Integrated decision-making and motion planning system for emergency autonomous vehicles

Abstract

Autonomous vehicle applications are becoming increasingly widespread, with rapid technological advancements driven by the joint efforts of academia and industry, allowing these vehicles to play a vital role in daily life. Beyond fundamental perception and prediction technologies, decision-making and planning for civilian vehicles have seen significant development. However, emergency autonomous vehicles (EAVs), driven by the need for speed, present unique decision-making and planning challenges. This thesis addresses two key aspects of EAVs: speed-oriented decision-making and safety-critical motion planning.

EAVs, such as ambulances, require specialized decision-making for high-interaction scenarios in multi-lane traffic. They demand real-time decisions with high sensitivity to dynamic traffic conditions, proactive speed-oriented exploration, and a more aggressive focus on efficiency. Current decision-making methods, such as data-driven approaches or reinforcement learning for lane-change decisions, are mostly designed for regular vehicles and struggle to meet the specific needs of EAVs, especially in high-stakes, time-sensitive situations. Another challenge arises when coupling speed-oriented decision-making with safety-critical motion planning. In this case, motion planning must explore a larger solution space to enhance speed efficiency and ensure effective collision avoidance with multiple surrounding vehicles. Safety becomes paramount given the high stakes of EAVs' operation in dynamic, multi-vehicle environments. In particular, forward invariance ensures that the system stays within a safe set of states over time, maintaining safety constraints throughout the maneuver, which prevents the vehicle from entering unsafe regions or violating constraints as it progresses along its trajectory.

This thesis proposes agile and reliable systems for EAVs, addressing the unique challenges they face in high-speed, dynamic traffic environments. Unlike existing systems, our approach integrates decision-making with motion planning, enabling the efficient implementation of speed-oriented strategies across both domains. Furthermore, the safety-critical motion planner leverages quadratic optimization to ensure effective collision avoidance, thereby guaranteeing the vehicle's safety in complex scenarios. This dual focus on efficiency and safety allows for flexible and rapid exploration of speed benefits in mixed-traffic environments, where multiple lanes are shared with human-driven vehicles. Simulation results demonstrate notable improvements in both efficiency and progress, while also ensuring robust safety under various traffic conditions.