Development of a genetic algorithm based nonlinear model predictive scheme

Model predictive controller (MPC) has demonstrated its competency in controlling autonomous vehicles. But to apply the current MPC-based schemes, certain compromise or approximations have to be made in order to fit the optimization solvers, e.g. linearizing the nonlinear vehicle model. To eliminate the gaps, in this paper, we propose a nonlinear MPC controller which controls the vehicle velocity and steering simultaneously. The optimization solver is based on genetic algorithms (GA), which provides a flexible structure to design the cost function and constraints in a more accurate, meaningful and direct way. The simulation results showed that the vehicle under the control of the proposed nonlinear MPC is able to follow the road center line accurately and consistently, even at sharp corners. The average distance deviation from the road center is 18.4cm and moving direction deviation from the road tangent is 0.041rad. Moreover, the simulation results also showed that passengers´ safety and comfort can be well taken care of under the proposed MPC scheme as both the vehicle movement acceleration and steering acceleration are well confined within a safety range. The promising results from the simulation indicate that the proposed GA based nonlinear MPC can be a suitable solution to autonomous vehicle control.