Abstract
The transfer of information between a human and a robot is of vital importance for robotic telemanipulation systems.In this paper,we propose a novel strategy-assisted shared control(SASC)scheme driven by electromyography(EMG)signals for robot telemanipulation.First,we develop an EMG decoding scheme to achieve reliable online performance,where ten features are extracted and selected and then classified by machine learning algorithms.Several feature selection methods are compared to learn a compact representation from original feature sets.Then,a vision-based module is designed for object detection and localization,which helps to grasp or release objects autonomously.Moreover,online visual feedback of environmental states and audio feedback of EMG decoding results are used for better perception of environmental contexts.A platform integrated with a robot arm,an RGB-D camera,and a pneumatic sucker,among others,is developed to evaluate the proposed method.The best accuracy of EMG recognition is 97.18％±10.61％,achieved by the support vector machine and features selected by recursive feature elimination.Two paradigms driven by a joystick or EMG are compared with SASC in a robotic telemanipulation task.The performance shows that SASC reaches a successful rate of 100％with the least time of 61.97±9.89 s.The completion times of SASC are only 67.78％and 55.71％of the two contrast methods,respectively.This shows that the proposed scheme has the potential to replace classical methods with reliable performance and less time,which provides a novel solution for this field of study.
万方数据