mcMotionChallengeGitHub ↗

MODEL / CHECKPOINT

Horizon Robotics

Model

Checkpoint artifacts and an independent MotionChallenge inference adapter.

Organization
Horizon Robotics
Type
MoE
Reference-routed mixture-of-experts Transformer policy.
Version
v2.0.0
author checkpoint
Robot support
Unitree G1 · 29 DoF
Author source revision
ebb3747ad3b6dab57e4b621611efef87317bdc3b

Checkpoint artifacts

Author release
HorizonRobotics/HoloMotion v2.0.0 ↗
Artifact type
Author-released checkpoint
Original file
HoloMotion_motion_tracking_model_v2.0.0/exported/model_10500.onnx ↗

SHA256 pins the author artifact. Native inference parity is recorded separately.

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 ↗
b0681976135eed572d1d20aefc52ec65c149bdc636f2594a6436fa7136a3fc86

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

Actual control-point offsets are local FK. Reconstruct with current joints/orientation at a fixed zero origin; compare the complete 1378D released WholeBody14 inference path. Reference positions remain relative to the reference root.

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

Evaluations