Context Attention: Human Motion Prediction Using Context Information and Deep Learning Attention Models

This work proposes a human motion prediction model for handover operations. The model uses a multi-headed attention architecture to process the human skeleton data together with contextual data from the operation. This contextual data consists on the position of the robot's End Effector (REE). The model input is a sequence of 5 s skeleton position and it outputs the predicted 2.5 future seconds position. We provide results of the human upper body and the human right hand or Human End Effector (HEE). The attention deep learning based model has been trained and evaluated with a dataset created using human volunteers and an anthropomorphic robot, simulating handover operations where the robot is the giver and the human the receiver. For each operation, the human skeleton is obtained using OpenPose with an Intel RealSense D435i camera set inside the robot's head. The results show a great improvement of the human's right hand prediction and 3D body compared with other methods.