Deep learning-based smith predictor design for a remote grasping control system

In this study, a robotic hand control system was designed based on data gloves, aiming to provide more intuitive control and improved operational performance for a remote robotic hand. Compensation measures were proposed for the time lag effect on the remote-control system to address the input and feedback time delays of the remote robot system. A Smith predictor structure was modified by replacing the linear estimator with a recurrent neural network. A convolutional neural network was applied to the long short-term memory (LSTM) model, as it had a better convergence time and learning performance than the multi-layer perceptron model during training. The experimental results demonstrate that the control effect of this scheme is approximately 0.5 s faster than the normal Smith predictive control, proving its effectiveness.