Safety Verification of Model Based Reinforcement Learning Controllers
This addresses safety verification for real-world systems like robotics and autonomous driving, but it is incremental as it builds on existing reachable set analysis methods.
The paper tackles the problem of verifying safety constraints for model-based reinforcement learning controllers, which is challenging due to the non-linear nature of neural networks, and presents a framework using reachable set analysis that can efficiently handle neural network models and identify safe initial states.
Model-based reinforcement learning (RL) has emerged as a promising tool for developing controllers for real world systems (e.g., robotics, autonomous driving, etc.). However, real systems often have constraints imposed on their state space which must be satisfied to ensure the safety of the system and its environment. Developing a verification tool for RL algorithms is challenging because the non-linear structure of neural networks impedes analytical verification of such models or controllers. To this end, we present a novel safety verification framework for model-based RL controllers using reachable set analysis. The proposed frame-work can efficiently handle models and controllers which are represented using neural networks. Additionally, if a controller fails to satisfy the safety constraints in general, the proposed framework can also be used to identify the subset of initial states from which the controller can be safely executed.