This academic article focuses on the development of neural network controllers for autonomous systems with unknown and stochastic dynamics. The authors propose a new approach to check and compute a temporal composition of trained neural network controllers that can satisfy complex tasks encoded with Linear Temporal Logic (LTL). The proposed method integrates automata theory and data-driven reachability analysis tools and provides safety guarantees. The results show that the system can generate safe behaviors for unseen complex temporal logic tasks in a zero-shot fashion. The effectiveness of the method is demonstrated through numerical simulations and hardware experiments on robot navigation tasks.
Publication date: 21 Nov 2023
Project Page: Not Provided
Paper: https://arxiv.org/pdf/2311.10863
