Multi-Algorithm Fusion Framework for Trajectory Prediction of Human Upper Limb Motion

To achieve the accuracy and security controlling of human-robot collaborative in the medical laboratory, a multi-algorithm fusion framework for human upper limb trajectory prediction is proposed. The main idea can be described as follows. A fusion strategy for integrating three algorithms including partial time series classification, representative trajectory, and polynomial fitting that can recognize human behavior and provide trajectory prediction is given. It can improve the accuracy of human-robot cooperation. The framework is divided into the online phase and the offline phase. The offline phase is designed to learn the parameters required by the predictive model from the task data directly, and the fusion method combines different algorithms; In the online phase, based on partial trajectory classification, polynomial fitting algorithms and representative trajectory algorithms have been used to predict the trajectory within a given time node. We discuss its accuracy in simulations. Furthermore, the framework was evaluated on a well-designed human upper limb dataset that collects the videos of humans-robot collaboration in a real scenario. The results show that our method could predict human trajectories better than the state-of-the-art forecasting algorithms.