Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The authors use compact stochastic world models with physically grounded latents—built from verifier-friendly operations—as perception surrogates to enable closed-loop verification of vision-based neural control systems. To verify these systems, they introduce a unified analysis procedure combining falsification, adaptive refinement, symbolic execution, and backward analysis. Their approach outperforms GAN surrogates using up to 130× fewer parameters, resolves 100% of an emergency braking benchmark that left 38% unresolved by prior SOTA, and successfully verifies over 80% of the state space on an RGB benchmark where verification was previously intractable.
Closed-loop vision verification is no longer bottlenecked by intractable generative surrogates: compact stochastic world models cut parameter counts by 130× while resolving over 80% of previously unverified RGB state spaces.
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.