Optimal Collision-Free Robot Trajectory Generation Based on Time Series Prediction of Human Motion

In this letter, we propose that the joint motion of a human worker doing repetitive work could be predicted using a time series model. With a motion capture system, the elbow joint rotation data are collected and used to fit an autoregressive model. An online parameter adaptation algorithm is employed to update model parameters in real time. A safety index with a distance factor is defined to describe the level of safety during physical human-robot interaction. An optimization problem is formulated to generate a collision-free trajectory for the manipulator based on human motion prediction to make the generated trajectory smoother. Planar and three-dimensional simulations are conducted to validate the efficacy of the algorithm by comparing the trajectories with and without real-Time human motion prediction. Experiments are conducted on a 6-DOF manipulator Universal Robots UR5 with a Microsoft Kinect sensor (ver. 1). The optimization problem can be solved within the sampling time of the system, which suggests this algorithm can be applied in real-Time trajectory generation. The trajectories generated with different control horizons are compared, and the results show that the manipulator can achieve the goal faster via a smoother trajectory with the help of human motion prediction.