Integration of problem-solving and learning in intelligent robots

Abstract

In the real world, there are many kinds of uncertainties in the motions of robots. Moreover, robots cannot always get sufficient knowledge about tasks in advance. Intelligent robots therefore have to possess both problem-solving ability, to decide on the proper motions with insufficient knowledge, and learning ability, to acquire knowledge from experiences. Effective integration of these two kinds of ability is also important. In this paper, we describe an experimental system named ARPEX-L (Automatic Robot Planning and Execution System with Learning Ability). ARPEX-L consists of the integration of reactive planning and learning. We applied ARPEX-L to two kinds of robot task: a pick-and-place task and a pushing block task. The former is a simple example of automatic robot programming. The latter is an attempt to make an actual robot learn by a symbolic framework. These applications show the validity of our approach in constructing intelligent robot systems. Current defects of the system and future work are also described.