Research map / Reinforcement learning / Visual and representation RL
Model-Based Reinforcement Learning With Isolated Imaginations
Minting Pan, Xiangming Zhu, Yitao Zheng, Yunbo Wang, Xiaokang Yang
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 Iso-Dream++ instead of DreamerV2 and Denoised-MDP | Iso-Dream++ · CARLA (500k steps) | average return: around 60 vs 10 (DreamerV2) and 25 (Denoised-MDP) | Improved |
| add inverse dynamics (Inverse Cell) (ablation) | Iso-Dream++ · CARLA | return | Improved |
| add rollouts of future noncontrollable states as action model input (ablation) | Iso-Dream++ · CARLA | return | Improved |
| add future-state attention instead of concatenation (ablation) | Iso-Dream++ · CARLA | return | Improved |
| add separate static-information branch (ablation) | Iso-Dream++ · CARLA | return: decreases by about 15% without it | Improved |
| add min-max variance constraints and sparse dependency modeling (ablation) | Iso-Dream++ · CARLA | return | Improved |
| add sparse dependency modeling (dependency gate) (ablation) | Iso-Dream++ · CARLA | return | Improved |
| rollout window tau 5 vs 10 vs 15 (ablation) | Iso-Dream++ · CARLA | return: no remarkable differences among tau in 5, 10, 15 | No clear effect |
| add rollout of noncontrollable states (tau 5 vs tau 0) (ablation) | Iso-Dream++ · CARLA | return: tau=0 significantly worse than tau in 5, 10, 15 | Improved |
| add min-max variance constraints (ablation) | Iso-Dream++ · DMC video_easy (10 seeds) | return: improves most tasks, especially Finger Spin | Improved |
| evaluate on video_hard after training on video_easy: Iso-Dream++ vs DreamerPro | Iso-Dream++ · DMC train video_easy test video_hard | return: Finger Spin 628 -> 692; Hopper Stand 180 -> 643; Walker Walk 533 -> 642; Cheetah Run 244 -> 441 | Improved |
| transfer pretrained action-free branch from Day to Night mode | Iso-Dream++ · CARLA Day to Night | return | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CURL — CURL: Contrastive Unsupervised Representations for Reinforcement Learning (ICML 2020) · better on 2 of 2 shared results
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
- Open-World Reinforcement Learning over Long Short-Term Imagination (ICLR 2025 · similarity 0.87)
- Reinforcing VLAs in Task-Agnostic World Models (NeurIPS 2026 · similarity 0.87)
- Dynamics-Aligned Latent Imagination in Contextual World Models for Zero-Shot Generalization (NeurIPS 2025 · similarity 0.87)
- Learning Latent Dynamic Robust Representations for World Models (ICML 2024 · similarity 0.86)
- Zero-shot World Models via Search in Memory (NeurIPS 2025 · similarity 0.86)
- AdaWM: Adaptive World Model based Planning for Autonomous Driving (ICLR 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.
