Efficient Continuous Space BeliefMDP Solutions for Navigation and Active Sensing

Autonomous robot teams have the potential to revolutionize the way we approach many problems, ranging from transportation to active sensing for weather science. However, to accomplish these missions, the robots must operate in environments with more threats and uncertainty than current autonomous systems can handle. The Belief Markov Decision Process framework (BeliefMDP) is a systematic and robust mathematical framework that can be used to obtain policies for these agents while reasoning over different kinds of uncertainties in the environment. Since computing optimal policies for a BeliefMDP exactly is intractable, this doctoral proposal focuses on solving them approximately by leveraging tree search techniques and guiding them using smart heuristics and learning algorithms for long-horizon continuous space problems.