Human Arm Trajectory Prediction Framework for Safe and Efficient Human-Robot Collaboration

Accurately prediction of human arm trajectories is necessary to the safe and efficient execution of human-robot collaborative tasks. Traditional algorithms often struggling with low accuracy and uncertainty estimation. To address these challenges, a framework tailored for human arm trajectory prediction in collaborative scenarios is proposed. Firstly, one collects three-dimensional spatial data of trajectory points on the human upper limb by utilizing depth camera and complementary devices, and constructing a dataset with five operational tasks. Preprocessing involves outlier removal and, smoothing to enhance data quality. To bolster early-stage position estimation accuracy, one integrates observed and predicted arm trajectories, and employing Gaussian Mixture Model for current hand motion target prediction. Subsequently, one develops an Informer prediction model for human arm trajectories. Compared with related works, the result on predicted and actual data curves demonstrating superior performance over four alternative approaches.