Deep Learning Based Human-Robot Co-Manipulation for a Mobile Manipulator

In this paper, a deep learning based human-robot co-manipulation technique using a mobile manipulator is proposed, which employs a CNN-LSTM network to generate robot motions through the short-term human motions. To learn a human motion, the CNN-LSTM network is trained on a library of motions, which is produced from human-robot demonstrations based on the positions of the human arm and the robot joint actions. To make the robot precisely following the generated motions under uncertain dynamics, an adaptive neural network controller is designed to ensure that the robot outputs track the generated trajectory within a small neighborhood of zero. The effectiveness of the proposed approach is demonstrated by performing co-manipulation tasks with a mobile manipulator.