基于模仿学习的变刚度人机协作搬运控制

针对人-机器人协作搬运，现有的控制策略难以同时保证搬运过程的柔顺性和搬运终点位置的精确性，而且对不同搬运任务适应性不够. 基于模仿学习提出变刚度协作搬运控制策略. 使用任务参数化的高斯混合模型(TP-GMM)对多次搬运示教数据进行编码，学习不同搬运工况下的搬运轨迹概率模型；结合导纳控制建立机械臂末端变刚度交互模型，实现柔性搬运操作，并基于交互力阈值实现不同搬运任务的切换；搭建协作搬运平台进行实验验证. 实验结果表明，提出的策略在实现柔性协作搬运的同时将特定搬运任务的终点位置精度提高到1.9 mm，且保证了特定搬运任务中机械臂末端在期望区域内运动以及搬运任务的切换.

Variable stiffiness control for human-robot cooperative transportation based on imitation learning

Aiming at human-robot cooperative transportation, current control strategies are difficult to guarantee both the compliant control for cooperative transportation and the accuracy of the end point, and lack sufficient flexibility to different tasks. A variable stiffiness control strategy was proposed for cooperative transportation based on imitation learning. Firstly, several human demonstrations of cooperative transportation were encoded in task parameterized Gaussian mixture model (TP-GMM) and a probabilistic model of trajectories under different transport conditions was learned. Secondly, combined with admittance control, an interactive model of variable stiffness at the end of the manipulator was established for transportation to realize compliant control. Besides, a strategy of switching between different transportation tasks was developed with a threshold method of interaction force. Finally, a cooperative transportation platform was built for experiments. Results show that the proposed approach can improve the position accuracy of the end point to 1.9 mm in specific tasks while achieving compliant control for cooperative transportation, and ensure that the end effector of the robot in specific tasks is controlled to moved in desired space and the switching of transportation tasks is available.