Summary: The Center for Control, Dynamical Systems, and Computation
University of California at Santa Barbara
Fall 2006 Seminars
Constant-Factor Approximation Algorithms for
by Sanjay Lall
Friday, October 20th, 2006 4:00 - 5:00 PM ESB 2001
For many control problems, computation of the optimal controller is intractable. Our objective in this
research is to develop simple algorithms for computing approximately optimal polices, and show that
the resulting cost achieved is close to the optimal achievable cost.
We present a a simple way to compute upper and lower bounds on the performance of stochastic con-
trol systems. We consider Markov decision processes over general state spaces, and our approach
allows any function to be used as an approximate Hamilton-Jacobi solution.
We give a number of examples including event-based sampling, dynamic planning for multiple ve-
hicles, decentralized decision problems and queuing. For each of these we construct a decentralized
policy and give a bound on the ratio of the cost achieved to the optimal achievable cost.
About the Speaker: