Variable admittance control based on iterative learning scheme with adaptive gradient methods for human-robot collaborative manipulation

Abstract

Human-robot cooperation systems can combine the skills of humans and the power of robots, which can improve productivity and flexibility and reduce the physical burden on workers. Admittance control has often been applied to the collaborative task with physical human-robot interactions. However, in the conventional methods, the admittance parameter was adjusted based on heuristic methods. The authors have proposed an iterative learning control scheme that can update admittance parameters to reduce the physical burden on the operator in the collaborative task. However, there was a problem that the learning performance was significantly influenced by uncertain data such as noise and outliers because the steepest descent method, which has a fixed learning rate, is employed in the updating law. Furthermore, the manual learning-rate adjustment by trial and error was required to improve learning performance. In recent years, research on adaptive gradient methods that vary the learning rate has been actively conducted in the fields of machine learning, aiming at improving learning performance. In this paper, we propose a novel iterative learning control scheme with adaptive gradient methods for human-robot collaborative manipulation to improve the learning performance against uncertain data and lower the cost of adjusting the learning rate. The validity of the proposed method is demonstrated through extensive experiments, including 1) cooperative operations in the presence of obstacles and 2) cooperative transport of heavy objects.