Multi-Objective Deep Reinforcement Learning with Priority-based Socially Aware Mobile Robot Navigation Frameworks

摘要

Socially aware robot navigation is a multi-objective decision-making problem. Nonetheless, the attempt to address this issue using single-objective reinforcement learning raises substantial challenges when it comes to extending navigation policies to accommodate the intricate demands of human social interactions. Our work presents a versatile framework for expanding existing robot navigation capabilities into the realm of multi-objective decision-making. This framework empowers users to easily specify their preferences without the burden of extensive engineering efforts. Experimental findings clearly demonstrate the effectiveness of our framework in guiding an existing reinforcement learning system toward the optimization of navigation policies in alignment with user-defined preferences.