As robots become increasingly advanced, the tasks we expect them to complete have become increasingly complicated. By increasing their available actions and what they can observe, we are creating an increasingly difficult world in which they must find and execute a plan. In this project, we are planning in the belief space of probability distribution over states. We use a partially observable Markov decision process (POMDP) to model the current belief state of the robot. This project will complete the implementation of a symbolic planner that attempts to solve these increasingly complex POMDPs. This planner will determinize and replan in the belief space, and will have a structured representation of its belief state. Both of these characteristics should yield a fast and efficient planner.

Im particularly interested in the areas of machine learning and planning algorithms. Ill be working with Professor Kaelbling to implement and evaluate a symbolic planner that attempts to solve increasingly complex POMDPs. My interest in planning algorithms began last fall when I took 6.S078, Planning Algorithms. I’m excited to apply what Ive learned about planning algorithms in a research setting.