Automated Planning and Reinforcement Learning

The Automated Planning and Reinforcement Learning area studies how to build AI programs that plan and execute sequences of actions to achieve goals in deterministic, stochastic, and adversarial environments. Some approaches are “model-based” in that they are fed models or derive models of the environment and use them to achieve goals and rewards, while others are “model-free” in that they start with a random policy and gradually adapt it to maximize the expected environmental rewards. The challenges in this area include coping with incorrect models, efficient exploration and information gathering, and learning to coordinate when multiple agents are involved.

The research is applicable to a variety of domains, including logistics, scheduling, autonomous navigation, dialogue management, and game playing. It also plays a key role in applications such as drug synthesis, efficient manufacturing, the design of novel materials, plant disease management, and the improvement of farm operations in agriculture.