Inverse Dynamics Model Learning for Robotic Manipulator Based on Swarm Optimization AdaBoost-LSTM

Shichao Ai, Yunhan Lin, Wenhui Huang, Huasong Min

Abstract: Inverse dynamics models of robotic manipulator can be derived using neural network training. However, in previous studies of inverse dynamics neural network model learning, there is the overfitting problem of good training effect but average estimation performance, while the configuration of hyperparameters in the model always depends on the traditional experience, which has a large room for improvement of the estimation accuracy of the model. Therefore, in this paper, we propose a method to optimize the LSTM (Long Short-Term Memory) model using AdaBoost (Adaptive Boosting) as well as Swarm Optimization Algorithm, which utilizes AdaBoost's ability to minimize the overfitting and Swarm Optimization Algorithm to optimize the model hyperparameters. In the experimental part, the method proposed in this paper is compared with previous methods using publicly available inverse dynamics datasets. Experimental results show that the method proposed in this paper has higher torque estimation accuracy and improves the performance of the inverse dynamics model.