A temporal difference method for multi-objective reinforcement learning

Abstract

This work describes MPQ-learning, an temporal-difference method that approximates the set of all non-dominated policies in multi-objective Markov decision problems, where rewards are vectors and each component stands for an objective to maximize. Unlike other approximations to Multi-objective Reinforcement Learning, MPQ-learning does not require additional parameters or preference information, and can be applied to non-convex Pareto frontiers. We also present the results of the application of MPQ-learning to some benchmark problems and compare it to a linearization procedure.