Скачать или смотреть Lars Grune: Using Redundancy of the Dynamics in Nonlinear Optimal Feedback Control

Date: 15 June 2021
Speaker: Lars Grune
Title: Using Redundancy of the Dynamics in Nonlinear Optimal Feedback Control
Abstract: Optimal Feedback Control is a computationally intensive task. The classical approach to a numerical solution of this problem via the Hamilton-Jacobi-Bellman PDE in continuous time or the Bellman equation in discrete time may fail already for very low dimensional systems because of its high computational complexity.

In contrast to this, modern optimisation based control methods like model predictive control or deep reinforcement learning are able to solve such problems approximately with much larger state dimension. In this talk we argue that this is possible because the dynamics of the problem exhibits some kind of redundancy, which can efficiently be exploited by these algorithms. We will explain which kind of redundancy this can be and why it helps to reduce the computational complexity of these methods.