Visual Feedback To Predict Obstacle Motion For On-line Collision-free Trajectory Planning Of Cylindrical Robots

Abstract

ABSmACT When a robot moves in a partially known environment, the desired trajectory generated in advance (off-line) may lead to collisions with nonstationary obstacles. Therefore, the on-line detection of obstacles in the workspace from sensor information, testing for potential collisions of the robot with obstacles, and possible on-line modifications of the planned trajectory are necessary. This paper describes a method that uses computer vision to detect a moving obstacle in the three-dimensional robot workspace. The movements of the obstacle of unknown dynamics are predicted by means of a recursive autoregressive (AR) time series model, in which the parameters are estimated by the least mean squared error method. This algorithm is combined with an on-line robot trajectory planning algorithm to generate a collision-free trajectory for the cylindrical robot. The approach is demonstrated by laboratory