mlsanityData to 7 Oct 2026

Learning Implicit Causal World Models from Multi-Agent Demonstrations

arXiv 2607.26336 ↗First posted 28 Jul 2026cs.LG cs.MA cs.ROcited by 0

On the map: World models → World models for language agents

Abstract (arXiv)

In model-based reinforcement learning, world models exist as internal simulators, but their training often conflates statistical correlations with causal mechanisms. This problem is exacerbated in multi-agent systems where physical transitions are intertwined with strategic agent intents, causing world models to fail under distribution shift. We introduce Implicit Causal World Models to recover environmental dynamics from offline demonstrations without requiring pre-defined causal graphs. By incorporating policy variance, we render world models discoverable via the sequential backdoor condition. Evaluations across coordination tasks (Two-Door, Navigation, and Giveway) demonstrate that these models provide interpretable causal representations under both full and partial observability, with model accuracy scaling directly with interventional strength.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in World models for language agents, one of 11 directions in World models. Builds on 46 of its 94 references that are in the map. First posted on arXiv 28 Jul 2026.

Builds on

46 of 94 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025What Has a Foundation Model Found? Using Inductive Bias to Probe for World Models9 Jul
  2. 2024Evaluating the World Model Implicit in a Generative Model6 Jun
  3. 2024COMBO: Compositional World Models for Embodied Multi-Agent Cooperation16 Apr
  4. 2024Robust agents learn causal world models16 Feb

Other references · 42

  1. 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,064
  2. 2016Categorical Reparameterization with Gumbel-Softmaxcited by 2,708
  3. 2019Invariant Risk Minimizationcited by 1,832
  4. 2017Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environmentscited by 1,519
  5. 2018Learning Latent Dynamics for Planning from Pixelscited by 1,221
  6. 2023Mastering Diverse Domains through World Modelscited by 1,193
  7. 2016The Concrete Distribution: A Continuous Relaxation of Discrete Random Variablescited by 1,135
  8. 2023Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecturecited by 982
  9. 2021The Surprising Effectiveness of PPO in Cooperative, Multi-Agent Gamescited by 903
  10. 2018Recurrent World Models Facilitate Policy Evolutioncited by 761
  11. 2018DAGs with NO TEARS: Continuous Optimization for Structure Learningcited by 745
  12. 2024Genie: Generative Interactive Environmentscited by 706
30 more
  1. 2015Causal inference using invariant prediction: identification and confidence intervalscited by 533
  2. 2017Safe Reinforcement Learning via Shieldingcited by 427
  3. 2022Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Taskcited by 417
  4. 2024DINO-WM: World Models on Pre-trained Visual Features enable Zero-shot Planningcited by 357
  5. 2019Causal Confusion in Imitation Learningcited by 286
  6. 2020Deep Structural Causal Models for Tractable Counterfactual Inferencecited by 210
  7. 2021Towards Causal Representation Learningcited by 207
  8. 2019Contrastive Learning of Structured World Modelscited by 206
  9. 2019Multi-Agent Pathfinding: Definitions, Variants, and Benchmarkscited by 197
  10. 2020Discovering contemporaneous and lagged causal relations in autocorrelated nonlinear time series datasetscited by 136
  11. 2020CausalWorld: A Robotic Manipulation Benchmark for Causal Structure and Transfer Learningcited by 109
  12. 2022VMAS: A Vectorized Multi-Agent Simulator for Collective Robot Learningcited by 63
  13. 2023TorchRL: A data-driven decision-making library for PyTorchcited by 59
  14. 2020Confounding-Robust Policy Evaluation in Infinite-Horizon Reinforcement Learningcited by 56
  15. 2022Neural Causal Models for Counterfactual Identification and Estimationcited by 54
  16. 2019Causal Reasoning from Meta-reinforcement Learningcited by 53
  17. 2020CausalVAE: Structured Causal Disentanglement in Variational Autoencodercited by 47
  18. 2025Evaluating Newtonian Mechanics in Video Generative Models with Real Physical Systemscited by 44
  19. 2020Model-based Reinforcement Learning: A Surveycited by 39
  20. 2023Amortized Variational Inference: When and Why?cited by 28
  21. 2019Importance Resampling for Off-policy Predictioncited by 26
  22. 2022Sequential Causal Imitation Learning with Unobserved Confounderscited by 25
  23. 2026Causal-JEPA: Learning World Models through Object-Level Latent Maskingcited by 25
  24. 2022Offline Reinforcement Learning with Causal Structured World Modelscited by 19
  25. 2025Revisiting Multi-Agent World Modeling from a Diffusion-Inspired Perspectivecited by 9
  26. 2026ShareVerse: Multi-Agent Consistent Video Generation for Shared World Modelingcited by 9
  27. 2025What's Producible May Not Be Reachable: Measuring the Steerability of Generative Modelscited by 5
  28. 2022Causal discovery for time series with latent confounderscited by 4
  29. 2024Generative Modeling of Neural Dynamics via Latent Stochastic Differential Equationscited by 4
  30. 2025Energy-Based Transfer for Reinforcement Learningcited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2024Boosting Efficiency in Task-Agnostic Exploration through Causal Knowledge
  2. 2025Curious Causality-Seeking Agents Learn Meta Causal World
  3. 2023Explainable Reinforcement Learning via a Causal World Model
  4. 2026A Unifying Perspective on Causal World Models: From Observations to Representations to Structure
  5. 2022Variational Causal Dynamics: Discovering Modular World Models from Interventions
  6. 2025Better Decisions through the Right Causal World Model
  7. 2020Causal World Models by Unsupervised Deconfounding of Physical Dynamics
  8. 2025Causal Process Models: Reframing Dynamic Causal Graph Discovery as a Reinforcement Learning Problem