mlsanityData to 14 Oct 2026

ARB4WM: An Adversarial Robustness Benchmark for World Models in Continuous Control

arXiv 2606.16605 ↗First posted 15 Jun 2026cs.AIcited by 2

On the map: World models → World model security and GUI agents

also in 1 more
Abstract (arXiv)

World models are widely used in robotic and agentic engineering control systems due to their ability to learn latent dynamics for planning and decision-making. As these systems are increasingly deployed in safety-critical settings, understanding their robustness under adversarial conditions has become essential. However, existing evaluations lack a unified benchmark for testing adversarial threats across the policy, value, and latent-dynamics levels of world-model agents. To fill this gap, we present ARB4WM, a unified evaluation framework for pre-deployment robustness and risk assessment of world-model agents under visual perturbations. ARB4WM defines five white-box loss objectives across these three levels and studies their effects when combined with single-step or multi-step perturbation strategies and temporal attack modes, including full-frame, half-sequence, and sparse-frame exposure. Specifically, we evaluate four Dreamer-style agents across 20 tasks from MetaWorld and the DeepMind Control Suite under different loss objectives, perturbation strategies, and temporal attack modes. Results show that attacks targeting value estimation, latent representations, and RSSM dynamics can be as damaging as direct policy disruption, and that early or frequent perturbations are especially harmful, while input-level defenses provide limited recovery under adaptive attacks. These findings suggest that safety, risk, and reliability assessment for world models should cover multiple component-oriented attack objectives and temporal exposure protocols rather than relying solely on action-space robustness. Source code is available at https://github.com/zaoanguai/ARB4WM.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in World model security and GUI agents, one of 11 directions in World models. Builds on 38 of its 48 references that are in the map. First posted on arXiv 15 Jun 2026.

Builds on

38 of 48 references.

Closest ancestors · 1

Same direction, newest first

  1. 2026WMAttack: Automated Attack Search for Adversarial Evaluation of World-Model Agents22 May

Other references · 37

  1. 2018Representation Learning with Contrastive Predictive Codingcited by 7,777
  2. 2014Explaining and Harnessing Adversarial Examplescited by 7,364
  3. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,415
  4. 2013Intriguing properties of neural networkscited by 5,155
  5. 2019Dream to Control: Learning Behaviors by Latent Imaginationcited by 1,653
  6. 2018World Modelscited by 1,513
  7. 2020Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attackscited by 1,451
  8. 2019Certified Adversarial Robustness via Randomized Smoothingcited by 1,340
  9. 2019Meta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement Learningcited by 1,204
  10. 2020Mastering Atari with Discrete World Modelscited by 1,007
  11. 2016Practical Black-Box Attacks against Machine Learningcited by 903
  12. 2019Square Attack: a query-efficient black-box adversarial attack via random searchcited by 685
25 more
  1. 2020RobustBench: a standardized adversarial robustness benchmarkcited by 661
  2. 2017Adversarial Patchcited by 517
  3. 2023TD-MPC2: Scalable, Robust World Models for Continuous Controlcited by 512
  4. 2023Learning Interactive Real-World Simulatorscited by 458
  5. 2017Adversarial Examples: Attacks and Defenses for Deep Learningcited by 351
  6. 2019HopSkipJumpAttack: A Query-Efficient Decision-Based Attackcited by 333
  7. 2022Transformers are Sample-Efficient World Modelscited by 328
  8. 2017Adversarial Attacks on Neural Network Policiescited by 320
  9. 2020Robust Deep Reinforcement Learning against Adversarial Perturbations on State Observationscited by 238
  10. 2023OccWorld: Learning a 3D Occupancy World Model for Autonomous Drivingcited by 233
  11. 2019Adversarial Policies: Attacking Deep Reinforcement Learningcited by 214
  12. 2018Disentangling Adversarial Robustness and Generalizationcited by 172
  13. 2017Tactics of Adversarial Attack on Deep Reinforcement Learning Agentscited by 166
  14. 2019Adversarial Sensor Attack on LiDAR-based Perception in Autonomous Drivingcited by 152
  15. 2017Delving into adversarial attacks on deep policiescited by 104
  16. 2021DreamerPro: Reconstruction-Free Model-Based Reinforcement Learning with Prototypical Representationscited by 74
  17. 2024Advancing Humanoid Locomotion: Mastering Challenging Terrains with Denoising World Model Learningcited by 56
  18. 2023SafeDreamer: Safe Reinforcement Learning with World Modelscited by 54
  19. 2025Adversarial Attacks on Robotic Vision Language Action Modelscited by 32
  20. 2024Learning Latent Dynamic Robust Representations for World Modelscited by 21
  21. 2024Towards Robust Model-Based Reinforcement Learning Against Adversarial Corruptioncited by 12
  22. 2026R2-Dreamer: Redundancy-Reduced World Models without Decoders or Augmentationcited by 10
  23. 2024World Models: The Safety Perspectivecited by 9
  24. 2024Safe Exploration Using Bayesian World Models and Log-Barrier Optimizationcited by 4
  25. 2025World Model Robustness via Surprise Recognitioncited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Diagnosing JEPA World Models with Action-Conditioned Predictive Consistencycited by 3
  2. 2026Beyond Patch Removal: Persistent Adversarial Effects in Vision-Language-Action Policiescited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Safety, Security, and Cognitive Risks in World Models
  2. 2026BadWAM: When World-Action Models Dream Right but Act Wrong
  3. 2026BadWorld: Adversarial Attacks on World Models
  4. 2026RoAd-RL: A Unified Library and Benchmark for Robust Adversarial Reinforcement Learning
  5. 2026JailWAM: Jailbreaking World Action Models in Robot Control
  6. 2024Dissecting Adversarial Robustness of Multimodal LM Agents
  7. 2026Security of World-Model-Based Embodied AI: A Lifecycle of Threats, Defenses, and Evaluation
  8. 2026TAPDreamer: Transferable Adversarial Patches for World Action Models