MODEL / CHECKPOINT
Humanoid-GPT 264
Galbot / Tsinghua & collaborators
Model
The released 264 MLP checkpoint with the author offline-motion hardware inference, reference filtering and feature kinematics.
- Organization
- Galbot / Tsinghua & collaborators
- Type
- Dense
Released 136-input MLP with hidden widths 2048/1024/512 and 29 action outputs. This checkpoint is distinct from the Transformer described in the paper. - Version
- 264 · bbaac94b
author-released feed-forward checkpoint - Robot support
- Unitree G1 · 29 DoF
- Author source revision
- 9f9e7b74ecadb532abbb34b6a779d87191a9bbb6
Checkpoint artifacts
- Author release
- GalaxyGeneralRobotics/Humanoid-GPT ↗
- Artifact type
- Author-released checkpoint
- Original file
- pns_wo_priv264.onnx ↗
Original author-recommended264 ONNX; no conversion or weight modifications. Download bytes verified by SHA256 and author Git blob ff4b024a8f29a76f0c9453289eea48dff9120b7c.
pns_wo_priv264.onnx: a 136-input MLP with hidden widths 2048 / 1024 / 512 and 29 motor outputs.
Download links deliver the adapter files. The release links and verification notes distinguish author files from conversions; SHA256 identifies the bytes used.
Download ↗
bbaac94ba08d30bd96879e0c5878d60a38180f1cb819a9336cc6b56d49f333f9Inference 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
Author infer_onnx_real hardware route: fixed synthetic root [0,0,0.78], author reference FK, EMA and action history. No measured robot position or position integration.
Exact checkpoint & evidence ↗- Control interval
- 20 ms
- Reference history
- 0 ms
- Reference preview
- 20 ms
- Output interface
- joint-pd-feedforward
- Configuration
- policy.json