Obstacle Detection in the surrounding Environment of manipulators based on Point Cloud data

Abstract

In the field of production and life, robots can avoid obstacles autonomously at work, not only to ensure safety, but also to improve their space utilization and work efficiency. The robot can detect obstacles in real time as the basis for collision avoidance. This paper studies the self-identification of the manipulator and the obstacle detection method in the point cloud environment. First, the calibration of the depth camera and the manipulator was performed using the least squares method. Secondly, aiming at the environmental requirements of manipulator collision avoidance and human-machine cooperation, a Locally Convex Connected Patches (LCCP) segmentation method based on kinematics model is proposed to realize the self-identification of the robot. Finally, by setting the control points of the manipulator model, the closest point and the closest distances to the obstacle are obtained. The effectiveness of the method was verified on the Kinect and UR robot.