Safe Hierarchical Autonomous Driving Framework: Integrating SAC-D Planning With HOCLF-HOCBF-QP Constrained Low-Level Control

Autonomous driving requires decision-making and motion control algorithms that can ensure both safety and efficiency in complex traffic environments. To address these challenges, this study proposes a hierarchical decision making and control framework that integrates a Soft Actor-Critic Discrete (SAC-D) planner with a High-Order Control Lyapunov Function and High-Order Control Barrier Function Quadratic Programming (HOCLF-HOCBF-QP) based controller. In this approach, the high-level SAC-D planning agent operates on a discrete action space, generates lane-level maneuver commands (such as lane keeping, left lane change, or right lane change), and learns an optimal decision policy through various traffic interaction scenarios. In the low-level control phase, HOCLFs enforce system stability, ensuring the vehicle follows a desired trajectory, while HOCBFs define safety boundaries, preventing collisions between the ego vehicle and other road users. The optimal control input is then obtained by solving a QP problem, which ensures that both stability and safety constraints are satisfied. A simulation study comprehensively evaluates the proposed framework using both roundabout and intersection traffic scenarios. Simulation results demonstrate that the SAC-D and HOCLF-HOCBF-QP control framework enables the ego-vehicle to perform robust and collision-free maneuvers, offering a reliable solution for safe autonomous navigation in complex environments. Real-time implementation capability of the SAC-D and HOCLF-HOCBF-QP is demonstrated using a Hardware-in-the-Loop system, where the low-level control algorithm and the high-level decision making are successfully executed in an automotive electronic control unit and a separate computational unit, respectively.