This article presents an approach to validate the safety of autonomous trains using deep generative models. The authors focus on Operational Design Domain (ODD) validation, which describes the specific conditions under which a system is required to operate properly. They demonstrate the use of data simulation to create realistic test cases, particularly for a camera-based rail-scene segmentation system. The system’s performance is evaluated under different lighting and weather conditions. The authors conclude that deep generative models can effectively simulate complex data structures, making them useful for scenario-based testing of autonomous trains.
Publication date: 16 Oct 2023
Project Page: https://arxiv.org/abs/2310.10635v1
Paper: https://arxiv.org/pdf/2310.10635
