Research map / Reinforcement learning / Offline RL
Offline Reinforcement Learning of High-Quality Behaviors Under Robust Style Alignment
Mathieu Petitbois, Rémy Portelas, Sylvain Lamprier
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 style conditioning (CBC vs BC) (ablation) | BC · circle2d-inplace-v0 | style alignment (%): 29.1 -> 58.6 | Improved |
| add style conditioning (CBC vs BC) (ablation) | BC · humenv-complex-v0 | style alignment (%): 33.3 -> 47.1 | Improved |
| add hindsight style relabeling (SCBC vs CBC) (ablation) | CBC · circle2d-inplace-v0 | style alignment (%): 58.6 -> 68.6 | Improved |
| add hindsight style relabeling (SCBC vs CBC) (ablation) | CBC · humenv-simple-v0 | style alignment (%): 89.1 -> 99.6 | Improved |
| add hindsight style relabeling (SCBC vs CBC) (ablation) | CBC · halfcheetah-vary-v0 | style alignment (%): 52.0 -> 31.7 | Hurt |
| add hindsight style relabeling (SCBC vs CBC) (ablation) | CBC · humenv-complex-v0 | style alignment (%): 47.1 -> 33.2 | Hurt |
| add value learning to SCBC (SCIQL vs SCBC) (ablation) | SCBC · halfcheetah-vary-v0 | style alignment (%): 31.7 -> 78.9 | Improved |
| add value learning to SCBC (SCIQL vs SCBC) (ablation) | SCBC · humenv-complex-v0 | style alignment (%): 33.2 -> 83.5 | Improved |
| add value learning to SCBC (SCIQL vs SCBC) (ablation) | SCBC · humenv-simple-v0 | style alignment (%): 99.6 -> 99.6 | No clear effect |
| use SCIQL instead of CBC | CBC · halfcheetah-fix-v0 | style alignment (%): 51.2 -> 78.0 | Improved |
| add MI weighting (BCPMI vs CBC) (ablation) | CBC · halfcheetah-vary-v0 | style alignment (%): 52.0 -> 52.6 | No clear effect |
| add GAWR reward-based gating to SCIQL (ablation) | SCIQL · halfcheetah-stitch-v0 | style alignment (%) | Hurt |
Closest papers by meaning
- C2IQL: Constraint-Conditioned Implicit Q-learning for Safe Offline Reinforcement Learning (ICML 2025 · similarity 0.89)
- Asymmetric REINFORCE for off-Policy Reinforcement Learning: Balancing positive and negative rewards (NeurIPS 2025 · similarity 0.89)
- Adaptive $Q$-Aid for Conditional Supervised Learning in Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.88)
- Model-based Offline Reinforcement Learning with Lower Expectile Q-Learning (ICLR 2025 · similarity 0.88)
- A Perspective of Q-value Estimation on Offline-to-Online Reinforcement Learning (AAAI 2024 · similarity 0.88)
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · 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.
