A model is still running

A model is still running or downloading. Leaving will stop the current operation. Stay on this page to let it finish.

Skip to content
LocalAI.toolsBeta

Join the waitlist

Get an email when the new apps and model runs launch.

No account needed.

MotionBricks Unitree G1

Generate a short Unitree G1 motion transition from four recent poses and four target poses with the official MotionBricks ONNX graph. Returns complete pose, foot-contact and token tensors.

Browser inferencePowered by WebAI · Private · Unlimited runs
About this model

Generate a short Unitree G1 motion transition from four recent poses and four target poses with the official MotionBricks ONNX graph. Returns complete pose, foot-contact and token tensors.

Created by
NVIDIA
Model released
Unknown
Source repository created
Oct 23, 2026
Model weights · FP32
734.3 MB

Best for

  • Offline animation and pose-transition exploration on the exact G1 skeleton.
  • Inspect raw model-space motion and predicted contact timelines.

Keep in mind

  • G1 joint coordinates only, in Z-up,+X-forward model space, meters,radians and unit XYZW root quaternions; no text, video, scene geometry or arbitrary skeleton input.
  • Raw output contains boundary and unused frames. Only the leading4×pred_num_tokens frames are meaningful; trim three frames at each end for playback. Raw foot-contact regression can exceed0–1; the visualization clamps display values, leaving raw tensors unchanged.
  • No endpoint correction, root smoothing, world alignment, tracking controller or physics simulation is applied. Not a physical robot controller or safety/limit validation.
  • Quality scope: internal provider walking recording plus two authored pose cases; not broad motion-quality certification.
  • Measured WASM only in Chrome157 on Apple M3 Max. Official FP32 WebGPU initialization fails Cast provider assignment. Atomic graph has no incremental streaming.
  • Complete verified execution-dependency backup and cold model/package source recovery remain pending. Normal production browser and native CPU app/history/replay scopes are qualified; broader backends/platforms are separate.

Processed on your deviceInference runs in this browser. Input/output sharing requires consent for each run. Basic run logs include device information and approximate connection location. Model and runtime files download from external hosts. Saved downloads may be cleared by your browser.

Sources & technical details
Original model
https://huggingface.co/nvidia/MotionBricks
Release verification
Unknown
Release evidence
Exact original public checkpoint release day not established; repository creation, paper date and ONNX update are distinct.
Source updated
Oct 23, 2026
Browser model files
NVlabs/motionbricks
Files updated
Oct 26, 2026
Artifact revision
54fc563aeaaf819a753d72cc9fdc6713989cdd43

Input

TEXT

Sample result

MotionBricks Unitree G1

Settings

Render root trajectory and predicted foot contacts while retaining all native tensors. Adds a small SVG rendering cost; raw inference output is unchanged.

Output

Your result will appear here

Add input or try the sample, then run the model.

Benchmark evidence

Saved worker evidence is separate from benchmarks run on your device. It does not guarantee speed or quality on every browser.

MotionBricks Unitree G1

2 configurations tested
Passed2
Failed0
TestedNov 16, 2026
Runtimeonnxruntime-web 1.31.0-dev.20260914-8d85527a0

Quality criteria

{ "scope": "G1 internal NVIDIA-technology evaluation on provider walk recording; not redistributable sample media. Hand-authored independent pose fixtures required separately for published app samples.", "minimumTokenCount": 6, "maximumTokenCount": 16, "maximumInferenceMs": 10000, "maximumEndpointRootErrorM": 0.15, "maximumAdjacentRootStepM": 0.15, "maximumMeanEndpointJointErrorRad": 0.35, "maximumAdjacentJointStepRad": 0.8, "fixtureSource": "NVlabs/motionbricks@54fc563aeaaf819a753d72cc9fdc6713989cdd43/assets/motions/unitree_g1/locomotion/bs_1d_linear/Neutral_walk_forward_002__A057.csv", "fixtureLicense": "NVIDIA Sample Data Evaluation License; retain locally for internal evaluation only; never publish or mirror recording or derivatives." }

