A PINN based control for stochastic nonlinear systems with input-deadzone

This paper investigates a novel controller for stochastic nonlinear system with input deadzone by leveraging Physics Informed Neural Network (PINN). The unknown system nonlinearities are approximated by using PINN through backstepping. A quartic Lyapnov function is utilised to construct virtual and actual controller, ensuring boundedness in probability of all closed loop signals. Deadzone nonlinearity is reformed into a linear input form to complete the backstepping process. Simulation result on a robotic manipulator system demonstrate excellent tracking performance and improved accuracy compared to existing method, confirming the effectiveness of proposed PINN based control strategy.