Learning Vision-Based Neural Network Controllers with Semi-Probabilistic Safety Guarantees
Xinhang Ma, Junlin Wu, Hussein Sibai, Yiannis Kantaros, Yevgeniy Vorobeychik
TL;DR
This work tackles the challenge of ensuring safety for vision-based autonomous control by introducing a semi-probabilistic verification (SPV) framework that couples reachability analysis with a conditional observation generator to model unknown perception. It then presents a gradient-based learning pipeline that optimizes a safety-aware loss, uses adaptive data selection, and employs curriculum learning to synthesize provably safe controllers while preserving nominal performance. Key contributions include formal SPV guarantees that bound safety over a distribution of initial states, a differentiable proxy for the safety objective, and an adaptive training regimen that emphasizes safety-critical states; these are validated across X-Plane, CARLA, a physical F1Tenth setup, and AirSim drone scenarios. The results demonstrate stronger probabilistic safety guarantees and competitive nominal performance, highlighting the practical potential of SPV for scalable safety in vision-based control, with acknowledged limitations like grayscale, low-resolution inputs and relatively small latent dimensions that future work could address.
Abstract
Ensuring safety in autonomous systems with vision-based control remains a critical challenge due to the high dimensionality of image inputs and the fact that the relationship between true system state and its visual manifestation is unknown. Existing methods for learning-based control in such settings typically lack formal safety guarantees. To address this challenge, we introduce a novel semi-probabilistic verification framework that integrates reachability analysis with conditional generative networks and distribution-free tail bounds to enable efficient and scalable verification of vision-based neural network controllers. Next, we develop a gradient-based training approach that employs a novel safety loss function, safety-aware data-sampling strategy to efficiently select and store critical training examples, and curriculum learning, to efficiently synthesize safe controllers in the semi-probabilistic framework. Empirical evaluations in X-Plane 11 airplane landing simulation, CARLA-simulated autonomous lane following, F1Tenth vehicle lane following in a physical visually-rich miniature environment, and Airsim-simulated drone navigation and obstacle avoidance demonstrate the effectiveness of our method in achieving formal safety guarantees while maintaining strong nominal performance.