Stochastic World Models for Verifying Vision-Based Neural Feedback Systems

2026-09-29

AI Summary

This study addresses the challenge of balancing high fidelity and analyzability in perception models for verifying vision-based neural feedback systems. We propose a stochastic world model grounded in physical latent variables to serve as a perception surrogate. By replacing conventional GANs with a verifiable stochastic world model and integrating a multi-strategy closed-loop verification framework, the approach combines falsification, adaptive refinement, and symbolic analysis to enable efficient state-space exploration. Experimental results demonstrate that the proposed model achieves superior reproduction accuracy with fewer parameters. Furthermore, in an emergency braking benchmark, it resolves over 80% of previously undetermined state spaces, significantly enhancing both the efficiency and reliability of system safety verification.

Abstract

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.

Problem

• Neural feedback systems
• Verification
• Stochastic world models
• Perception surrogates
• Closed-loop analysis

Innovation

• Stochastic World Models
• Perception Surrogates
• Neural Feedback Verification
• Falsification and Refinement
• Closed-loop Analysis