Spatial Repetitive Learning Control for Trajectory Learning in Human-Robot Collaboration

Waypoints updating based on Adam and ILC for path learning in physical human-robot interaction

“…In an ideal case, if we can achieve F h = 0, it indicates that the robot's actual position X = X h . Different from the previous work [14], we will not update the robot's reference trajectory in the time domain but will introduce a waypoints updating method to generate the robot's reference path. As shown in Fig.…”

“…In this section, simulation results are presented to demonstrate the advantages of the proposed waypoints updating method by comparing with the existing ones: Adam optimization in [17] and ILC in [14]. The aforementioned HRI scenario is considered, where a human user guides a robotic manipulator to complete a path following task.…”

“…4(a) and Fig. 4(b) show the path updating in seven iterations by the Adam optimization method [17] and spatial ILC method [14], respectively. It is found that the Adam optimization method needs five iterations and the spatial ILC method needs seven iterations for following the desired path.…”

“…It is further developed in [13] for nonlinear dynamic systems with random changes in the iteration period. In our previous work [14], we propose a method based on spatial iterative learning to learn the desired path of the robot's human partner, which is assumed to be periodic in space instead of in time. In [15], we develop a period-varying ILC scheme for HRI where the human movement is assumed to be periodic but with uncertain time durations.…”

“…2) The proposed method does not require the assumption of repetitive task with a certain period, and addresses the limitation of the ILC method in [14]. 3) By combining the improved Adam path optimization algorithm with ILC, our method achieves faster learning and reduced human effort, compared to the algorithm in [17].…”