Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : May 10, 2017 - May 13, 2017
Traditional motion planners always generate trajectories from scratch, which is computationally expensive and fails to use previous knowledge of already encountered environments. Therefore, we propose to use a robot motion dataset to decrease the cost. We made an algorithm that learned the dataset distribution and approximate it with a Gaussian Mixture Model method to generate initial trajectories. The proposed method is evaluated on the STOMP algorithm in the simulation of a seven degree of freedom industrial robot.
