 2.5.3. Native stepping is CPU-only.

54 simulated seconds took **12.3616 wall seconds**, about **4.37×** simulated throughput including each fresh asset verification, compilation and state extraction. Run wall stops before serialization; rendering is separate. The slowest complete attempt took **6.7047s**, retained in the receipt.

Across 27,000 native `mj_step` calls, median **8.50µs**, p95 **15.75µs**, p99 **23.42µs**, maximum **111.92µs**. These exclude observation extraction, rendering, sensors, transport and actuator response. Median compilation was **65.66ms**. Peak process RSS **1,111,916,544bytes** includes allocations in this process and is not a minimum RAM specification. No real-time guarantee is made.

## Scope and reproduction

These are synthetic rigid proxies with untested mass/contact assumptions. Native loading, gravity and passive support do not establish shoe grasping, deformation, learned competence or physical transfer. Geometry is intentionally outside the held robot's immediate contacts; reachability and graspability are unqualified.

See the [installation and scene-loading guide](../../docs/asset-library.md) and [benchmark protocol](../../docs/asset-library-benchmark-protocol.md). Fetch UR5e into a fresh root first, then add PiPER and SO-101. After provisioning:

```sh
dvidia-assets benchmark --assets ./robot-assets --seed 612047 --repeats 3 --source-commit 441ee2871620104246ab9caeb89825f12a14b889 --output ./asset-benchmark
dvidia-assets inspect ./asset-benchmark
dvidia-assets scene --robot piper --assets ./robot-assets --pair all --output ./shoe-preview
python -m http.server 8360 --bind 127.0.0.1 --directory ./shoe-preview
```

The library includes exact upstream model license notices. DVIDIA tooling and original shoe proxies are MIT; robot models retain their individual licenses. The binary viewer uses recorded native body poses and performs no physics or hardware control in the browser.

## Corrected viewer source

The installable viewer and publication tooling use [`17a7c81784362697f09757712efbe00d98743501`](https://github.com/Dvidia-Inference/dvidia-training/tree/17a7c81784362697f09757712efbe00d98743501). Browser acceptance caught an animation timestamp race that could request a negative frame. Playback now clamps indices and cannot move backwards on stale callbacks. A deterministic test forces a timestamp 10 seconds old before checking start, resume and pause.

The original benchmark was **not rerun**. Only the browser bundle and its preview seals changed; all native geometry, motion, scene descriptors and model notices are byte-identical to the frozen exports. [Installed corrected wheel and preservation proof](viewer-build.json). The library records both benchmark and publication source revisions.

[Final browser acceptance](browser-acceptance.json): all three arms passed at 1440, 390 and 320 pixels, with rendered geometry, scene downloads, controls and stale-animation-timestamp regression checks. No JavaScript errors, unexpected remote imports or horizontal overflow were observed.

The final packaging and clarified setup guide use [`45bef6315c9632b33915e1ce993babd2926b8f7d`](https://github.com/Dvidia-Inference/dvidia-training/tree/45bef6315c9632b33915e1ce993babd2926b8f7d). Viewer code and all native evidence remain unchanged from the revisions above.

