Recurrent neural network for motion trajectory prediction in human-robot collaborative assembly

Abstract

Effective and safe human-robot collaboration in assembly requires accurate prediction of human motion trajectory, given a sequence of past observations such that a robot can proactively provide assistance to improve operation efficiency while avoiding collision. This paper presents a deep learning-based method to parse visual observations of human actions in an assembly setting, and forecast the human operator's future motion trajectory for online robot action planning and execution. The method is built upon a recurrent neural network (RNN) that can learn the time-dependent mechanisms underlying the human motions. The effectiveness of the developed method is demonstrated for an engine assembly.