Catching Flying Objects Based on Trajectory Prediction Method with Long Short-Term Memory Neural Networks

With the rapid development of robotics technology, catching flying objects has become a crucial research focus in the field. The accurate prediction of trajectories for flying objects is pivotal for successful capture. Robots need to accurately and swiftly predict the trajectory based on the observed points to move to the landing point and complete the catch process. Traditional prediction methods include polynomial fitting and predicting object motion equations based on force conditions. However, due to the complexity of flying object motion, traditional methods often exhibit significant accuracy errors and poor generalization to different objects. In response to these challenges, we propose a prediction method based on Long Short-Term Memory (LSTM) neural networks. Experimental results show the superior accuracy of the LSTM prediction algorithm. We conducted tests on a robot using this prediction algorithm, and achieve catch accuracy of 89.7%, surpassing the traditional method’s accuracy of 83.4%.