Neural Networks based Human Intent Prediction for Collaborative Robotics Applications

Abstract

Industry 5.0 has laid the necessity to relocate the human at the center of the manufacturing cycle, where everything should be designed to ease his/her work. This implies one to redefine the human-robot collaboration, making it not only safe, but also more inclusive for the operator. Nowadays, this goal is viable thanks to the integration of new sensors in the work cell. Coupled with advanced control strategies, they give the robot a better understanding of both the surrounding environment and the human movement, allowing a more organic cooperation. This paper exploits an RGBD camera and Deep Learning algorithms to retrieve the 3D positions of the manipulated objects in the workspace and to infer the most likely future human destinations in a pick and place case study. Merging this information, a proper control logic is defined and tested in a real robotics application, with the final intent of minimizing human-robot collisions during the collaboration and making the process more reliable and efficient.