In order to improve the trajectory tracking control performance of unmanned vehicles under different road adhesion coefficients and vehicle speeds, an adaptive horizon model predictive control (MPC) algorithm was proposed. First, the MPC trajectory tracking controller was designed based on a three-degree-of-freedom vehicle dynamics model. Second, a particle swarm optimization (PSO) algorithm incorporating quasi-reflective-based learning and Gaussian variation was introduced to optimize the horizon parameters and obtain the offline optimal horizon data sets under different operating conditions. Then, the adaptive network-based fuzzy inference system (ANFIS) was utilized to train the dataset to obtain a control system that can adaptively adjust the horizon. Finally, it was verified by Carsim and Simulink joint simulation and real vehicle verification. The results show that the trajectory tracking accuracy and stability of the MPC controller with adaptive horizon under different working conditions are improved substantially, which shows that the algorithm has good practicality.