Research map / Reinforcement learning / Visual and representation RL
Iso-Dream: Isolating and Leveraging Noncontrollable Visual Dynamics in World Models
Minting Pan, Xiangming Zhu, 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 |
|---|---|---|---|
| remove action-free branch (ablation) | Iso-Dream · BAIR bouncing balls, predict 18 frames | PSNR: 21.43 -> 20.47 | Hurt |
| remove action-free branch (ablation) | Iso-Dream · BAIR bouncing balls, predict 28 frames | PSNR: 19.51 -> 18.51 | Hurt |
| remove action-free branch (ablation) | Iso-Dream · BAIR bouncing balls, predict 28 frames | SSIM: 0.768 -> 0.690 | Hurt |
| remove Inverse cell (ablation) | Iso-Dream · BAIR bouncing balls, predict 18 frames | PSNR: 21.43 -> 21.42 | No clear effect |
| remove Inverse cell (ablation) | Iso-Dream · BAIR bouncing balls, predict 28 frames | PSNR: 19.51 -> 19.34 | Hurt |
| remove Inverse cell (ablation) | Iso-Dream · BAIR bouncing balls, predict 28 frames | SSIM: 0.768 -> 0.759 | Hurt |
| remove Inverse cell (ablation) | Iso-Dream · CARLA | episode return | Hurt |
| concatenate controllable and noncontrollable states directly instead of rolling out noncontrollable states with attention (ablation) | Iso-Dream policy · CARLA | episode return | Hurt |
| remove separate network branch for static time-invariant information (ablation) | Iso-Dream · CARLA | episode return: about 15% | Hurt |
| replace DreamerV2 with Iso-Dream | DMC video_easy agent · Walker walk | final return: 655 $\pm$ 47 -> 911 $\pm$ 50 | Improved |
| replace DreamerV2 with Iso-Dream | DMC video_easy agent · hopper stand | final return: 260 $\pm$ 366 -> 746 $\pm$ 312 | Improved |
| replace SVG with Iso-Dream | video predictor · BAIR bouncing balls | PSNR: 18.12 -> 19.51 | 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 4 of 4 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
- Zero-shot World Models via Search in Memory (NeurIPS 2025 · similarity 0.88)
- Learning Latent Dynamic Robust Representations for World Models (ICML 2024 · similarity 0.87)
- From Forecasting to Planning: Policy World Model for Collaborative State-Action Prediction (NeurIPS 2025 · similarity 0.86)
- DINO-WM: World Models on Pre-trained Visual Features enable Zero-shot Planning (ICML 2025 · similarity 0.86)
- DreamVLA: A Vision-Language-Action Model Dreamed with Comprehensive World Knowledge (NeurIPS 2025 · similarity 0.86)
- Act to See, See to Act: Diffusion-Driven Perception-Action Interplay for Adaptive Policies (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.
