Manipulator Smooth Control Method Based on LSTM-XGboost and Its Optimization Model Construction

Abstract

With the rapid development of computer and artificial intelligence technology, robots have been widely used in assembly, sorting, and other work scenarios, gradually changing the human-oriented mechanical assembly line working mode. Traditional robot control methods often rely on application fields and mathematical models, and they cannot meet the emerging requirements of versatility and flexibility in many fields, such as intelligent manufacturing and customized production. Therefore, aiming at the relationship between the manipulator’s smooth control command parameters and the manipulator’s actual motion stability in the multi-step object sorting task, this paper proposes a method for predicting the stability of the manipulator based on Long Short-Term Memory Extreme Gradient Boosting. The acquisition signal of the manipulator vibration is segmented according to the action, and the boost model is used to learn the relationship between the control command parameters and the stability characteristic indexes. Next, the Extreme Gradient Boosting algorithm establishes a feature index-stationarity score prediction model. The minimum Mean Absolute Error predicted by the five indicators is 0.0024 so that the model can predict the manipulator’s motion stability level according to the manipulator’s command parameters.