A Risk-aware Spatial-temporal Trajectory Planning Framework for Autonomous Vehicles Using QP-MPC and Dynamic Hazard Fields

TL;DR

This paper introduces a risk-aware trajectory planning framework for autonomous vehicles that combines QP-MPC with dynamic hazard fields, improving safety, efficiency, and comfort in complex environments.

Contribution

It presents a novel cost function integrated into QP-MPC using dynamic hazard fields, enabling better risk management and performance validation across diverse scenarios.

Findings

Outperforms benchmark methods in efficiency, stability, and comfort.

Effective risk assessment using dynamic hazard fields.

Validated across complex driving tasks like lane-changing and intersection crossing.

Abstract

Trajectory planning is a critical component in ensuring the safety, stability, and efficiency of autonomous vehicles. While existing trajectory planning methods have achieved progress, they often suffer from high computational costs, unstable performance in dynamic environments, and limited validation across diverse scenarios. To overcome these challenges, we propose an enhanced QP-MPC-based framework that incorporates three key innovations: (i) a novel cost function designed with a dynamic hazard field, which explicitly balances safety, efficiency, and comfort; (ii) seamless integration of this cost function into the QP-MPC formulation, enabling direct optimization of desired driving behaviors; and (iii) extensive validation of the proposed framework across complex tasks. The spatial safe planning is guided by a dynamic hazard field (DHF) for risk assessment, while temporal safe planning is based on a space-time graph. Besides, the quintic polynomial sampling and sub-reward of comforts are used to ensure comforts during lane-changing. The sub-reward of efficiency is used to maintain driving efficiency. Finally, the proposed DHF-enhanced objective function integrates multiple objectives, providing a proper optimization tasks for QP-MPC. Extensive simulations demonstrate that the proposed framework outperforms benchmark optimization methods in terms of efficiency, stability, and comfort across a variety of scenarios likes lane-changing, overtaking, and crossing intersections.