Research map / Reinforcement learning / Visual and representation RL
Recurrent World Models Facilitate Policy Evolution
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the MDN-RNN hidden state to the controller input alongside the VAE representation (ablation) | CarRacing-v0 controller · CarRacing-v0 | average score: 632 ± 251 -> 906 ± 21 | Helped |
| Add a hidden layer to the controller while using only the VAE representation (ablation) | CarRacing-v0 controller · CarRacing-v0 | average score: 632 ± 251 -> 788 ± 141 | Helped |
| Train the controller in a virtual environment with τ=0.10 rather than τ=1.15 (ablation) | DoomTakeCover-v0 World Model · DoomTakeCover-v0 actual environment | score: 1092 ± 556 -> 193 ± 58 | Hurt |
| Train the controller in a virtual environment with τ=0.50 rather than τ=1.15 (ablation) | DoomTakeCover-v0 World Model · DoomTakeCover-v0 actual environment | score: 1092 ± 556 -> 196 ± 50 | Hurt |
| Train the controller in a virtual environment with τ=1.00 rather than τ=1.15 (ablation) | DoomTakeCover-v0 World Model · DoomTakeCover-v0 actual environment | score: 1092 ± 556 -> 868 ± 511 | Hurt |
| Train the controller in a virtual environment with τ=1.30 rather than τ=1.15 (ablation) | DoomTakeCover-v0 World Model · DoomTakeCover-v0 actual environment | score: 1092 ± 556 -> 753 ± 139 | Hurt |
| Increase virtual-environment temperature to τ=1.15 | DoomTakeCover-v0 World Model · DoomTakeCover-v0 actual environment | score: 1092 ± 556 vs. 820 ± 58 | Helped |
Closest papers by meaning
- PWM: Policy Learning with Multi-Task World Models (ICLR 2025 · similarity 0.88)
- Neurosymbolic World Models for Sequential Decision Making (ICML 2025 · similarity 0.86)
- RLVR-World: Training World Models with Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- Recurrent Reinforcement Learning with Memoroids (NeurIPS 2024 · similarity 0.86)
- Continual Reinforcement Learning by Planning with Online World Models (ICML 2025 · similarity 0.86)
- Reinforcement World Model Learning for LLM-based Agents (NeurIPS 2026 · 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.