Dynamic Systems & Controls

In this presentation we present an iterative forward-backward algorithm which aims to solve discrete time stochastic optimal control problems. It is inspired by both max-plus numerical methods introduced by McEneaney and the Stochastic Dual Dynamic Programming (SDDP) algorithm of Pereira and Pinto.

At each time step, our algorithm builds two sequences of approximations of the (Bellman) value function respectively as a max-plus linear combinations of basic functions and as a min-plus linear combinations of other basic functions.

Designing autonomous systems that are simultaneously high-performing, adaptive, and provably safe remains an open problem.  In this talk, we will argue that in order to meet this goal, new theoretical and algorithmic tools are needed that blend the stability, robustness, and safety guarantees of robust control with the flexibility, adaptability, and performance of machine and reinforcement learning.  We will highlight our progress towards developing such a theoretical foundation of robust learning for safe control in the context of two case studies: (i) efficiently learning stability certif