# Optimal Control with Adversarial Events

advisor: | Doc. Dipl. Ing. Stefan Ratschan, Dr. Tech. |

e-mail: | show e-mail |

type: | phd thesis |

branch of study: | MI_MM, MI_AMSM, MINF |

link: | http://www2.cs.cas.cz/~ratschan/ |

description: | Assume an ordinary differential equation with inputs. Classical optimal control is the problem of finding inputs such that resulting solution of the differential equation optimizes a given objective function. The thesis will study an extension of this problem where some random events beyond our control can influence the differential equation, and the goal is to optimize the expected value of the objective function. This amounts to optimal control for a certain class of stochastic hybrid dynamical systems [1].
The goal of this thesis topic will be the design of an algorithm for solving this problem that can efficiently solve benchmark problems of a size that goes beyond the capabilities of current methods. Suggested basic ingredients will be: - Multiple shooting: a method that can solve highly non-trivial classical optimal control problems [2], but creates inputs that are not able to react to external events. - Approximate dynamic programming [3]: a method that efficiently solves the optimal control problem for Markov decision processes, which corresponds to the given problem, but without the presence of differential equations. |

references: | [1] Christos G. Cassandras, John Lygeros: Stochastic Hybrid Systems, CRC Press, 2006
[2] John T. Betts: Practical Methods for Optimal Control and Estimation Using Nonlinear Programming, SIAM, 2010 [3] Warren B. Powell: Approximate Dynamic Programming, Wiley, 2011 |

last update: | 27.04.2018 11:42:34 |

administrator for this page:
Ľubomíra Dvořáková | last update: 09/12/2011