Datasets for Data-Driven Reinforcement Learning

This course focuses on offline reinforcement learning (RL) and provides an extensive benchmark for evaluating offline RL algorithms in various settings. The learning outcomes include understanding the offline RL problem, learning how to train a policy from a dataset of previously collected data, and evaluating algorithms in offline RL scenarios. The course teaches skills such as designing benchmark tasks and datasets for offline RL evaluation, analyzing properties of datasets relevant to offline RL applications, and driving improvements in offline RL methods. The teaching method involves presenting benchmark tasks and datasets tailored to offline RL, with examples including datasets from hand-designed controllers and human demonstrators. The course is intended for individuals interested in advancing their knowledge of offline RL, particularly researchers and practitioners in the field of reinforcement learning.