Research map / Reinforcement learning / RL from feedback
Towards Adaptive Humanoid Control via Multi-Behavior Distillation and Reinforced Fine-Tuning
Yingnan Zhao, Xinmiao Wang, Dewei Wang, Xinzhe Liu, Dan Lu, Qilong Han, Peng Liu, Chenjia Bai
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 RL fine-tuning on diverse terrains after distillation | Multi-Behaviors Policy · Locomotion, Slope terrain | traversal success rate: 0.160 -> 0.975 | Improved |
| add RL fine-tuning on diverse terrains after distillation | Multi-Behaviors Policy · Locomotion, Discrete terrain | traversal success rate: 0.702 -> 0.969 | Improved |
| add RL fine-tuning on diverse terrains after distillation | Multi-Behaviors Policy · Locomotion, Hurdle terrain | traversal success rate: 0.756 -> 0.922 | Improved |
| add RL fine-tuning on diverse terrains after distillation | Multi-Behaviors Policy · Locomotion, Plane terrain | traversal success rate: 0.993 -> 0.992 | No clear effect |
| add RL fine-tuning on diverse terrains after distillation | Multi-Behaviors Policy · Fail recovery, Slope terrain | recovery success rate: 0.904 -> 1.000 | Improved |
| add RL fine-tuning on diverse terrains after distillation | Multi-Behaviors Policy · Fail recovery, Hurdle terrain | recovery success rate: 0.997 -> 0.985 | Hurt |
| distill walking and recovery into one multi-behavior policy | Walking Policy · Locomotion, Hurdle terrain | traversal success rate: 0.161 -> 0.756 | Improved |
| distill walking and recovery into one multi-behavior policy | Fail Recovery Policy · Fail recovery, Slope terrain | recovery success rate: 0.757 -> 0.904 | Improved |
| add PCGrad gradient projection with a shared critic (ablation) | AHC · Gradient cosine similarity | cosine similarity: 0.247 -> 0.519 | Improved |
| add PCGrad gradient projection with behavior-specific critics (ablation) | AHC · Gradient cosine similarity | cosine similarity: 0.334 -> 0.535 | Improved |
| use behavior-specific critics instead of a shared critic (ablation) | AHC · Gradient cosine similarity (no PCGrad) | cosine similarity: 0.247 -> 0.334 | Improved |
| use behavior-specific critics instead of a shared critic (ablation) | AHC · Stage-2 value loss and returns | value loss / episode return | Improved |
Closest papers by meaning
- Hierarchical World Models as Visual Whole-Body Humanoid Controllers (ICLR 2025 · similarity 0.85)
- SUGAR: A Scalable Human-Video-Driven Generalizable Humanoid Loco-Manipulation Learning Framework (NeurIPS 2026 · similarity 0.85)
- Hybrid Internal Model: Learning Agile Legged Locomotion with Simulated Robot Response (ICLR 2024 · similarity 0.85)
- Humanoid Horizon: Extending Task Horizon in Whole-Body Loco-Manipulation via Parallel Training, Dynamic Starting, and Reward Gating (NeurIPS 2026 · similarity 0.85)
- From Experts to a Generalist: Toward General Whole-Body Control for Humanoid Robots (NeurIPS 2025 · similarity 0.84)
- KungfuBot: Physics-Based Humanoid Whole-Body Control for Learning Highly-Dynamic Skills (NeurIPS 2025 · similarity 0.84)
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.
