Summary: | 碩士 === 國立臺灣海洋大學 === 機械與機電工程學系 === 95 === In general, a robot requires being equipped with the vision-aided positioning capability in order to execute automatic peg-in-hole task. However, the vision-aided positioning capability will become inefficient or incapable when the working environment is too obscure or dark. In view of the above, the main objective of the thesis focuses on the problems of applying force feedback algorithm to correct positional error between the peg’s center and hole center.
We have adopted force/torque sensor for the agent of robot arm and the working environment. To achieve the purpose of automatic positioning correction, we utilized the nonlinear mapping characteristic of the neural network. The training data of the experiments comprised the contact positions of the peg on the edge of the hole and the reaction moments measured by the force sensor mounted on the wrist. And then the data was fed into the neural network, and the relationship was memorized using back propagation algorithm. Finally, when the robot was executing the insertion task, it would first acquire reaction moment signals from the sensor, and estimated the positional error through the previously trained neural network.
It has been shown by the experimental result that the percentage of tests with positioning error less than 5mm is up to 85% (totally 270 training data). It showed that the robot arm can execute the preliminary peg-in-hole insertion task using the proposed force-guided method.
|