Deep Learning Technology-Based Exoskeleton Robot Controller Development

For higher degrees of freedom robot, mass matrix, Coriolis and centrifugal force and gravity matrix are computationally heavy and require a long time to execute. Due to the sequential structure of the programs, multicore processors cannot boost performance. High processing power is required to maintain a higher sampling rate. Neural network-based control is a great approach for developing a parallel equivalent model of a sequential model. In this paper, Deep learning algorithm-based controller is designed for 7 degrees of freedom exoskeleton robot. A total of 49 densely connected neurons are arranged in four layers to estimate joint torque requirements for tracking trajectories. For training, the deep neural network analytical model-based data generation technique is presented. A PD controller is added to handle prediction errors. Since a deep learning network has a parallel structure, using a multicore CPU/GPU can significantly improve controller performance. Simulation results show very high trajectory tracking accuracies.