Research map / Reinforcement learning / Visual and representation RL
Beyond Single-Speed Reasoning: Coordinating Fast and Slow Dynamics for Efficient World Modeling
Hongwei Wang, Yangru Huang, Guangyao Chen, Xu Wang, Yi Jin
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 slow branch (ablation) | SF-RSSM · CARLA and DMC average | episode return: 523 -> 612 | Improved |
| add residual prediction in fast branch (ablation) | SF-RSSM · CARLA and DMC average | episode return: 542 -> 612 | Improved |
| add cross-timescale distillation (ablation) | SF-RSSM · CARLA and DMC average | episode return: 557 -> 612 | Improved |
| add curiosity module (ablation) | SF-RSSM · CARLA and DMC average | episode return: 575 -> 612 | Improved |
| use dual-timescale design instead of single-scale RSSM (ablation) | SF-RSSM · CARLA and DMC average | episode return: 460 -> 612 | Improved |
| make slow branch slow rather than a second fast branch (ablation) | SF-RSSM · CARLA Variable Weather, DMC Quadruped-run, Atari Amidar | return | Improved |
| replace DreamerV3 with SF-RSSM | DreamerV3 · CARLA Variable weather | episode return: 442 -> 606 | Improved |
| replace DreamerV3 with SF-RSSM | DreamerV3 · DMC Quadruped run | episode return: 586 -> 832 | Improved |
| replace TWISTER with SF-RSSM | TWISTER · CARLA HardRainNoon | episode return: 526 -> 614 | Improved |
| use SF-RSSM instead of DreamerV3 | DreamerV3 · Atari 100k Battle Zone | score: 12250 -> 14984 | Improved |
| use SF-RSSM instead of CURL | CURL · Atari 100k Battle Zone | score: 14870 -> 14984 | Improved |
| use SF-RSSM instead of DreamerV3 | DreamerV3 · DMC Finger turn hard | episode return: 962 -> 966 | No clear effect |
Closest papers by meaning
- Simplifying Latent Dynamics with Softly State-Invariant World Models (NeurIPS 2024 · similarity 0.89)
- Overcoming Slow Decision Frequencies in Continuous Control: Model-Based Sequence Reinforcement Learning for Model-Free Control (ICLR 2025 · similarity 0.88)
- PWM: Policy Learning with Multi-Task World Models (ICLR 2025 · similarity 0.88)
- Neurosymbolic World Models for Sequential Decision Making (ICML 2025 · similarity 0.88)
- ReLaX: Reasoning with Latent Exploration for Large Reasoning Models (CVPR 2026 · similarity 0.88)
- WPT: World-to-Policy Transfer via Online World Model Distillation (CVPR 2026 · 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.
