mlsanityData to 7 Oct 2026

World Model Robustness via Surprise Recognition

arXiv 2512.01119 ↗First posted 30 Nov 2025cs.LG cs.AIcited by 3

On the map: World models → Latent dynamics models and model-based RL

Abstract (arXiv)

AI systems deployed in the real world must contend with distractions and out-of-distribution (OOD) noise that can destabilize their policies and lead to unsafe behavior. While robust training can reduce sensitivity to some forms of noise, it is infeasible to anticipate all possible OOD conditions. To mitigate this issue, we develop an algorithm that leverages a world model's inherent measure of surprise to reduce the impact of noise in world model--based reinforcement learning agents. We introduce both multi-representation and single-representation rejection sampling, enabling robustness to settings with multiple faulty sensors or a single faulty sensor. While the introduction of noise typically degrades agent performance, we show that our techniques preserve performance relative to baselines under varying types and levels of noise across multiple environments within self-driving simulation domains (CARLA and Safety Gymnasium). Furthermore, we demonstrate that our methods enhance the stability of two state-of-the-art world models with markedly different underlying architectures: Cosmos and DreamerV3. Together, these results highlight the robustness of our approach across world modeling domains. We release our code at https://github.com/Bluefin-Tuna/WISER .

Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Placed in Latent dynamics models and model-based RL, one of 11 directions in World models. Builds on 20 of its 37 references that are in the map. First posted on arXiv 30 Nov 2025.

Builds on

20 of 37 references.

Closest ancestors · 4

Same direction, newest first

  1. 2024Learning Latent Dynamic Robust Representations for World Models10 May
  2. 2023Novelty Detection in Reinforcement Learning with World Models12 Oct
  3. 2023SafeDreamer: Safe Reinforcement Learning with World Models14 Jul
  4. 2018Recurrent World Models Facilitate Policy Evolution4 Sep

Other references · 16

  1. 2017CARLA: An Open Urban Driving Simulatorcited by 2,431
  2. 2021Restormer: Efficient Transformer for High-Resolution Image Restorationcited by 1,618
  3. 2019Leveraging Procedural Generation to Benchmark Reinforcement Learningcited by 478
  4. 2022Diffusion Models for Adversarial Purificationcited by 469
  5. 2020Learning Invariant Representations for Reinforcement Learning without Reconstructioncited by 424
  6. 2021Benchmarking the Spectrum of Agent Capabilitiescited by 193
  7. 2023Safety-Gymnasium: A Unified Safe Reinforcement Learning Benchmarkcited by 135
  8. 2023Robust Evaluation of Diffusion-Based Adversarial Purificationcited by 65
  9. 2022Mask-based Latent Reconstruction for Reinforcement Learningcited by 39
  10. 2017Learning End-to-end Multimodal Sensor Policies for Autonomous Navigationcited by 16
  11. 2024CarDreamer: Open-Source Learning Platform for World Model based Autonomous Drivingcited by 13
  12. 2021The Role of Pretrained Representations for the OOD Generalization of Reinforcement Learning Agentscited by 12
4 more
  1. 2023MaDi: Learning to Mask Distractions for Generalization in Visual Deep Reinforcement Learningcited by 9
  2. 2024LoRID: Low-Rank Iterative Diffusion for Adversarial Purificationcited by 7
  3. 2024Policy-shaped prediction: avoiding distractions in model-based reinforcement learningcited by 6
  4. 2025Continual Reinforcement Learning via Autoencoder-Driven Task and New Environment Recognitioncited by 5

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 3.

3 papers.

  1. 2026WMAttack: Automated Attack Search for Adversarial Evaluation of World-Model Agentscited by 6
  2. 2026Learning from Contrasts: Synthesizing Reasoning Paths from Diverse Search Trajectoriescited by 4
  3. 2026ARB4WM: An Adversarial Robustness Benchmark for World Models in Continuous Controlcited by 2

Similar, not cited

Close in wording, no citation either way

  1. 2026Safety, Security, and Cognitive Risks in World Models
  2. 2026When World Models Lie: Adaptive Safety Analysis Under Wrong Imaginations
  3. 2023Reward-Free Curricula for Training Robust World Models
  4. 2025World Models for Anomaly Detection during Model-Based Reinforcement Learning Inference
  5. 2023Exploring the Potential of World Models for Anomaly Detection in Autonomous Driving
  6. 2026Modeling Latent Disturbances for Robust Decision-Making in World Models
  7. 2026What Capable Agents Must Know: Selection Theorems for Robust Decision-Making under Uncertainty
  8. 2022Denoised MDPs: Learning World Models Better Than the World Itself