In this paper, a safety-guaranteed game-theoretical velocity planning framework in a hierarchical manner is proposed to generate safe, ride comfort, and travel efficiency-balanced velocity for autonomous vehicles (AVs). In the upper layer, a bang-bang decision-making method is utilized to determine which planning mode to be implemented based on acceleration and jerk constraints, including a comfort mode, an efficiency mode, and a game mode. In the lower layer, asymmetric jerk limits based on comfort characteristics sensibility analysis and safe velocity simultaneously considering longitudinal and lateral stability are firstly developed to maintain ride comfort and driving safety, respectively on curve roads, especially sharp curves where vehicle stability may be not fully considered in most researches. Based on these, a non-cooperative game-theoretical velocity planning method is presented to solve the conflict between comfort mode and efficiency mode by optimizing his own objective based on the other’s action. Finally, for the sake of solving efficiency and accuracy, a chaos optimization-based algorithm (COA) is designed to solve for the Stackelberg equilibrium solution of the bilevel game optimization problem. Three experimental tests are carried out to comprehensively demonstrate the effectiveness, robustness, and real time of the proposed framework. The results show that the proposed method can provide the great performance of ride comfort, travel efficiency, and longitudinal-lateral stability in real time in the velocity planning process.