mlsanityData to 7 Oct 2026

Latent dynamics models and model-based RL

298 papers5 this weekHigh confidence in this grouping: the papers form one clear topicHigh confidence in this grouping: the papers form one clear topic · About

Keywords: reinforcement, reinforcement learning, model-based, policy, performance

Key facts. Latent dynamics models and model-based RL: 298 papers on arXiv since 2015 (data through 7 Oct 2026). 153 of its 298 mapped papers (51%) cite “Dream to Control: Learning Behaviors by Latent Imagination” (2019, arXiv 1912.01603). 129 of its 298 mapped papers (43%) cite “Mastering Atari with Discrete World Models” (2020, arXiv 2010.02193). 23 new papers 17 Sep–7 Oct.

What’s new

17 Sep – 7 Oct 2026

23 new papers, 17 Sep – 7 Oct; 5 this week.

  1. This week · 5
  2. World-Model Policy Arbiter for Goal-Conditioned Reinforcement Learning
  3. RIWANav: Recursive World-Action Models with Self-Improvement for Urban Navigation also fits: World models for navigation
  4. Learning in Dreams, Winning in Reality: A Continuous Dyna Loop for a Ten-Hero MOBA also fits: World models for language agents
  5. Independent Multi-Agent Reinforcement Learning with Counterfactual Semantic-Social World Models
  6. Considering Context: When World Models Need Context Encoders
  7. Earlier · 18
  8. Action-Consequence Alignment for Reliable Planning and Self-Improving in Latent World Models also fits: Latent predictive world models
  9. DreamTest: World-Model Surrogates for Search-Based Testing of Deep Reinforcement Learning Agents
  10. Network World Models as Environments for Algorithm Design on Complex Systems also fits: World models for language agents
  11. Calibration-risk routing for controlled world-model adaptation
  12. DashVMC: Real-Time Discrete World Model Control in Geometry Dash also fits: Latent predictive world models
All 23
  1. Beyond Policy Alignment: Closing the Planning-Learning Loop for Robot Control with Learned World Models also fits: Humanoid and legged locomotion world models
  2. Linear Recurrent Memory Suffices to Distil a World-Model Policy for Robot Air Hockey also fits: Latent predictive world models
  3. Beyond Prediction: Steering VLM Agents with Retrospective World Modeling
  4. Code to Control: Synthesizing Parameterized Reactive Controllers
  5. Think Fast, Plan Selectively: Adaptive Deliberation for Efficient Data-Driven MPC
  6. MA-JEPA: Joint-Embedding World Models for Multi-Agent Reinforcement Learning
  7. Beyond Conservatism: Recoverability-Conditioned Exploration for Model-Based Imitation Learning
  8. Does Learning to Predict the World Help Agents Act? Auditing World-Model Post-Training also fits: World models for language agents
  9. SwingRL: Adaptive Observation Reinforcement Learning with World-Model Prediction for Cable-Suspended Hoisting Control
  10. MA-WAM: Multi-Agent World-Action Model for Test-Time Planning also fits: World models for language agents
  11. Recommendation World Models for Future-State Control also fits: World models for language agents
  12. Prioritized Rollouts for Efficient World Model-based Vision-Language-Action Policy Optimization also fits: Embodied and robot world models
  13. Adaptive Rollout Truncation Based on Epistemic Uncertainty for Efficient Offline World Model Training also fits: Latent predictive world models

