Partial Planning Reinforcement Learning

Abstract : This project explored several problems in the areas of reinforcement learning, probabilistic planning, and transfer learning. In particular, it studied Bayesian Optimization for model-based and model-free reinforcement learning, transfer in the context of model-free reinforcement learning based on hierarchical Bayesian framework, probabilistic planning based on monte-carlo tree search, and new algorithms for learning task hierarchies. The algorithms were empirically evaluated in real-time strategy games and other standard benchmark tasks and were shown to perform better than the state of the art approaches. The project also developed new theoretical frameworks for learning deterministic action models and for decision theoretic assistance and proved new formal results in these areas. The project helped graduate two Ph.D. students and partially funded the research of two other students.