10206 移動ロボットによるサブタスクの自律的定義と制御則獲得の一手法(OS15 知能ロボット)

概要

論文の詳細を見る
From recent studies in machine learning, making a proper use of multiple simple controllers may show much better performance than using just one big controller. Thus, it is important for agents to obtain standards for separating assigned tasks into sub-tasks, which yet fully discussed. In this paper, we propose a method which enables to criticize the effectiveness of goals for each sub-task by agents themselves. Here, agents innovate criteria, a combined information available to agents through learning. Agents then decide a policy, which declares that one criterion has the highest influence over the result of sub-task learning. Through simulation and actual learning, agents attempt to match policies to candidates of sub-task goals. We have attempted experiment on sweeping task and verified how the proposed algorithm works comparing to conventional approaches.
2005-03-17