Research map / Reinforcement learning / Visual and representation RL
MaskAdapt: Learning Flexible Motion Adaptation via Mask-Invariant Prior for Physics-Based Characters
Soomin Park, Eunseong Lee, Kwang Bin Lee, Sung-Hee Lee
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 mask-invariant KL loss (lambda 1.0) to masked base-policy training (ablation) | MaskAdapt base policy · LAFAN1 coverage | normalized entropy: 0.7448 -> 0.9025 | Improved |
| use weaker MI loss weight (0.1 vs 1.0) (ablation) | MaskAdapt base policy · LAFAN1 coverage | normalized entropy: 0.8968 vs 0.9025 | No clear effect |
| apply stochastic body-part masking without MI loss (ablation) | base policy · LAFAN1 coverage | normalized entropy: 0.9124 -> 0.7448 | Hurt |
| use MaskAdapt residual policy instead of CML | motion composition · Case (a) preserved region | normalized entropy: 0.7835 -> 0.8580 | Improved |
| use MaskAdapt instead of CML | motion composition · Case (b) adapted region | normalized entropy: 0.7478 -> 0.8710 | Improved |
| use MaskAdapt instead of CML | motion composition · Case (c) preserved region | normalized entropy: 0.8461 -> 0.8980 | Improved |
| use MaskAdapt instead of CML on Strike task | goal-driven task · Strike | success rate (%): 46.5 -> 64.2 | Improved |
| use MaskAdapt instead of CML on Location task | goal-driven task · Location | success rate (%): 100.0 -> 99.5 | No clear effect |
| use MaskAdapt instead of CML on Heading task | goal-driven task · Heading | success rate (%): 75.2 -> 76.9 | No clear effect |
| residual on mask-invariant base instead of from-scratch policy | partial goal tracking · Partial tracking (Arms) | tracking error (m): 0.076 -> 0.055 | Improved |
| use MaskAdapt instead of CML | partial goal tracking · Partial tracking (Arms) | DTW pose error (m): 0.079 -> 0.047 | Improved |
| use arm-only masked residual vs MaskedMimic arm tracking | partial goal tracking · Arm tracking | tracking error (m): 0.0849 -> 0.0423 | Improved |
Closest papers by meaning
- Universal Humanoid Motion Representations for Physics-Based Control (ICLR 2024 · similarity 0.87)
- Humanoid Horizon: Extending Task Horizon in Whole-Body Loco-Manipulation via Parallel Training, Dynamic Starting, and Reward Gating (NeurIPS 2026 · similarity 0.86)
- Adversarial Locomotion and Motion Imitation for Humanoid Policy Learning (NeurIPS 2025 · similarity 0.85)
- KungfuBot: Physics-Based Humanoid Whole-Body Control for Learning Highly-Dynamic Skills (NeurIPS 2025 · similarity 0.85)
- Meta-Controller: Few-Shot Imitation of Unseen Embodiments and Tasks in Continuous Control (NeurIPS 2024 · similarity 0.85)
- SUGAR: A Scalable Human-Video-Driven Generalizable Humanoid Loco-Manipulation Learning Framework (NeurIPS 2026 · similarity 0.85)
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.
