Research map / Vision-language and multimodal / Robot action and navigation
End-to-End Language-Action Model for Humanoid Whole Body Control
Yuxuan Wang, Haobin Jiang, Shiqing Yao, Ziluo 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 |
|---|---|---|---|
| replace direct motion generation + retargeting with end-to-end training on simulated robot trajectories | SENTINEL · Text-based WBC test set | Success rate (%): 94.94 -> 99.45 vs MDM + Retarget | Improved |
| add multi-scale long-term observation history (4Hz over 10s) (ablation) | SENTINEL · Text-based WBC test set | R@1: 0.153 -> 0.582 | Improved |
| use multi-scale history instead of 20-frame 10Hz history (ablation) | SENTINEL · Text-based WBC test set | R@1: 0.489 -> 0.582 | Improved |
| add auxiliary state prediction (ablation) | SENTINEL · Text-based WBC test set | Success rate (%): 98.67 -> 99.45 | Improved |
| add auxiliary state prediction (ablation) | SENTINEL · Text-based WBC test set | R@1: 0.589 -> 0.582 | No clear effect |
| add done prediction token (ablation) | SENTINEL · Text-based WBC test set | MMD (1e-2): 5.852 -> 3.438 | Improved |
| increase action chunk size H from 5 to 50 (ablation) | SENTINEL · Text-based WBC test set | MMD and success rate | Improved |
| use rollout step K=5 rather than other values (ablation) | SENTINEL · Text-based WBC test set | R@1 and success rate | Improved |
| scale model from 60M to 600M parameters (ablation) | SENTINEL · Text-based WBC test set | Success rate (%): 22.73 -> 99.45 | Improved |
| scale model from 200M to 600M parameters (ablation) | SENTINEL · Text-based WBC test set | R@1: 0.371 -> 0.582 | Improved |
| add residual action head via PPO post-training (ablation) | SENTINEL · Domain randomization, K=10 | Success rate (%): 95.44 -> 99.11 | Improved |
| add residual action head via PPO post-training (ablation) | SENTINEL · Domain randomization, K=10 | R@1: 0.315 -> 0.392 | Improved |
Closest papers by meaning
- Humanoid Locomotion as Next Token Prediction (NeurIPS 2024 · similarity 0.89)
- Hierarchical World Models as Visual Whole-Body Humanoid Controllers (ICLR 2025 · similarity 0.88)
- HumanoidArena: Benchmarking Egocentric Hierarchical Whole-body Learning (NeurIPS 2026 · similarity 0.87)
- SUGAR: A Scalable Human-Video-Driven Generalizable Humanoid Loco-Manipulation Learning Framework (NeurIPS 2026 · similarity 0.86)
- H-GAP: Humanoid Control with a Generalist Planner (ICLR 2024 · similarity 0.86)
- Adversarial Locomotion and Motion Imitation for Humanoid Policy Learning (NeurIPS 2025 · 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.
