Collision Avoidance in Collaborative Robotics Based on Real-Time Skeleton Tracking

Human-robot interaction in manufacturing is increasing more and more due to the spread of collaborative robots. However, it is important to ensure operator safety and improve the operator's physical and mental well-being. In this paper, an obstacle avoidance algorithm is implemented exploiting a vision system for human motion tracking. Once the positions of the human body joints relative to the robot's base are known in real time, the manipulator is able to avoid collisions by changing the trajectory and ensuring task execution at the same time. The motion tracking system is realized with three RGB-D cameras and customized software based on artificial intelligence and machine learning techniques. The whole system was tested in a real human-robot interaction scenario with the Universal Robot UR5e manipulator. The robot is able to avoid collisions over its entire kinematic chain. The distance between the human body and the manipulator always remains above a safety distance that is a settable parameter of the algorithm.