Dynamic High-Order Control Barrier Functions with Diffuser for Safety-Critical Trajectory Planning at Signal-Free Intersections

TL;DR

This work tackles safe and efficient autonomous driving through signal-free intersections by framing trajectory planning as conditional generation guided by task labels and explicit goals. It combines a diffusion-based DSC-Diffuser with a Dynamic High-Order Control Barrier Function (DHOCBF) safety layer to learn multi-task behaviors for left turns, straight movements, and right turns while guaranteeing forward invariance in dynamic environments. The approach demonstrates strong generalization to unseen scenes, achieving low displacement errors and high safety (SR≈1) across trained and untrained scenarios, with DHOCBF reducing conservatism compared to traditional CBFs. The proposed framework offers a practical path toward robust, safety-verified autonomous navigation at complex, unsignalized intersections, with potential extensions to dynamic parameter tuning and vehicle connectivity.

Abstract

Planning safe and efficient trajectories through signal-free intersections presents significant challenges for autonomous vehicles (AVs), particularly in dynamic, multi-task environments with unpredictable interactions and an increased possibility of conflicts. This study aims to address these challenges by developing a unified, robust, adaptive framework to ensure safety and efficiency across three distinct intersection movements: left-turn, right-turn, and straight-ahead. Existing methods often struggle to reliably ensure safety and effectively learn multi-task behaviors from demonstrations in such environments. This study proposes a safety-critical planning method that integrates Dynamic High-Order Control Barrier Functions (DHOCBF) with a diffusion-based model, called Dynamic Safety-Critical Diffuser (DSC-Diffuser). The DSC-Diffuser leverages task-guided planning to enhance efficiency, allowing the simultaneous learning of multiple driving tasks from real-world expert demonstrations. Moreover, the incorporation of goal-oriented constraints significantly reduces displacement errors, ensuring precise trajectory execution. To further ensure driving safety in dynamic environments, the proposed DHOCBF framework dynamically adjusts to account for the movements of surrounding vehicles, offering enhanced adaptability and reduce the conservatism compared to traditional control barrier functions. Validity evaluations of DHOCBF, conducted through numerical simulations, demonstrate its robustness in adapting to variations in obstacle velocities, sizes, uncertainties, and locations, effectively maintaining driving safety across a wide range of complex and uncertain scenarios. Comprehensive performance evaluations demonstrate that DSC-Diffuser generates realistic, stable, and generalizable policies, providing flexibility and reliable safety assurance in complex multi-task driving scenarios.

Figures (11)

• Figure 1: The framework of the DSC-Diffuser planner. In the training process, the labeled data are used to train the Diffuser for multi-task learning. Blue arrows represent the planning process of the proposed DSC-Diffuser in horizon $H$, while red arrows represent the generation process in the Diffuser. Orange points is are the goals. Green arrows indicate the input of labeled trajectory data used to train the model during the training phase.
• Figure 2: The optimal trajectory for ego vehicle A $p_A(t:t+H|t)$ to avoid a collision with other vehicle B $p_B(t:t+H|t)$, whose trajectory is predicted, at time $t$ over $H$ future steps. The safe distance, $d_{safe}(t)$ is dynamic. The HOCBF zone is the safe region, limited by HOCBF bounds, ensuring $h(\boldsymbol{s_t},\boldsymbol{s_{obs}(t)}) \geq 0$, so that the ego vehicle $A$ does not approach the other vehicle $B$ too closely.
• Figure 3: MA and GL scenarios for given situations. The orange lines are stop lines and pedestrian lines.
• Figure 4: Performance Comparison of HOCBF and DHOCBF with Dynamic Obstacles of Varying Speeds. When the obstacle is stationary (v=$0$$m/s$), DHOCBF and HOCBF generate the same trajectories, resulting in the green and blue lines overlapping in the figure. The speed of obstacles is set as $0, 1$$m/s$, $3$$m/s$. In this case, the ego vehicle is traveling in the same direction.
• Figure 5: Comparative Analysis of HOCBF and DHOCBF with Varying Obstacle Radii. The obstacles move at a speed of $2$$m/s$ in the positive x-direction.