Trajectory planning is a core technology for autonomous vehicle directlyreflecting the driving safety and efficiency. In this paper, a spatial‐speed decoupled planning method is studied for real‐time trajectory generation in the on‐road environment. The proposed approach mainly includes two parts: optimization‐based lateral planning and selection‐based longitudinal speed planning. The optimization‐based lateral planning is employed to generate an optimal collision‐free and smooth path by solving a quadratic programming problem. Specifically, the lateral planning first constructs a safe corridor by integrating obstacles and initial risky corridor together in the Frénet frame. The safe corridor is useful for quadratic programming problem formulation. The selecting‐based longitudinal speed planning is proposed to generate a suitable and continuous velocity trajectory. The novel speed selector considering four cases is designed to select a more suitable reference velocity in different lane situations. The continuous velocity trajectory is obtained by solving piecewise continuous quartic polynomial. As a result, the combination of the spatial path and the velocity trajectory is the final planning result. The proposed algorithm is tested extensively in a simulation environment. Experimental results demonstrate that the proposed algorithm has good real‐time property and practical validity.