Citation

Ji X, Fei C, Xu T, He X. Int. J. Veh. Des. 2020; 83(2/3/4): 240-257.

Copyright

(Copyright © 2020, Inderscience Publishers)

DOI

10.1504/IJVD.2020.115062

PMID

unavailable

Abstract

Autonomous vehicles are a research area of active interest. Collision avoidance system (CAS) is one of the central concerns to provide security protection for autonomous vehicles. This paper proposes a multi-mode collision avoidance system (mCAS), which combines a trajectory prediction module, a risk assessment module and a motion planning module into the closed-loop system. At each step, the trajectory prediction module predicts the trajectories of the vehicle and surrounding vehicles. The risk assessment model calculates the collision probability and chooses reasonable control mode. Then the motion planning module designs the desired deceleration profile based on it. The car takes the first step of the planned deceleration and repeats this cycle, achieving closed-loop control. The mCAS is tested in a closed-loop simulation setup and the results show that the proposed mCAS is of good effectiveness and feasibility, which can significantly reduce collision probability as well as false alarms.

Keywords: autonomous vehicles; CAS; collision avoidance system; trajectory prediction; LSTM; long short-term memory; network collision detection; Monte-Carlo methods; risk assessment; TTC; time-to-collision; multi-mode control strategy; motion planning.

Language: en