Research map / Reinforcement learning / Visual and representation RL
RePo: Resilient Model-Based Reinforcement Learning by Regularizing Posterior Predictability
Chuning Zhu, Max Simchowitz, Siri Gadipudi, Abhishek Gupta
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 |
|---|---|---|---|
| replace reconstruction-based Dreamer with RePo posterior-predictability objective | visual model-based RL · Lazy TurtleBot | success rate (%): 0.0\% -> 62.5\% | Improved |
| replace reconstruction-based Dreamer with RePo | visual model-based RL · Lazy TurtleBot | return: -61.7 -> -24.3 | Improved |
| add a scoreboard distractor showing cumulative reward to the observation (ablation) | RePo · Distracted DMC Walker Walk | episode return: 868.72 -> 872.86 | No clear effect |
| replace support-matching constraint with distribution matching for test-time adaptation (ablation) | RePo test-time adaptation · Walker Stand, Walker Walk, Cheetah Run (static to distracted) | adaptation performance | Hurt |
| remove calibration (paired) examples from adaptation (ablation) | RePo test-time adaptation · Walker Stand, Walker Walk, Cheetah Run | adaptation performance | Hurt |
| add dynamics consistency (ILA) to distribution-matching adaptation | test-time adaptation · Walker Stand, Walker Walk, Cheetah Run | adaptation performance | No clear effect |
| use calibration alone for adaptation | test-time adaptation · Walker Stand, Walker Walk, Cheetah Run | adaptation performance | Hurt |
| vary information bottleneck KL target epsilon (too small) (ablation) | RePo · Distracted DMC Walker Walk | reward loss / return | Hurt |
| vary information bottleneck KL target epsilon (too large) (ablation) | RePo · Distracted DMC Walker Walk | return | Hurt |
| increase learning rate of Lagrange multiplier beta (too high) (ablation) | RePo · Distracted DMC Walker Walk | return | Hurt |
| lower learning rate of Lagrange multiplier beta (too low) (ablation) | RePo · Distracted DMC Walker Walk | return | Hurt |
| increase KL balancing parameter (train prior more) to a moderate level (ablation) | RePo · Distracted DMC Walker Walk | return | Improved |
Later papers that compared with this method
3 later accepted papers report RePO as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Discounted Beta–Bernoulli Reward Estimation for Sample-Efficient Reinforcement Learning with Verifiable Rewards (ICML 2026 · better on 4 of 4 shared results)
- VPA-WM: Vision-Priors-Aligned World Models for Robust Visual Reinforcement Learning (ECCV 2026 · better on 3 of 3 shared results)
- Buffer Matters: Unleashing the Power of Off-Policy Reinforcement Learning in Large Language Model Reasoning (ICLR 2026 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- The Surprising Ineffectiveness of Pre-Trained Visual Representations for Model-Based Reinforcement Learning (NeurIPS 2024 · similarity 0.91)
- VIGOR: Zero-Shot Visual Generalization via Latent-Space Consistency in Model-Based Reinforcement Learning (NeurIPS 2026 · similarity 0.91)
- Learning Latent Dynamic Robust Representations for World Models (ICML 2024 · similarity 0.89)
- Seek Commonality but Preserve Differences: Dissected Dynamics Modeling for Multi-modal Visual RL (NeurIPS 2024 · similarity 0.89)
- A Simple Framework for Generalization in Visual RL under Dynamic Scene Perturbations (NeurIPS 2024 · similarity 0.89)
- Focus On What Matters: Separated Models For Visual-Based RL Generalization (NeurIPS 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.
