Human–Robot Shared Control Based on Locally Weighted Intent Prediction for a Teleoperated Hydraulic Manipulator System

Abstract

Human–robot shared control provides a promising opportunity to improve operational safety and efficiency for a teleoperated hydraulic manipulator system that works in unstructured environments, such as rescue response and underwater exploration. However, the poor dynamic of the heavy-duty manipulator (e.g., oscillation tendency and response delay) brings a challenging problem on intent prediction and effective assistance. To overcome this problem, a blended shared control method is proposed via locally weighted intent prediction. First, a task learning method from demonstration is presented by task-based action primitives with the Bayesian nonparametric clustering to extract subgoals of the remote arm trajectory, especially a locally weighted chattering cancellation method is proposed to reduce unintended action primitives due to velocity fluctuations. A dynamic intent prediction method is presented based on the empirical and real-time motion information by introducing an iterative transition matrix and a locally weighted dynamic angle, respectively. A prediction-based arbitration rule is finally established to blend the controllers of the human and robot seamlessly. Comparative tests in two typical scenes were carried out on a teleoperated hydraulic manipulator system. The results validate the proposed shared control method with fewer collision accidents and reduced task completion time, and subjective feelings on operation are also improved.