Research map / Reinforcement learning / Visual and representation RL
Model-Based Reinforcement Learning via Imagination with Derived Memory
Yao Mu, Yuzheng Zhuang, Bin Wang, Guangxiang Zhu, Wulong Liu, Jianyu Chen, Ping Luo, Shengbo Eben Li, Chongjie Zhang, Jianye HAO
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 derived memory, R2S2 decoding constraint and evaluator (IDM vs Dreamer) | Dreamer · Walker-Walk, image uncertainty, 500K | average return: 876.3 ± 48 -> 914.9 ± 61 | Improved |
| add IDM to Dreamer | Dreamer · Hopper-Stand, image uncertainty, 500K | average return: 430.2 ± 128 -> 841.2 ± 54 | Improved |
| add IDM to Dreamer | Dreamer · Walker-Walk, no uncertainty, 100K | average return: 277 ± 12 -> 630 ± 113 | Improved |
| IDM vs model-free DrQ | DrQ · Cheetah-Run, no uncertainty, 500K | average return: 660 ± 96 vs 631 ± 103 | Hurt |
| IDM vs DrQ under test-time image rotation | DrQ · Walker-Walk, image uncertainty, 500K | average return: 151.9 ± 23 -> 914.9 ± 61 | Improved |
| remove trajectory evaluator (IDM w/o EVA) (ablation) | IDM · Hopper-Stand, image uncertainty | average return | Hurt |
| remove trajectory evaluator (IDM w/o EVA) (ablation) | IDM · Walker-Walk, image uncertainty | average return | No clear effect |
| remove derived memory, keep constraint and evaluator (IDM w/o DM) (ablation) | IDM · Walker-Walk, image uncertainty, 100K | improvement over Dreamer: 26.6% vs 41.6% | Hurt |
| remove decoding constraint and derived memory, keep evaluator only (ablation) | IDM · DMControl, image uncertainty | average return | No clear effect |
| double Dreamer batch size (Doubled-Dreamer) (ablation) | Dreamer · DMControl, image uncertainty | average return | No clear effect |
| derived memory contribution on stable Walker-Walk (ablation) | IDM · Walker-Walk | average return | Improved |
| evaluator contribution on unstable Hopper-Stand (ablation) | IDM · Hopper-Stand | average return | Improved |
Closest papers by meaning
- Mastering Memory Tasks with World Models (ICLR 2024 · similarity 0.88)
- Dynamics-Aligned Latent Imagination in Contextual World Models for Zero-Shot Generalization (NeurIPS 2025 · similarity 0.87)
- Open-World Reinforcement Learning over Long Short-Term Imagination (ICLR 2025 · similarity 0.87)
- DrM: Mastering Visual Reinforcement Learning through Dormant Ratio Minimization (ICLR 2024 · similarity 0.86)
- Stable Hadamard Memory: Revitalizing Memory-Augmented Agents for Reinforcement Learning (ICLR 2025 · similarity 0.86)
- Memo: Training Memory-Efficient Embodied Agents with Reinforcement Learning (NeurIPS 2025 · 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.
