Development and testing of a collision avoidance algorithm for industrial applications

Safe interaction of the operator and the cobot in the co-working space is one fundamental requirement for introducing automation in small and medium enterprises characterized by unstructured cell layouts. In addition to the exploitation of cobots, a fully safe interaction requires the adoption of collision avoidance systems, to enable the real-time re-planning of the end-effector trajectories thus avoiding to stop the machine in case of collision with humans. This is fundamental for industrial applications, in order to maintain the production rate as constant as possible even in unstructured production environments. However, collision avoidance applications in actual industrial production cells are still limited, due to some limitations which are not yet completely solved. It is indeed still likely to happen that the recalculation of the obstacle-free trajectory takes a too long time, not compatible with industrial applications. This paper describes the development of a collision avoidance program using the Lazy PRM* algorithm. The program is tested on a physical robot, the Mitsubishi Melfa RV-5AS-D, to perform some exemplary pick and place tasks. The developed algorithm can alter online, during the execution, the road-map from dense to sparse in those cases in which the search for an alternative trajectory is detected to take too long time. Compared to other commercial collision avoidance systems, in which the evaluation of road-maps can only be done offline once and for all, this feature would prevent the robot to get stucked in case a feasible solution is not found in a fast enough way. Furthermore, the developed program guarantees greater flexibility in creating the road-map, trying to match the user's needs, and eliminating the superfluous parts of the road-map to reduce computational time. This marks a difference with present-day solutions for commercial collision avoidance systems which, to the author knowledge, connects all the nodes together.