認識と計画の実時間統合に基づくロボットの作業レベルフィードバック制御

Abstract

This paper discusses a methodology to allow a robot to execute the task autonomously with high confidence and its method of the systematization. Taking up the mobility of a robot without collision in the dynamic environment as an example, a concept of the feedback control for the execution of the task is introduced. The key idea is to incorporate the real-time recognition and motion planning functions into the task control loop. For real-time recognition of the environment, mark-based vision is proposed to be systematized. In additoin, algorithm to do a global motion planning in the environment where movable obstacles exist is proposed to be implemented. By integrating the real-time recognition and motion planning, we succeed in forming a control loop for the task. It is verified by experiments that mobility of a robot controled by the proposed method is done without failure except that a human handler intentionally disturbed the movement of the robot.