Spatial-temporal risk field-based coupled dynamic-static driving risk assessment and trajectory planning in weaving segments

TL;DR

This paper introduces a novel spatial-temporal risk field for better anticipatory risk assessment and trajectory planning in weaving segments, improving safety and efficiency in autonomous vehicle navigation.

Contribution

It proposes a new 3D risk field model, a parameter calibration method using real-world data, and a STRF-based trajectory planning approach that enhances safety and reduces computation time.

Findings

STRF outperforms traditional risk fields in situational awareness.

The proposed planning method improves safety and reduces lane-change time.

Real-world tests show significant safety and efficiency gains.

Abstract

In this paper, we first propose a spatial-temporal coupled risk assessment paradigm by constructing a three-dimensional spatial-temporal risk field (STRF). Specifically, we introduce spatial-temporal distances to quantify the impact of future trajectories of dynamic obstacles. We also incorporate a geometrically configured specialized field for the weaving segment to constrain vehicle movement directionally. To enhance the STRF's accuracy, we further developed a parameter calibration method using real-world aerial video data, leveraging YOLO-based machine vision and dynamic risk balance theory. A comparative analysis with the traditional risk field demonstrates the STRF's superior situational awareness of anticipatory risk. Building on these results, we final design a STRF-based CAV trajectory planning method in weaving segments. We integrate spatial-temporal risk occupancy maps, dynamic iterative sampling, and quadratic programming to enhance safety, comfort, and efficiency. By incorporating both dynamic and static risk factors during the sampling phase, our method ensures robust safety performance. Additionally, the proposed method simultaneously optimizes path and speed using a parallel computing approach, reducing computation time. Real-world cases show that, compared to the dynamic planning + quadratic programming schemes, and real human driving trajectories, our method significantly improves safety, reduces lane-change completion time, and minimizes speed fluctuations.