Research map / Continual learning / Online and replay-based
Multi-world Model in Continual Reinforcement Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a Multi-world Model, a foundation-style world model extending World Models (VAE + RNN) to be trained across many environments for generalizable state-sequence prediction in continual RL; the paper is a proposal with no experiments.
Problem
World Models are environment-specific and do not generalize to unseen environments, which limits their use in continual reinforcement learning where catastrophic forgetting and task diversity are issues.
Approach
Build a single multi-environment world model from the World Models architecture (VAE vision + RNN memory), scaled up and possibly replacing the RNN with Transformers, and aligned with text representations of environment states, so it can predict future image sequences in both seen and novel environments.
Key results
- No experimental results are reported; the paper is a research proposal describing planned evaluation on Continual World against clonEx-SAC and PackNet.
Limitations
- No implementation or experiments; claims are prospective.
- Planned evaluation design and metrics are not yet defined in detail.
- Architecture choices (Transformer replacement, text alignment) are proposed but not validated.
Experimental setup
Baselines: clonEx-SAC PackNet
Builds on: World Models DreamerV3 Transformer Video Prediction by Efficient Transformers
Closest papers by meaning
- Continual Reinforcement Learning by Planning with Online World Models (ICML 2025 · similarity 0.84)
- Revisiting Multi-Agent World Modeling from a Diffusion-Inspired Perspective (NeurIPS 2025 · similarity 0.84)
- Grounded Answers for Multi-agent Decision-making Problem through Generative World Model (NeurIPS 2024 · similarity 0.83)
- Zero-shot World Models via Search in Memory (NeurIPS 2025 · similarity 0.83)
- RLVR-World: Training World Models with Reinforcement Learning (NeurIPS 2025 · similarity 0.83)
- PWM: Policy Learning with Multi-Task World Models (ICLR 2025 · similarity 0.82)
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.