Implementing a Model Predictive Control for a self-driving car

The MPC considers the task of following a trajectory as an optimization problem in which the solution is the path the car should take. The idea is to simulate different actuator inputs (steering, acceleration and braking) and predict a resulting trajectory by selecting the one with the minimum cost. The car follows that trajectory and gets new input to calculate a new set of trajectories to optimize. The model utilizes the called “receding horizon controller” which performs a trajectory recalculation for every new state, since the defined trajectory is just an approximation.

Website