An open extended reality platform supporting dynamic robot paths for studying human–robot collaboration in manufacturing

Human–robot collaboration (HRC) in manufacturing allows advantageous distribution of tasks, e.g. exploiting robot accuracy and human dexterity, safety being of paramount importance. Safety is mostly linked to avoiding collisions between the human and the robot but the pertinent measures adopted should prolong task duration as little as possible. In order to test such measures in HRC pertinent algorithms need to be applied, which is made possible without jeopardising human safety only in an Extended Reality environment. In order to implement path planning algorithms and human–robot interaction rules freely the environment must be open. In this work, the development of such an environment is presented and demonstrated by example of laying up carbon fibre fabric sheets in a mould. An existing open platform was substantially extended by embedding robot control functionality concerning motion, path and trajectory planning emphasizing static and dynamic obstacle detection, interactive input and manipulation and real-time path planning, whereas trajectory planning focused on ensuring acceptability of joint motion solutions using inverse kinematics. Two different real-time path planning methods are embedded in the environment as representative examples. The first one is the established ‘Rapidly exploring Random Tree’ (RRT) algorithm followed by path optimization. The second one is ‘Machine-Learned Path Planning’ (MLPP) a prototype machine learning model trained using linear regression with Gaussian noise based on safe path planning data generated by users. The evaluation criteria of these methods were the number and severity of collisions as well as the total completion time of the manufacturing task. In the particular case examined, the machine learning technique proved much faster than RRT but not as safe, despite its potential. However, the openness of the XR platform enables testing of any other strategy supporting HRC in manufacturing before it is actually transcribed to the real robot controller.