To help the autonomous vehicle plan a safe， comfortable and efficient driving trajectory， a trajectory planning approach based on model predictive control was proposed. First， to simplify the planning environment， a safe and feasible “three-circle” expansion of the safety zone was introduced， which also eliminates the collision issues caused by the idealized model of the vehicle. Then， the trajectory planning was decoupled in lateral and longitudinal space. A model prediction method was applied for lateral planning to generate a series of candidate trajectories that met the driving requirements， and a dynamic planning approach was utilized for longitudinal planning， which improved the efficiency of the planning process. Eventually， the factors affecting the selection of optimal trajectories were considered comprehensively， and an optimal trajectory evaluation function was proposed for path planning and speed planning more compatible with the driving requirements. The effectiveness of the proposed algorithm was verified by joint simulation with Matlab/Simulink， Prescan and Carsim software. Experimental results indicate that the vehicle achieves the expected effects in terms of comfort metrics， steering wheel angle variation and localization accuracy， and the planning curve also perfectly matches the tracking curve， which validates the advantage of the proposed algorithm.