Assembly Pattern Recognition for Modular Robot based on Neural Networks

Abstract

Unlike other general robot systems which have limited purpose, the modular robot can be used for many purpose by reconfiguring its structure appropriating to various work environments. Modular robots can be categorized into two groups, based on the purpose: the configuration-based and the operation-based. This paper proposes a modular robot system which can identify and memorize its load directions for performing motion depending on its assembly pattern. This robot system is designed to measure the angle of the joint in each of it, 1-DOF modules. After developing the hardware, memorizing kinetic information method and recognizing load direction method are implemented. The modular robot system can recognize the load direction of its configuration based on a neural network pattern recognition with the information on the measured trajectories that is repeated swaying motions of each module in sequence. Users can teach specific motion to each module by demonstrating. Then, the modular robot can perform the learned motion. Performance evaluation was conducted with four different load directions. As a result, the modular robot is able to correctly recognize the load direction in 90% of trials.