Stochastic Control: From Theory to Parallel Computation and Applications

Abstract

For autonomous systems to operate in stochastic environments, they have to be equipped with fast decision-making processes to reason about the best possible action. Grounded on first principles in stochastic optimal control theory and statistical physics, the path integral framework provides a mathematically sound methodology for decision making under uncertainty. It also creates opportunities for the development of novel sampling-based planning and control algorithms that are highly parallelizable. In this talk, I will present results in the area of sampling-based control that go beyond classical formulations and show applications to robotics and autonomous systems for tasks such as manipulation, grasping, and high-speed navigation. In addition to sampling-based stochastic control, alternative methods that rely on uncertainty propagation using stochastic variational integrators and polynomial chaos theory will be presented and their implications to trajectory optimization and state estimation will be demonstrated. At the end of this talk, and towards closing the gap between high-level reasoning/decision making and low-level organization/computation, I will highlight the interdependencies between theory, algorithms, and forms of computation and discuss future computational technologies in the area of autonomy and robotics.