mlsanityData to 7 Oct 2026

Stochastic World Models for Verifying Vision-Based Neural Feedback Systems

arXiv 2609.38120 ↗First posted 29 Sep 2026cs.AI cs.SY eess.SYcited by 0

On the map: World models → Latent predictive world models

Abstract (arXiv)

Verifying a vision-based neural feedback system requires a model of the observations its controller acts upon. Such a model must capture the variation the sensor produces, while remaining tractable for closed-loop analysis. Generative adversarial networks (GANs) have served as perception surrogates, but they are large, reproduce complex scenes poorly, and are hard to verify. We explore stochastic world models as a richer class of perception surrogates. We train a world model with physically grounded latents, built from operations that standard verifiers bound. It reproduces held-out frames more faithfully than GAN surrogates with up to 130 times as many parameters. To verify these surrogates, we develop a procedure that combines falsification, adaptive refinement, symbolic, and backward analyses. On an emergency braking benchmark with a GAN surrogate, our procedure resolves the entire state space, 38% of which the state-of-the-art verifier left unresolved. On the RGB version of the benchmark, where no verification results have previously been reported, our procedure resolves over 80% of the state space with a world model surrogate.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Latent predictive world models, one of 11 directions in World models. Builds on 32 of its 41 references that are in the map. First posted on arXiv 29 Sep 2026.

Builds on

32 of 41 references.

Closest ancestors · 2

Same direction, newest first

  1. 2025Four Principles for Physically Interpretable World Models4 Mar
  2. 2024Physically Interpretable World Models via Weakly Supervised Representation Learning17 Dec

Other references · 30

  1. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 6,979
  2. 2017FiLM: Visual Reasoning with a General Conditioning Layercited by 2,286
  3. 2014Conditional Generative Adversarial Netscited by 2,152
  4. 2015Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networkscited by 2,048
  5. 2020Mastering Atari with Discrete World Modelscited by 990
  6. 2018Self-Attention Generative Adversarial Networkscited by 809
  7. 2023DriveDreamer: Towards Real-world-driven World Models for Autonomous Drivingcited by 258
  8. 2021Beta-CROWN: Efficient Bound Propagation with Per-neuron Split Constraints for Complete and Incomplete Neural Network Robustness Verificationcited by 224
  9. 2017Hamilton-Jacobi Reachability: A Brief Overview and Recent Advancescited by 186
  10. 2020Fast and Complete: Enabling Complete Neural Network Verification with Rapid and Massively Parallel Incomplete Verifierscited by 183
  11. 2023Real-World Humanoid Locomotion with Reinforcement Learningcited by 132
  12. 2019An Abstraction-Based Framework for Neural Network Verificationcited by 68
18 more
  1. 2021OVERT: An Algorithm for Safety Verification of Neural Network Control Policies for Nonlinear Systemscited by 37
  2. 2017Compositional Falsification of Cyber-Physical Systems with Machine Learning Componentscited by 31
  3. 2021Verification of Image-based Neural Network Controllers Using Generative Modelscited by 27
  4. 2019Guaranteeing Safety for Neural Network-Based Aircraft Collision Avoidance Systemscited by 26
  5. 2022Open- and Closed-Loop Neural Network Verification using Polynomial Zonotopescited by 24
  6. 2021Verifying Controllers with Convolutional Neural Network-based Perception: A Case for Intelligible, Safe, and Precise Abstractionscited by 19
  7. 2022NNLander-VeriF: A Neural Network Formal Verification Framework for Vision-Based Autonomous Aircraft Landingcited by 19
  8. 2022Backward Reachability Analysis of Neural Feedback Loops: Techniques for Linear and Nonlinear Systemscited by 18
  9. 2026Foundational World Models Accurately Detect Bimanual Manipulator Failurescited by 11
  10. 2024Scalable Surrogate Verification of Image-based Neural Network Control Systems using Composition and Unrollingcited by 8
  11. 2024Constraint-Aware Refinement for Safety Verification of Neural Feedback Loopscited by 4
  12. 2025Polyhedral Enclosures: An Efficient Combinatorial Abstraction for Nonlinear Neural Feedback Systemscited by 3
  13. 2026The Era of End-to-End Autonomy: Transitioning from Rule-Based Driving to Large Driving Modelscited by 3
  14. 2025Scalable and Interpretable Verification of Image-based Neural Network Controllers for Autonomous Vehiclescited by 2
  15. 2025Deterministic World Models for Closed-loop Reachability Analysis of End-to-End Vision-based Controlcited by 2
  16. 2025RampoNN: A Reachability-Guided System Falsification for Efficient Cyber-Kinetic Vulnerability Detectioncited by 1
  17. 2026The FABRIC Strategy for Verifying Neural Feedback Systemscited by 1
  18. 2026Counterexample Guided Branching via Directional Relaxation Analysis in Complete Neural Network Verificationcited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2020RetinaGAN: An Object-aware Approach to Sim-to-Real Transfer
  2. 2024Towards Evaluating the Robustness of Visual State Space Models
  3. 2025RoboMonkey: Scaling Test-Time Sampling and Verification for Vision-Language-Action Models
  4. 2026TaskSense: Focusing on What Matters in World Models
  5. 2026Contrastive World Models
  6. 2018Sample-Efficient Imitation Learning via Generative Adversarial Nets
  7. 2024Interpretable Imitation Learning via Generative Adversarial STL Inference and Control
  8. 2026ComplexityWorld: Benchmarking Vision-Language Models on Verifiable Visual Decision Making