[4D2-OS-18c] [Organized Session] OS-18

[4D2-OS-18c-02] Finding compromise solutions for multi-stage and multi-objective optimization problem

〇Keisuke Yamamoto¹, Sachiyo Arai¹ (1. Chiba University)

Keywords:multi-objective reinforcement learning, waste collection

Many problems in real world could be formalized as the multi-stage and multi-objective optimization problems (MOP) where there exist mutual conflicts among the objective functions. It is said that people hard to find a compromise solution without a sufficient number of solutions as the candidates. Thus, we propose a method to find the pareto-optimal exhaustively. Our approach is based on multi-objective reinforcement learning (MORL) because the real-world problem requires multiple action-sequences until getting the reward. We evaluate our proposed method by applying it to "waste collection problem" where there are two conflicting objective functions, "capacity of collecting vehicles" and "time".