mcMotionChallengeGitHub ↗

MODEL / CHECKPOINT

Horizon Robotics

Model

The original v1.3.1 checkpoint using the author runtime legacy RoPE window handling.

Organization
Horizon Robotics
Type
MoE
Reference-routed mixture-of-experts Transformer policy.
Version
v1.3.1
author checkpoint
Robot support
Unitree G1 · 29 DoF
Author source revision
adaacd50ada6ba9338f6c3f539093f85453e2554

Checkpoint artifacts

Author release
HorizonRobotics/HoloMotion ↗
Artifact type
Author-released checkpoint
Original file
HoloMotion_motion_tracking_model_v1.3.1/exported/motion_tracking_model.onnx ↗

The adapter pins this published artifact by SHA256. Artifact identity is separate from inference parity.

Original checkpoint with an 8192-position RoPE table. The author runtime resets its KV cache at 8128 steps while retaining reference time and physical state.

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 ↗
ae28e8e9bf39bc752e82ade5e0ea0436689dafa07af41fa39af696f6f6a2bbfe

Inference interface

The adapter declares its control clock and reference window below. Source and configuration describe the observation and action transforms.

Observation requirements

Observation rules ↗

Model-specific handling

Released actor uses joint state, angular velocity/gravity or orientation and reference features. Retain author normalization, history/recurrent state, action processing and declared lookahead; validate numerical independence from actual translation.

Exact checkpoint & evidence ↗
Control interval
20 ms
Reference history
0 ms
Reference preview
200 ms
Output interface
joint-pd-feedforward
Configuration
policy.json

Evaluations