Research map / Reinforcement learning / Visual and representation RL
Scalable and General Whole-Body Control for Cross-Humanoid Locomotion
Yufei Xue, Yunfeng Lin, Wentao Dong, Yang Tang, Jingbo Wang, Jiangmiao Pang, Ming Zhou, Minghuan Liu, Weinan Zhang
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 |
|---|---|---|---|
| one generalist policy instead of per-robot specialists, zero-shot | XHugWBC · 12 simulated humanoids | return: approximately 85% of specialist return | Mixed |
| fine-tune the generalist per robot instead of training specialists from scratch | XHugWBC Generalist-FT · 12 simulated humanoids | return: approximately 10% higher return than specialist | Improved |
| fine-tune the generalist per robot instead of training from scratch | XHugWBC Generalist-FT · 12 simulated humanoids | sample efficiency | Improved |
| use GNN or Transformer policy encoder instead of MLP (ablation) | XHugWBC · generalist training | episodic return and episode length | Improved |
| use physics-consistent morphological randomization instead of naive box-primitive randomization (ablation) | XHugWBC · zero-shot unseen robots | survival rate | Improved |
| use hybrid-mask Transformer with embodiment graph vs MetaMorph policy | XHugWBC · zero-shot unseen robots | survival rate | Improved |
| use MorAL simple architecture and proprioception only vs XHugWBC | MorAL · zero-shot unseen robots | survival rate | Hurt |
| fine-tune MetaMorph and MorAL baselines | MetaMorph, MorAL · per-robot fine-tuning | return | Mixed |
| zero-shot deploy sim-trained policy on real robots | XHugWBC · 7 real humanoids | forward velocity tracking error: Dobot Atom 0.063 -> 0.131 m/s | Hurt |
| generalist vs specialist velocity tracking | XHugWBC · 12 simulated humanoids average | Ev_x error: 0.060 -> 0.084 | Hurt |
| generalist vs specialist pelvis angle tracking | XHugWBC · 12 simulated humanoids average | pelvis angle error: 0.138 -> 0.097 | Improved |
Closest papers by meaning
- From Experts to a Generalist: Toward General Whole-Body Control for Humanoid Robots (NeurIPS 2025 · similarity 0.91)
- Hierarchical World Models as Visual Whole-Body Humanoid Controllers (ICLR 2025 · similarity 0.90)
- Adversarial Locomotion and Motion Imitation for Humanoid Policy Learning (NeurIPS 2025 · similarity 0.89)
- HumanoidArena: Benchmarking Egocentric Hierarchical Whole-body Learning (NeurIPS 2026 · similarity 0.89)
- Humanoid Horizon: Extending Task Horizon in Whole-Body Loco-Manipulation via Parallel Training, Dynamic Starting, and Reward Gating (NeurIPS 2026 · similarity 0.89)
- Universal Humanoid Motion Representations for Physics-Based Control (ICLR 2024 · similarity 0.88)
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.
