Learning the insertion operation of a flexible beam into a hole with a manipulator

Abstract

Flexible objects are widely used in the aerospace, automobile, electronics, and medical industries, but automated assembly of flexible objects is difficult to realize. In most cases, flexible objects are still handled and assembled by people. This article researches a typical flexible objects assembly operation, i.e., to insert a flexible beam into a hole. A learning method is proposed to learn the mapping from the sensed force to the end-effector's motion, by which the insertion operation can be achieved efficiently. The mapping is decomposed in Cartesian space. Agents based on a learning automaton are defined between the input space, formed by the sensed force, and the output space, formed by the end-effector's motion, to implement the functions of the decomposed mappings. The input space is partitioned into different contexts. Through learning, agents can learn optimal actions according to different contexts, and then fulfill the insertion task cooperatively and efficiently. Simulation results of a 2D insertion operation prove the feasibility of the proposed method.