Effective action selection under multi task by ignoring tasks and limiting tasks

Abstract

In recent years, research has been conducted to treat multitask. In this field, task priorities change optimal solution and varies according to the situation. There is the problem that when priority is low, It can't do action selection well. We solve these problem by restrict tasks. We made a proposal that two types of restriction. Thereby, we assumed to can select action according to the situation. We carried out an experiment that set three tasks to a robot applied proposal technique. From results of the experiment, we showed the proposed technique improved learning performance and confirmed the usefulness of proposal technique.