Learning discrete Bayesian models for autonomous agent navigation

Partially observable Markov decision processes (POMDPs) are a convenient representation for reasoning and planning in mobile robot applications. We investigate two algorithms for learning POMDPs from series of observation/action pairs by comparing their performance in fourteen synthetic worlds in conjunction with four planning algorithms. Experimental results suggest that the traditional Baum-Welch algorithm learns better the structure of worlds speci cally designed to impede the agent, while a bestrst model merging algorithm originally due to Stolcke and Omohundro [?, SO]erforms better in more benign worlds, including such that model typical real-world robot fetching tasks.