Multi-Objective Reinforcement Learning

Roxana Rădulescu

Reinforcement learning (RL) is a rapidly growing field in machine learning concerned with agents learning how to act optimally in sequential decision-making settings. Such settings translate to numerous real-world problems, where one has to take a sequence of actions in order to reach a desired goal. Real-world problems are generally complex and require trade-offs between multiple, often conflicting, objectives. Despite this, the majority of research in reinforcement learning either assumes only a single objective, or that multiple objectives can be adequately handled via a simple linear combination. Such approaches may oversimplify the underlying problem and hence produce sub-optimal results. Multi-objective reinforcement learning (MORL) is the more general setting that explicitly allows for multiple objectives to be taken into account during the learning process. The goal of this course is to provide an in-depth overview of multi-objective reinforcement learning and a guide to the application of MORL methods. It identifies the factors that may influence the nature of the desired solution, and illustrates how these should be used to guide the design of multi-objective decision-making systems for complex problems.