MODEL / CHECKPOINT
GRIT 99k
GRIT Team
Model
Checkpoint artifacts and an independent MotionChallenge inference adapter.
- Organization
- GRIT Team
- Type
- Dense
Graph-based reference encoder and dense policy head. - Version
- 99k · 39d573e9
post-v0.0.1 checkpoint; differs from the tagged release - Robot support
- Unitree G1 · 29 DoF
- Author source revision
- 5660177cf63f72803367d1e24eb7385a8cdc877c
Checkpoint artifacts
- Author release
- mrzuang/GRIT_teleop_deploy ↗
- Artifact type
- Author-released checkpoint
- Original file
- sim2real/checkpoints/policy.onnx ↗
Downloaded directly from the pinned author repository. Its SHA256 exactly matches both the author checksum and the former Drive mirror; the evaluated bytes are unchanged.
Download links deliver the adapter files. The release links and verification notes distinguish author files from conversions; SHA256 identifies the bytes used.
Policy checkpoint
Download ↗
Download ↗
39d573e970b7088275ed68ec8a9597fa9c402d98d5ad211991db7d8d0f5a82b9Inference interface
The adapter declares its control clock and reference window below. Source and configuration describe the observation and action transforms.
Observation requirements
Model-specific handling
Nine-frame reference context and 990D proprioceptive history use permitted orientation, joint state and angular velocity; preserve author reset warmup and raw-action history.
Exact checkpoint & evidence ↗- Control interval
- 20 ms
- Reference history
- 0 ms
- Reference preview
- 260 ms
- Output interface
- joint-pd-feedforward
- Configuration
- policy.json