Research map / Reinforcement learning / Offline RL
Cross-Embodiment Offline Reinforcement Learning for Heterogeneous Robot Datasets
Haruki Abe, Takayuki Osa, Yusuke Mukuta, Tatsuya Harada
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 |
|---|---|---|---|
| use IQL instead of BC | cross-embodiment training · 70% Suboptimal Forward | final return: 30.52 -> 36.62 | Improved |
| use IQL instead of BC | cross-embodiment training · 70% Suboptimal Backward | final return: 41.42 -> 38.69 | Hurt |
| use IQL instead of BC | cross-embodiment training · Expert Forward/Backward | final return: 63.31 -> 63.39 | No clear effect |
| add embodiment grouping to IQL | IQL · 70% Suboptimal Forward | final return: 36.62 -> 51.19 | Improved |
| add embodiment grouping to IQL | IQL · Mean over six datasets | final return: 52.05 -> 57.29 | Improved |
| add embodiment grouping to IQL vs adding PCGrad or SEL | IQL · 70% Suboptimal datasets | relative improvement over IQL: 7.15% for PCGrad, 18.33% for SEL and 33.99% for EG | Improved |
| add embodiment grouping to TD3+BC | TD3+BC · 70% Suboptimal datasets | final return: +19.5% on average; mean 46.43 -> 51.39 | Improved |
| add embodiment grouping to BC | BC · 70% Suboptimal splits | final return: +26.3% | Improved |
| add embodiment grouping to IQL | IQL · Expert Forward | final return: 63.39 -> 63.52 | No clear effect |
| group robots by FGW embodiment distance vs random grouping (ablation) | IQL+EG · 70% Suboptimal Forward | final return: 38.73 -> 51.98 | Improved |
| group robots by FGW embodiment distance vs heuristic split (ablation) | IQL+EG · 70% Suboptimal Forward | final return: 34.45 -> 51.98 | Improved |
| add embodiment grouping with compute-normalized IQL baseline (ablation) | IQL · 70% Suboptimal Forward | final return: 44.20 -> 51.98 | Improved |
Closest papers by meaning
- PEAC: Unsupervised Pre-training for Cross-Embodiment Reinforcement Learning (NeurIPS 2024 · similarity 0.91)
- Uni-O4: Unifying Online and Offline Deep Reinforcement Learning with Multi-Step On-Policy Optimization (ICLR 2024 · similarity 0.90)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.89)
- Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models (NeurIPS 2025 · similarity 0.89)
- Robust Policy Learning via Offline Skill Diffusion (AAAI 2024 · similarity 0.88)
- Beyond OOD State Actions: Supported Cross-Domain Offline Reinforcement Learning (AAAI 2024 · similarity 0.88)
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.
