Neural Network and Striped Lighting Pattern- Based Autograsping Technology for Flexible Robotic Assembly

摘要

A flexible robotic assembly cell in which fixtures are deemed unnecessary was developed at the Automation Laboratory of the Chung-Hua Polytechnic Institute. The crucial innovation is the development of a robotic autograsping capability, constructed in two steps. The first step performs an off-line neural network training procedure. Laser-striped light projected onto a part is graphically simulated by the SGI computer in advance. The geometric centers of striped lighting patterns with different part orientations are calculated and used as neural network training data. After training the neural network, the second step accomplishes the on-line testing procedure. When an unheld part arrives at the robotic cell, the optimal orientation adjustment for robot grasping is predicted by sending the geometric center of the real striped light pattern to the previously trained neural network. According to the results of experiments, the prediction error of trained neural networks are only 1° to 2? when 41 training patterns for each part have been trained and the possible range of each part's orientation is 40°. This proposed technology not only simplifies control strategies but also makes the autograsping capability more reliable and flexible. It is especially suitable for production involving small volumes and many varieties.