基于模仿学习的人机协作的研究

An Imitation-Learning-Based Human-Robot Collaboration

人机协作是机器人领域的一个重要课题,通过人机协作可合并人机优势,从而更好地完成现场任务。然而,人机协作系统的建立需要大量的工作,特别是手动编程的工作需要有十分专业的机器人技术背景,这不利于人机协作的应用和推广。提出利用模仿学习的方式来对人机协作任务进行建模,通过交互概率模型对任务中人和机器人运动的时空差异性进行表达,并辅以肌电信号来提高人的任务识别率。该模型将任务识别和机器人协作运动轨迹的生成统一起来,仅通过任务的示范即可对该交互概率模型进行训练,并在协作型机器人Baxter上进行了实验验证。

Human-robot collaboration（HRC）is an important topic in robotics.The tasks would be done better through HRC because of the combination of human and robot specialty.However,it takes tremendous effort to model an HRC task which requires the strongly professional programming skill.It goes against the application and promotion of HRC.In this paper,the model of HRC was set up through imitation learning method.This interaction probabilistic model was used to describe the spatiotemporal variance about the human and robot motion in the task,and the electromyography（EMG）signal was introduced to enhance the task recognition.This interaction probabilistic model combines the task recognition and the robot collaborative motion trajectory generation,so you just need to demonstrate the task to train model.The experiment on Baxter has verified the proposed method is feasible.