Stochastic Feedback Controller Design Considering the Dual Effect

Abstract

A new approach for the problem of nonlinear stochastic control is presented. A typical procedure for stochastic control takes a two step approach by separating estimation and control - estimate the system state using noise corrupted measurements, and control the system based on the estimated state. However, for nonlinear stochastic control systems and adaptive control systems with uncertain parameters, control and estimation are often not separable, since control inputs can afiect not only the system state, but also the quality of state estimation. Dynamic programming and search-based methods are general solution techniques for solving this type control problem, however, those solution techniques are often infeasible for real-time applications due to the huge computational requirements even for small problems. In this paper, a new suboptimal control algorithm is presented, which can be implemented as an online feedback controller for a stochastic control problem of moderate dimension. For an application example, automated docking of a nonholonomic vehicle using a bearing-only sensor is considered. The results from a series of numerical simulations are shown.