Collison Avoidance using Point Cloud Data Fusion from Multiple Depth Sensors: A Practical Approach

This paper presents a collision avoidance system based on vision sensors that is suitable for collaborative robotics scenarios. In fact, collaborative robotics foresees the possibility that humans and robots share the same workspace, so the safety of the human operators must be ensured. The collision avoidance algorithm here presented can modify the trajectory of the robot in order to avoid any collisions with a human operator. A fundamental element for the algorithm is the relative position between robot and human. In this work, the information of the position of a human operator is obtained by Microsoft Kinect sensor in the form of a point cloud. Two Microsoft Kinect are used and their point cloud data is merged to overcome the problems related to the possible occlusions of the sensors, obtaining a more reliable point cloud. Each Kinect works with a dedicated PC and the two PCs communicate via ethernet network in a master-slave mode. The layout of the acquiring system is described and the functions used for the communication between the PCs and the manipulation of the point clouds are presented. Results of simulation tests made to verify the performances of the system and collision avoidance based on point cloud compared with convex mesh are reported and discussed.