Neural Network Inverse Optimal Control of Ground Vehicles

Abstract In this paper an active controller for ground vehicles stability is presented. The objective of to force the vehicle track a desired reference, ensuring safe driving conditions in case adhesion loss during hazardous maneuvers. To aim, nonlinear discrete-time inverse optimal control based on neural network identification designed, using recurrent high order (RHONN) trained by Extended Kalman Filter. RHONN ensures error, while tracking errors. Moreover, reduced state observer utilized reconstruct lateral dynamic not usually available. For problem, references velocity and yaw rate are given system mimicking ideal having not-decreasing tire characteristics. proposed approach avoids Pacejka’s parameters tires, so simplifying input determination. optimize actuator effort power, bounded. Control gains determined “nature-inspired" algorithms such as particle swarm optimization. Test maneuvers, performed through full simulator CarSim ® , have been used test correctness, quality performances observer, identifier controller. Robustness also discussed different sample times. Finally, fair comparison between non-optimal schemes presented, highlighting numerical results obtained simulation.