Objective Weight Interval Estimation Using Adversarial Inverse Reinforcement Learning

Several real-world problems are modeled as multi-objective sequential decision-making with multiple competing objectives, and reinforcement learning (MORL) has garnered attention a solution to this problem. One of the challenges in obtaining desired policy using MORL is that priorities (hereafter, weights) for each objective must be designed advance scalarize reward vector. Determining weights through trial-and-error burdens system designers, methods estimate needed. The existing use inverse (IRL), which not scalable because it requires several times until an optimal obtained. This study proposes weight interval estimation (WInter) method adversarial IRL (AIRL). AIRL framework reduces computational complexity by simultaneously estimating rewards policies. WInter estimates expert neighborhoods obtained during training. We successfully estimated experiments benchmark environment continuous state space while reducing compared methods.