Multi-Joint Predictor for Human Upper Limbs in Human-Robot Collaboration Tasks

The safety of collaborators is the primary concern in human-robot collaboration tasks. Accurately perceiving the human body pose information is a crucial technology for achieving the safe execution of human-robot collaboration. This paper proposes a multi-joint predictor for the human upper limb, discusses the correlation between multiple joints and the predictor's performance under different motion speeds to provide precise obstacle avoidance information for collision detection. A combined dataset is created by collecting operation information from medical personnel in a cellular experiment context. Secondly, the sparrow search algorithm is utilized to optimize the Long Short Term Memory network for constructing a predictor for multiple joints while discussing the correlation between each joint to obtain real-time position updates. Finally, the usability and accuracy of the proposed predictor is validated.