Research map / Video / Video generation
RL from Physical Feedback: Aligning Large Motion Models with Humanoid Control
Junpeng Yue, Zepeng Wang, Jiangxing Wang, Yuxuan Wang, Yu Zhang, Xinrun Xu, Bin Cao, Sipeng Zheng, gang ding, Zongqing Lu
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 |
|---|---|---|---|
| add motion tracking reward (ablation) | RLPF · CMU test, IsaacGym | Success rate: 0.32 -> 0.95 | Improved |
| add motion tracking reward (ablation) | RLPF · AMASS test, IsaacGym | Success rate: 0.30 -> 0.92 | Improved |
| add motion tracking reward (ablation) | RLPF · CMU test, MuJoCo | Success rate: 0.24 -> 0.75 | Improved |
| use Exbody2 tracker instead of PHC for the reward (ablation) | RLPF · CMU test, IsaacGym | Success rate: 0.91 -> 0.95 | Improved |
| use Exbody2 tracker instead of PHC for the reward (ablation) | RLPF · AMASS test, IsaacGym | Success rate: 0.80 -> 0.92 | Improved |
| use Exbody2 tracker instead of PHC for the reward (ablation) | RLPF · AMASS test, high-level | FID: 1.84 -> 3.34 | Mixed |
| add alignment verification (ablation) | RLPF · CMU test, high-level | FID: 32.53 -> 3.61 | Improved |
| add alignment verification (ablation) | RLPF · AMASS test, high-level | R@1: 0.07 -> 0.28 | Improved |
| add alignment verification (ablation) | RLPF · CMU test, MuJoCo | Success rate: 0.90 -> 0.75 | Hurt |
| add alignment verification (ablation) | RLPF · AMASS test, MuJoCo | Success rate: 0.99 -> 0.63 | Hurt |
| supervised fine-tuning on full data | Base Model · CMU test, IsaacGym | Success rate: 0.43 -> 0.36 | Hurt |
| supervised fine-tuning on tracking-filtered data | Base Model · CMU test, IsaacGym | Success rate: 0.43 -> 0.57 | Improved |
Closest papers by meaning
- Humanoid Locomotion as Next Token Prediction (NeurIPS 2024 · similarity 0.86)
- HumanoidArena: Benchmarking Egocentric Hierarchical Whole-body Learning (NeurIPS 2026 · similarity 0.86)
- Universal Humanoid Motion Representations for Physics-Based Control (ICLR 2024 · similarity 0.86)
- Hierarchical World Models as Visual Whole-Body Humanoid Controllers (ICLR 2025 · similarity 0.86)
- HumanTOMATO: Text-aligned Whole-body Motion Generation (ICML 2024 · similarity 0.86)
- SUGAR: A Scalable Human-Video-Driven Generalizable Humanoid Loco-Manipulation Learning Framework (NeurIPS 2026 · 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.
