Research map / Reinforcement learning / Visual and representation RL
ModSkill: Physical Character Skill Modularization
Yiming Huang, Zhiyang Dou, Lingjie Liu
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 per-body-part modular low-level controllers (ablation) | ModSkill · AMASS-Test full-body tracking | E_g-mpjpe (mm): 41.1 -> 35.9 | Improved |
| add skill modularization attention layer (ablation) | ModSkill · AMASS-Test full-body tracking | E_g-mpjpe (mm): 35.9 -> 32.4 | Improved |
| add generative adaptive sampling (ablation) | ModSkill · AMASS-Test full-body tracking | E_mpjpe (mm): 23.2 -> 22.7 | Improved |
| add generative adaptive sampling (ablation) | ModSkill · AMASS-Test full-body tracking | success rate: 99.3% -> 99.3% | No clear effect |
| use ModSkill instead of PHC+ for tracking | AMASS-Test · AMASS-Test full-body tracking | E_g-mpjpe (mm): 36.1 -> 32.2 | Improved |
| use five body parts instead of two (ablation) | ModSkill · AMASS-Test | E_g-mpjpe (mm): 36.4 -> 32.2 | Improved |
| use modular skills instead of distilled PULSE embeddings for downstream reach | ModSkill · Downstream reach | normalized return: 0.77 -> 0.79 | No clear effect |
| apply part-wise adversarial priors to ASE (ASE-PMP) | ASE · Downstream steering | normalized return: 0.60 -> 0.31 | Hurt |
| use ModSkill instead of PULSE in VR tracking | AMASS-Test VR tracking · AMASS-Test VR tracking | success rate: 93.4% -> 93.4% | No clear effect |
| use ModSkill instead of PHC+ on AMASS-Train success | AMASS-Train · AMASS-Train full-body tracking | success rate: 100% -> 99.6% | Hurt |
Closest papers by meaning
- NIL: No-data Imitation Learning (CVPR 2026 · similarity 0.88)
- ManiSkill2: A Unified Benchmark for Generalizable Manipulation Skills (ICLR 2023 · similarity 0.88)
- InterMimic: Towards Universal Whole-Body Control for Physics-Based Human-Object Interactions (CVPR 2025 · similarity 0.86)
- Universal Humanoid Motion Representations for Physics-Based Control (ICLR 2024 · similarity 0.86)
- Meta-Controller: Few-Shot Imitation of Unseen Embodiments and Tasks in Continuous Control (NeurIPS 2024 · similarity 0.85)
- AnySkill: Learning Open-Vocabulary Physical Skill for Interactive Agents (CVPR 2024 · 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.
