Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Humanoid Generative Pre-Training for Zero-Shot Motion Tracking
Zekun Qi, Xuchuan Chen, Jilong Wang, Chenghuai Lin, Yunrui Lian, Wenyao Zhang, Xinqiang Yu, He Wang, Li Yi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace 3-layer MLP backbone with causal Transformer (small) (ablation) | Humanoid-GPT tracker · AMASS-test (MuJoCo sim) | MPKPE: 100.49 -> 62.65 | Improved |
| replace 8-layer TCN with small causal Transformer (ablation) | Humanoid-GPT tracker · AMASS-test (MuJoCo sim) | SR: 81.48 -> 83.26 | Improved |
| scale training tokens from 2M to 20M (small model) (ablation) | Humanoid-GPT-S · AMASS-test (MuJoCo sim) | SR: 83.26 -> 86.02 | Improved |
| scale training tokens from 200M to 2B (base model) (ablation) | Humanoid-GPT-B · AMASS-test (MuJoCo sim) | SR: 88.27 -> 90.43 | Improved |
| scale model from 22.1M to 80.4M params at 2B tokens (ablation) | Humanoid-GPT · AMASS-test (MuJoCo sim) | SR: 90.43 -> 92.58 | Improved |
| scale model from 22.1M to 80.4M params at 2B tokens (ablation) | Humanoid-GPT · AMASS-test (MuJoCo sim) | RootVelErr: 0.1756 -> 0.1785 | Hurt |
| scale data and model capacity together (ablation) | Humanoid-GPT · AMASS-test (MuJoCo sim) | MPKPE: 62.65 -> 40.99 | Improved |
| use larger Transformer (B vs S) on real G1 dances (ablation) | Humanoid-GPT · Real G1 dances (4 clips) | MPJPE: 0.1024 -> 0.0974 (Can Do Can Go); 0.1180 -> 0.1075 (Gokuraku); 0.0825 -> 0.0858 (HuoYuanJia) | Mixed |
| curated 2B-frame corpus vs AMASS for motion diversity | training data · HME embedding space | log-volume: approximately 4-5 increase in log-volume compared with AMASS | Improved |
Closest papers by meaning
- DiT4DiT: Jointly Modeling Video Dynamics and Actions for Generalizable Robot Control (NeurIPS 2026 · similarity 0.88)
- Unleashing Large-Scale Video Generative Pre-training for Visual Robot Manipulation (ICLR 2024 · similarity 0.88)
- Zero-Shot Whole-Body Humanoid Control via Behavioral Foundation Models (ICLR 2025 · similarity 0.87)
- SUGAR: A Scalable Human-Video-Driven Generalizable Humanoid Loco-Manipulation Learning Framework (NeurIPS 2026 · similarity 0.87)
- Humanoid Locomotion as Next Token Prediction (NeurIPS 2024 · similarity 0.87)
- Universal Humanoid Motion Representations for Physics-Based Control (ICLR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
