Occlusion-Aware Contingency Safety-Critical Planning for Autonomous Driving

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Autonomous vehicles (AVs) face significant challenges in dynamic and occluded environments, impacting safety and efficiency.
• Real-time risk assessment and trajectory planning are crucial for AV navigation in complex scenarios.

Purpose of the Study:

• To propose an occlusion-aware contingency safety-critical planning approach for real-time autonomous driving.
• To enhance the safety and travel efficiency of AVs in dynamic, occluded environments.

Main Methods:

• Utilized reachability analysis and forward reachable sets (FRSs) of phantom vehicles (PVs) for risk assessment.
• Developed a biconvex nonlinear programming (NLP) formulation with spatiotemporal barrier constraints for safety enforcement.
• Employed consensus alternating direction method of multipliers (ADMMs) for real-time computation and decomposition of the NLP problem.

Main Results:

• Demonstrated enhanced safety and improved travel efficiency in simulations and real-world experiments.
• Successfully enabled real-time safe trajectory generation in occluded intersections under varying obstacle conditions.
• Validated the effectiveness of risk-aware dynamic velocity boundaries derived from FRSs.

Conclusions:

• The proposed occlusion-aware planning approach effectively addresses safety and efficiency challenges for AVs in dynamic, occluded environments.
• The method facilitates real-time, safe trajectory generation, crucial for practical AV deployment.
• The integration of reachability analysis and ADMM-based optimization offers a robust solution for safety-critical autonomous driving.