MODEL / CHECKPOINT
TextOp Tracker 75k
TeleAI / SJTU / ECUST
Model
The released low-level universal motion tracker from TextOp. This entry evaluates reference-motion tracking, without the high-level text-to-motion generator.
Training data
The authors describe a mixture of public motion data and a small private dataset. Their README names AMASS, BABEL, TEACH and LAFAN1; exact overlap with this retargeted LAFAN1 suite is not established.
Author documentation ↗- Organization
- TeleAI / SJTU / ECUST
- Type
- Dense
Dense MLP Tracker; the upstream text-to-motion generator is not evaluated. - Version
- Tracker 75k · 8df6371c
author-released ONNX checkpoint - Robot support
- Unitree G1 · 29 DoF
- Author source revision
- ef6555fb174c9b5c44945a62c7ffc77b5ddbbf22
Checkpoint artifacts
- Author release
- TeleHuman/TextOp ↗
- Artifact type
- Author-released checkpoint
- Original file
- latest.onnx ↗
The author usage guide links this artifact repository. Checkpoint identity and inference parity are independently verified.
Download links deliver the adapter files. The release links and verification notes distinguish author files from conversions; SHA256 identifies the bytes used.
Download ↗
8df6371cce677a6ef8fd7124d082fbf98df11735b9e1aa5808f1bf9d3152b582Inference 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 C++ hardware route: initial-heading reference alignment, all15 future position inputs zeroed by the author, and last action initialized from the author start pose. The exact hardware weight matches latest.onnx. Canonical benchmark physical reset is unchanged; no robot position is read.
Exact checkpoint & evidence ↗- Control interval
- 20 ms
- Reference history
- 0 ms
- Reference preview
- 80 ms
- Output interface
- joint-pd-feedforward
- Configuration
- policy.json