A control framework for learning and performing safe human-robot collaborative assembly

The advent of collaborative robots has enabled new applications where human and robots can work together safely and effectively toward achieving a common goal. However, robots in the industry, and particularly in assembly lines, are still programmed with time-consuming methods, have little to no adaptability to changes in the environment, and still present safety issues in close collaboration with a human, due to stiff control methods utilized to achieve accuracy. In this chapter, we address the problem of learning a pick-and-place task by kinesthetic demonstration and executing it to new and even moving targets, aiming at collaborative assembly lines where operations take place on a moving conveyor. Moreover, the human safety aspect is addressed via a control framework that includes two layers. The first layer is in the form of collision avoidance by altering the executed trajectory, and the second one is achieved by a compliant robot behavior that preserves safety under unintentional contacts, without compromising target-reaching accuracy. For the learning part, we modify the traditional dynamic movement primitive (DMP) formulation for properly encoding the demonstrated task and generalizing it spatially to even moving targets. For the performance and safety part, we propose a control structure incorporating the DMP, the human avoidance signals, and the low stiffness controller of the robot to achieve safety and accuracy. Our approach is validated experimentally in a real-world industrial assembly scenario of a TV manufacturer, where a robot and a human collaboratively achieve the assembly of electronic boards on a TV back plate. This work contributes on a wider scale, as it aims to advance the way toward the adoption of human-robot collaboration in industrial assembly lines.