Research map / Reinforcement learning / Visual and representation RL
Learning to Assist: Physics-Grounded Human-Human Control via Multi-Agent Reinforcement Learning
Yuto Shibata, Kashu Yamazaki, Lalit Jayanti, Yoshimitsu Aoki, Mariko Isogawa, Katerina Fragkiadaki
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 |
|---|---|---|---|
| joint MARL training instead of sequential training with a frozen recipient (ablation) | AssistMimic · Inter-X specialist | Success rate (%): 62.4 -> 74.9 | Improved |
| initialize policies from single-person motion prior (ablation) | AssistMimic · Inter-X specialist | Success rate (%): 0.0 -> 74.9 | Improved |
| initialize policies from single-person motion prior (ablation) | AssistMimic · HHI-Assist specialist | Success rate (%): 19.1 -> 85.8 | Improved |
| add dynamic reference retargeting (ablation) | AssistMimic · Inter-X specialist | Success rate (%): 83.4 -> 74.9 | Hurt |
| add dynamic reference retargeting (ablation) | AssistMimic · HHI-Assist specialist | Success rate (%), seen dynamics: 85.4 -> 85.8 | No clear effect |
| add dynamic reference retargeting (ablation) | AssistMimic · HHI-Assist specialist (unseen dynamics) | Success rate (%): 49.1 -> 67.8 (mass x1.5); 62.9 -> 73.2 (hip torque x0.5) | Improved |
| add contact-promoting reward (ablation) | AssistMimic · HHI-Assist specialist | Success rate (%), seen dynamics: 97.7 -> 85.8 | Hurt |
| add contact-promoting reward (ablation) | AssistMimic · HHI-Assist specialist (unseen dynamics) | Success rate (%): 56.4 -> 67.8 (mass x1.5); 27.7 -> 73.2 (hip torque x0.5) | Improved |
| add contact-promoting reward (ablation) | AssistMimic · Inter-X specialist | Success rate (%): seen 81.6 -> 74.9; mass x1.2 66.3 -> 57.9; Kp/Kd x0.5 77.1 -> 72.8 | Mixed |
| distill specialists into a generalist with DAgger (ablation) | AssistMimic generalist · Inter-X generalist (30 clips) | Success rate (%): 77.3 -> 94.7 | Improved |
Closest papers by meaning
- CooHOI: Learning Cooperative Human-Object Interaction with Manipulated Object Dynamics (NeurIPS 2024 · similarity 0.87)
- Universal Humanoid Motion Representations for Physics-Based Control (ICLR 2024 · similarity 0.87)
- Habitat 3.0: A Co-Habitat for Humans, Avatars, and Robots (ICLR 2024 · similarity 0.86)
- SUGAR: A Scalable Human-Video-Driven Generalizable Humanoid Loco-Manipulation Learning Framework (NeurIPS 2026 · similarity 0.86)
- Humanoid Horizon: Extending Task Horizon in Whole-Body Loco-Manipulation via Parallel Training, Dynamic Starting, and Reward Gating (NeurIPS 2026 · similarity 0.86)
- AnySkill: Learning Open-Vocabulary Physical Skill for Interactive Agents (CVPR 2024 · 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.
