Speaker: Jon Oelfke
Candidate for Bachelor of Science in Computer Science
Time: 5:05 PM
Place: Trustee Hall 211
Supervisor: Dr. Anthony Cassandra
Title: Searching for Optimal Policies for Partially Observable Markov Decision Processes

Abstract: Decision-making under uncertainty is a difficult task that is an integral part of acting in the real world. Partially observable Markov decision processes (POMDP) are one way to model environments that contain uncertainty where a sequence of decisions must be made. This presentation describes experiments that were carried out to explore the effectiveness of various search heuristics in finding good decision policies for POMDP. The algorithms that were developed and the problem structures used in the experiments are characterized, and the results of the experiments are given.