Trajectory Error Prediction of Manipulator Using Seq2Seq Gated and Adaptive Continuous-Time Recurrent Neural Network (S2S-GACTRNN)

With the widespread application of manipulators, ensuring the trajectory tracking performance during their operation has drawn increasing attention. Due to the strong fitting capabilities, neural networks can be employed to predict and generalize the trajectory errors of manipulators. In this study, we introduce a sequence-to-sequence gated and adaptive continuous-time recurrent neural network (S2S-GACTRNN) for trajectory error prediction. This approach skips the mapping relationship between joint space and end-effector trajectory errors and is directly used for the generalization and prediction of periodic trajectories in the manipulator's workspace. Exper-imental results show that the network achieves a prediction accuracy exceeding 90% for identical end-effector speeds, with an average generalization accuracy of 84% across 7 different end-effector speeds. These results demonstrate that the network effectively learns the temporal and periodic characteristics of trajectory errors.