Path tracking for car-like robots based on feed-forward nonlinear model predictive control

Due to the low degree of parts standardization of car-like robots, tire mechanical parameters such as cornering stiffness are difficult to obtain accurately, and dynamic modeling is very difficult. Therefore, existing research work usually uses kinematic models as the control model of car-like robots. However, due to model mismatch in its kinematic model, it will lead to severe oscillations in the error between the car-like robot and the reference path, the front wheel angle and the front wheel angular speed. In response to the aforementioned problems, this paper is based on the rolling optimization principle of nonlinear model predictive control (NMPC), introduces feedforward angle information based on the inverse kinematics model, uses the front wheel angle as the fourth dimension of the prediction model, and proposes A car-like robot path tracking control algorithm based on feedforward nonlinear model predictive control (Feedforward NMPC, FNMPC), and jointly simulated through Simulink and CarSim. Experimental results show that FNMPC effectively reduces the oscillation phenomenon caused by model mismatch, while enabling the model to have higher tracking accuracy. Among them, the absolute value of the displacement error of the feedforward nonlinear model prediction controller does not exceed 0.1106 m, and the absolute value of the heading error does not exceed 0.1253 rad. Under the same working conditions, linear model predictive control, feedforward linear model predictive control, pure tracking control and Stanley control errors diverge. However, the FNMPC proposed in this paper has higher tracking accuracy than the existing NMPC, and the absolute cumulative value of the control increment is far away. Lower than the NMPC algorithm.