The Explainable Model to Multi-Objective Reinforcement Learning Toward an Autonomous Smart System
Tomohiro Yamaguchi · Advances in systems analysis, software engineering, and high performance computing book series · 2023
The mission of this chapter is to add an explainable model to multi-goal reinforcement learning toward an autonomous smart system to design both complex behaviors and complex decision making friendly for a human user. At the front of the introduction section, and a relation between reinforcement learning including an explainable model and a smart system is described. To realize the explainable model, this chapter formalizes the exploration of various behaviors toward sub-goal states efficiently and in a systematic way in order to collect complex behaviors from a start state towards the main goal state. However, it incurs significant learning costs in previous learning methods, such as behavior cloning. Therefore, this chapter proposes a novel multi-goal reinforcement learning method based on the iterative loop-action selection strategy. As a result, the complex behavior sequence is learned with a given sub-goal sequence as a sequence of macro actions. This chapter reports the preliminary work carried out under the OpenAIGym learning environment with the CartPoleSwingUp task.