Approximate Dynamic Programming Using Support Vector Regression

This paper presents a new approximate policy iteration algorithm based on support vector regression (SVR). It provides an overview of commonly used cost approximation architectures in approximate dynamic programming problems, explains some difficulties encountered by these architectures, and argues that SVR-based architectures can avoid some of these difficulties. A key contribution of this paper is to present an extension of the SVR problem to carry out approximate policy iteration by forcing the Bellman error to zero at selected states. The algorithm does not require trajectory simulations to be performed and is able to utilize a rich set of basis functions in a computationally efficient way. Computational results for an example problem are shown.