Learning Stabilizing Policies in Stochastic Control Systems
This work addresses the challenge of ensuring stability in AI-driven control systems, but it is incremental as it builds on prior certification methods without achieving broad SOTA gains.
The paper tackles the problem of learning provably stable neural network policies for stochastic control systems by jointly optimizing a policy and a martingale certificate, finding that random initialization leads to local minima and pre-training is necessary for successful verification.
In this work, we address the problem of learning provably stable neural network policies for stochastic control systems. While recent work has demonstrated the feasibility of certifying given policies using martingale theory, the problem of how to learn such policies is little explored. Here, we study the effectiveness of jointly learning a policy together with a martingale certificate that proves its stability using a single learning algorithm. We observe that the joint optimization problem becomes easily stuck in local minima when starting from a randomly initialized policy. Our results suggest that some form of pre-training of the policy is required for the joint optimization to repair and verify the policy successfully.