Thesis defense Sep Thijssen (Donders series 246)

28 November 2016

Promotor: Prof. dr. B. Kappen

Path Integral Control

The objective of stochastic control theory is to find an external input in order to move a noisy system into a desired state. In certain cases, the optimal way to control the system can be described in a mathematically convenient manner, so that the control can be approximated by the Monte Carlo method. Because the control takes place over a certain time interval, the random samples take the form of paths. The involved computations for the optimal control are in a sense numerical approximations of integrals. Hence the name Path Integral Control.

In this thesis we show how the optimal control can be computed efficiently as a function of the state. Furthermore we prove that finding the optimal control, and optimizing the sampling procedure are mathematically the same problem.