Semi-Adaptable Human Hand Motion Prediction Based on Neural Networks and Kalman Filter

Abstract This paper focuses on predicting trajectories of the human hand in order to improve the safety for human-robot interactions. In this work, the position and orientation are represented by two curves in the operation space such that the same algorithm can be used for both position and orientation prediction. The motion prediction is achieved in two steps. Firstly, the neural network (NN) model is applied for offline training to model the human hand motion. Secondly, the Kalman filter is added to adjust the weight coefficients of the NN model’s output layer online when a set of new data is measured, such that the NN model is adaptive to new data. An experiment study has been conducted to validate the effectiveness of the proposed algorithm. The result shows that the proposed algorithm achieves a higher prediction accuracy and requires a smaller amount of data to achieve optimal performance compared with the advanced method.