Static-host navigation was corrected in [`d291ee11f28b0d3c7a0b9ed062174e46fe764d15`](https://github.com/Dvidia-Inference/dvidia-training/tree/d291ee11f28b0d3c7a0b9ed062174e46fe764d15): robot and overview links use explicit `index.html` paths for Hugging Face. [Navigation proof](navigation-build.json) preserves the prior viewer receipt and binds only changed HTML and seals; all native and JavaScript bytes are unchanged.
161 changes: 161 additions & 0 deletions benchmarks/shoe-asset-library-v01-20261012/browser-acceptance.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,161 @@
{
"schema": "dvidia.asset-library-browser-acceptance/v0.1",
"source_commit": "17a7c81784362697f09757712efbe00d98743501",
"target": "final sealed publication previews served locally over loopback",
"robots": {
"ur5e": {
"records": [
{
"width": 1440,
"renderCount": 15,
"renderedPixels": {
"colored": 6089,
"darkNeutral": 7417,
"width": 1058,
"height": 581
},
"errors": [],
"remoteDownloads": 0,
"overflow": false
},
{
"width": 390,
"renderCount": 14,
"renderedPixels": {
"colored": 2392,
"darkNeutral": 2995,
"width": 364,
"height": 391
},
"errors": [],
"remoteDownloads": 0,
"overflow": false
},
{
"width": 320,
"renderCount": 16,
"renderedPixels": {
"colored": 1584,
"darkNeutral": 1981,
"width": 294,
"height": 391
},
"errors": [],
"remoteDownloads": 0,
"overflow": false
}
],
"timestampRegression": {
"forcedTimestampOffsetMs": -10000,
"resumeFrame": 60,
"pausedFrame": 60,
"initialFrame": 0,
"normalPlaybackResumed": true,
"errors": []
}
},
"piper": {
"records": [
{
"width": 1440,
"renderCount": 17,
"renderedPixels": {
"colored": 7906,
"darkNeutral": 4570,
"width": 1058,
"height": 581
},
"errors": [],
"remoteDownloads": 0,
"overflow": false
},
{
"width": 390,
"renderCount": 15,
"renderedPixels": {
"colored": 3100,
"darkNeutral": 1839,
"width": 364,
"height": 391
},
"errors": [],
"remoteDownloads": 0,
"overflow": false
},
{
"width": 320,
"renderCount": 15,
"renderedPixels": {
"colored": 2058,
"darkNeutral": 1230,
"width": 294,
"height": 391
},
"errors": [],
"remoteDownloads": 0,
"overflow": false
}
],
"timestampRegression": {
"forcedTimestampOffsetMs": -10000,
"resumeFrame": 60,
"pausedFrame": 60,
"initialFrame": 0,
"normalPlaybackResumed": true,
"errors": []
}
},
"so101": {
"records": [
{
"width": 1440,
"renderCount": 14,
"renderedPixels": {
"colored": 12621,
"darkNeutral": 4326,
"width": 1058,
"height": 581
},
"errors": [],
"remoteDownloads": 0,
"overflow": false
},
{
"width": 390,
"renderCount": 15,
"renderedPixels": {
"colored": 4961,
"darkNeutral": 1706,
"width": 364,
"height": 391
},
"errors": [],
"remoteDownloads": 0,
"overflow": false
},
{
"width": 320,
"renderCount": 21,
"renderedPixels": {
"colored": 3266,
"darkNeutral": 1138,
"width": 294,
"height": 391
},
"errors": [],
"remoteDownloads": 0,
"overflow": false
}
],
"timestampRegression": {
"forcedTimestampOffsetMs": -10000,
"resumeFrame": 60,
"pausedFrame": 60,
"initialFrame": 0,
"normalPlaybackResumed": true,
"errors": []
}
}
},
"scope": "Browser rendering and interaction acceptance, not a physics rerun or hardware test."
}
Original file line number Diff line number Diff line change
@@ -0,0 +1,44 @@
{
"records": [
{
"width": 1440,
"renderCount": 14,
"renderedPixels": {
"colored": 7893,
"darkNeutral": 4312,
"width": 1058,
"height": 581
},
"errors": [],
"remoteDownloads": 0,
"overflow": false
},
{
"width": 390,
"renderCount": 12,
"renderedPixels": {
"colored": 3102,
"darkNeutral": 1830,
"width": 364,
"height": 391
},
"errors": [],
"remoteDownloads": 0,
"overflow": false
},
{
"width": 320,
"renderCount": 13,
"renderedPixels": {
"colored": 2069,
"darkNeutral": 1274,
"width": 294,
"height": 391
},
"errors": [],
"remoteDownloads": 0,
"overflow": false
}
],
"scope": "Development PiPER browser acceptance; public final artifacts verified separately."
}
Loading
Loading