A Risk-Field Based Motion Planning Method for Multi-Vehicle Conflict Scenario

In the present studies of autonomous driving systems, motion planning of automated vehicles in complex traffic scenarios is a tough issue. The problem of adjusting automated vehicles driving style is a difficult one. To address the above issues, this article focuses on the spatio-temporal motion planning problem of automated vehicles in multi-vehicle conflict scenarios. A risk-field based motion planning method is proposed. By predicting the state of the vehicles in the traffic scenario, the risk field matrix is built for motion planning. The idea of path-velocity separation planning is adopted to divide the motion planning into two layers as path planning and trajectory planning. In path planning, static risk matrices serve as constraints. Considering the driving safety and comfort, the path cost function is designed to evaluate the candidate path group, obtained by longitudinal sampling and lateral sampling, to determine the target path. The dynamic risk matrices are used to construct the spatio-temporal graph corresponding to the target path, based on which dynamic planning and quadratic planning are introduced to solve the trajectory planning problem. Simulations were built to prove the practicability of the method. Results show that by adjusting risk threshold can change the aggressiveness of autonomous vehicles driving style.

I. Introduction

Automated vehicle technologies have received a lot of attention in recent years because of their potential to improve traffic safety and efficiency. Automated driving is a difficult task that necessitates a thorough grasp of the dynamic traffic environment, as well as the ability to drive safely in a variety of conditions [1]. Due to developments in artificial intelligence and computer vision, automated vehicles have seen some early success in the last few decades [2], [3]. But there are still some unresolved issues, such as detection problems on unstructured roads, balance between optimal vehicle control and drivers’ user experience and take-over issues for human-machine co-driving [4], [5], [6].