LSTM network in bilateral teleoperation of a skid-steering robot

Abstract

The paper analyses a control scheme aided by LSTM networks for the delayed bilateral teleoperation system of a skid-steering wheeled mobile robot. The strategy implemented at the local and remote sites combines a virtual force based on nonlinear impedance, nonlinear Proportional–Integral (PI) gains, spring-damper, and robust neural dynamics compensation, including a gradient-based adjustment law or critic-actor RL trained offline using the ADAM algorithm. To analyse the stated strategy, stability analysis is performed. A Lyapunov–Krasovskii functional is proposed for evaluation along the system trajectories to analyse the evolution of control errors and network errors. Human-in-the-loop simulations are conducted and evaluated as a case study to observe the responses of velocities and yaw rate errors, lateral velocity, and network parameters in the presence of time-varying delays, variable load, and different terrain frictions.