Foundations

Most cited by papers here

  1. Dream to Control: Learning Behaviors by Latent Imagination2019cited by 153 here
  2. Mastering Atari with Discrete World Models2020cited by 129 here
  3. Mastering Diverse Domains through World Models2023cited by 115 here
  4. Recurrent World Models Facilitate Policy Evolution2018cited by 83 here
  5. World Models2018cited by 77 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2017Neural Network Dynamics for Model-Based Deep Reinforcement Learning with Model-Free Fine-Tuningcited by 29 here
  2. 2018Model-Ensemble Trust-Region Policy Optimizationcited by 29 here
  3. 2018Deep Reinforcement Learning in a Handful of Trials using Probabilistic Dynamics Modelscited by 62 here
  4. 2018Learning Latent Dynamics for Planning from Pixelscited by 134 here
  5. 2019Dream to Control: Learning Behaviors by Latent Imaginationcited by 153 here
  6. 2020Planning to Explore via Self-Supervised World Modelscited by 47 here
  7. 2020Mastering Atari with Discrete World Modelscited by 129 here
  8. 2021Mastering Atari Games with Limited Datacited by 36 here
  9. 2022Transformers are Sample-Efficient World Modelscited by 61 here
  10. 2023Mastering Diverse Domains through World Modelscited by 115 here
  11. 2023STORM: Efficient Stochastic Transformer based World Models for Reinforcement Learningcited by 32 here
  12. 2024Diffusion for World Modeling: Visual Details Matter in Ataricited by 30 here
12 more it builds on, not in this chain
  1. 2018World Modelscited by 77 here
  2. 2018Recurrent World Models Facilitate Policy Evolutioncited by 83 here
  3. 2019When to Trust Your Model: Model-Based Policy Optimizationcited by 64 here
  4. 2020Language Models are Few-Shot Learnerscited by 20 here
  5. 2020Denoising Diffusion Probabilistic Modelscited by 22 here
  6. 2022TransDreamer: Reinforcement Learning with Transformer World Modelscited by 25 here
  7. 2022Temporal Difference Learning for Model Predictive Controlcited by 47 here
  8. 2022DayDreamer: World Models for Physical Robot Learningcited by 25 here
  9. 2023Diffusion Policy: Visuomotor Policy Learning via Action Diffusioncited by 18 here
  10. 2023Transformer-based World Models Are Happy With 100k Interactionscited by 45 here
  11. 2023TD-MPC2: Scalable, Robust World Models for Continuous Controlcited by 55 here
  12. 2024Genie: Generative Interactive Environmentscited by 18 here

The 24 most cited of 328.

Browse

30 of 298, ranked 6–35

  1. 6When to Trust Your Model: Model-Based Policy Optimization201964
  2. 7Deep Reinforcement Learning in a Handful of Trials using Probabilistic Dynamics Models201862
  3. 8Transformers are Sample-Efficient World Models202261
  4. 9TD-MPC2: Scalable, Robust World Models for Continuous Control202355
  5. 10Planning to Explore via Self-Supervised World Models202047
  6. 11Transformer-based World Models Are Happy With 100k Interactions202345
  7. 12Mastering Atari Games with Limited Data202136
  8. 13STORM: Efficient Stochastic Transformer based World Models for Reinforcement Learning202332
  9. 14Diffusion for World Modeling: Visual Details Matter in Atari202430
  10. 15Model-Ensemble Trust-Region Policy Optimization201829
All 30
  1. 16Neural Network Dynamics for Model-Based Deep Reinforcement Learning with Model-Free Fine-Tuning201729
  2. 17TransDreamer: Reinforcement Learning with Transformer World Models202225
  3. 18DayDreamer: World Models for Physical Robot Learning202225
  4. 19DreamerPro: Reconstruction-Free Model-Based Reinforcement Learning with Prototypical Representations202116
  5. 20Value Prediction Network201715
  6. 21Offline Reinforcement Learning as One Big Sequence Modeling Problem202113
  7. 22Dreaming: Model-based Reinforcement Learning by Latent Imagination without Reconstruction202012
  8. 23Algorithmic Framework for Model-based Deep Reinforcement Learning with Theoretical Guarantees201812
  9. 24Learning Transformer-based World Models with Contrastive Predictive Coding202511
  10. 25Efficient World Models with Context-Aware Tokenization202411
  11. 26Training Agents Inside of Scalable World Models20259
  12. 27EfficientZero V2: Mastering Discrete and Continuous Control with Limited Data20249
  13. 28Iso-Dream: Isolating and Leveraging Noncontrollable Visual Dynamics in World Models20229
  14. 29Exploring Model-based Planning with Policy Networks20199
  15. 30Diffusion World Model: Future Modeling Beyond Step-by-Step Rollout for Offline Reinforcement Learning20248
  16. 31SafeDreamer: Safe Reinforcement Learning with World Models20238
  17. 32On Learning to Think: Algorithmic Information Theory for Novel Combinations of Reinforcement Learning Controllers and Recurrent Neural World Models20158
  18. 33Improving Token-Based World Models with Parallel Observation Prediction20247
  19. 34Policy-Guided Diffusion20247
  20. 35Pre-training Contextualized World Models with In-the-wild Videos for Reinforcement Learning20237