Dynamic Movement Primitives-Based Human Action Prediction and Shared Control for Bilateral Robot Teleoperation

This article presents a novel shared-control teleoperation framework that integrates imitation learning and bilateral control to achieve system stability based on a new dynamic movement primitives (DMPs) observer. First, a DMPs-based observer is first created to capture human operational skills through offline human demonstrations. The learning results are then used to predict human action intention in teleoperation. Compared with other observers, the DMPs-based observer incorporates human operational features and can predict long-term actions with minor errors. A high-gain observer is established to monitor the robot’s status in real time on the leader side. Subsequently, two controllers on both the follower and leader sides are constructed based on the outputs of the observers. The follower controller shares control authorities to address accidents in real-time and correct prediction errors of the observation using delayed leader commands. The leader controller minimizes position-tracking errors through force feedback. The convergence of the predictions of the DMPs-based observer under the time delays and teleoperation system stability are proved by building two Lyapunov functions. Finally, two groups of comparative experiments are conducted to verify the advantages over other methods and the effectiveness of the proposed framework in motion prediction with time delays and obstacle avoidance.