PrecisionDeviceResult
FP32CPU · WASMPassedVery fast
FP32CPU · WASMPassedVery fast
Test setup & sample details
Quality gate
{ "scope": "G1 internal NVIDIA-technology evaluation on provider walk recording; not redistributable sample media. Hand-authored independent pose fixtures required separately for published app samples.", "minimumTokenCount": 6, "maximumTokenCount": 16, "maximumInferenceMs": 10000, "maximumEndpointRootErrorM": 0.15, "maximumAdjacentRootStepM": 0.15, "maximumMeanEndpointJointErrorRad": 0.35, "maximumAdjacentJointStepRad": 0.8, "fixtureSource": "NVlabs/motionbricks@54fc563aeaaf819a753d72cc9fdc6713989cdd43/assets/motions/unitree_g1/locomotion/bs_1d_linear/Neutral_walk_forward_002__A057.csv", "fixtureLicense": "NVIDIA Sample Data Evaluation License; retain locally for internal evaluation only; never publish or mirror recording or derivatives." }
Tested
2026-11-16T15:23:41.943Z
Environment
Apple M3 Max96GiB/macOS27.2/Chrome157; eight WASM threads; exact staged graph. Persistent B2 mirror active; separate CoreML compatibility compilation active during public timings. No duplicate same-configuration benchmark.
Runtime
onnxruntime-web 1.31.0-dev.20260914-8d85527a0
Artifact revision
54fc563aeaaf819a753d72cc9fdc6713989cdd43
Fixture
{
  "file": "model-samples/motionbricks-g1/standing-input.json",
  "size": 5678,
  "sha1": "402501f9b8d98a911549dab33c78722f7d1ac1ef",
  "sha256": "a9cd8eff52d8e7a7fe64be2c093760f999e99e78b6838649d8843cc2675879c1",
  "url": "https://assets.localai.tools/samples/v1/files/a9cd8eff52d8e7a7fe64be2c093760f999e99e78b6838649d8843cc2675879c1/standing-input.json"
}
Rerun when
artifact revision; runtime version; fixture checksum; quality gate; backend behavior
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "finite": true,
      "tokenCount": 11,
      "scope": "Authored pose contract case; independent real-walk quality and native/WASM parity are retained separately.",
      "passed": true,
      "endpointRootErrorM": 0.002416581245365056,
      "meanEndpointJointErrorRad": 0.01848469114268263,
      "maxRootStepM": 0.00020897215209185759,
      "maxJointStepRad": 0.034323155879974365
    },
    "runtime": {
      "durationMs": 65.24000000208616,
      "loadMs": 4790.829999998212,
      "inferenceMs": 65.35000000149012,
      "totalMs": 4856.179999999702,
      "framesPerSecond": 581.4843152124487,
      "grade": "good",
      "engine": "onnxruntime-web",
      "version": "1.31.0-dev.20260914-8d85527a0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "artifactRevision": "54fc563aeaaf819a753d72cc9fdc6713989cdd43",
      "precision": "fp32",
      "device": "wasm"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "finite": true,
      "tokenCount": 11,
      "scope": "Authored pose contract case; independent real-walk quality and native/WASM parity are retained separately.",
      "passed": true,
      "endpointRootErrorM": 0.026404865897954793,
      "meanEndpointJointErrorRad": 0.021903239541430152,
      "maxRootStepM": 0.015825882947598464,
      "maxJointStepRad": 0.04737657308578491
    },
    "runtime": {
      "durationMs": 42.36499999463558,
      "loadMs": 4790.829999998212,
      "inferenceMs": 42.40500000119209,
      "totalMs": 4833.234999999404,
      "framesPerSecond": 896.1207404535253,
      "grade": "good",
      "engine": "onnxruntime-web",
      "version": "1.31.0-dev.20260914-8d85527a0",
      "actualDevice": "wasm",
      "actualPrecision": "fp32",
      "artifactRevision": "54fc563aeaaf819a753d72cc9fdc6713989cdd43",
      "precision": "fp32",
      "device": "wasm"
    }
  }
]

Something not quite right?

Tell us what happened with MotionBricks. Loading, results, or the page itself—we’re here for all of it.

Attachments Optional

Up to 3 files · 50 MB each

Browser, OS, CPU/GPU, memory, screen, storage, and network details are included automatically. Model inputs, results, and browsing history are not collected.