Collision Avoidance for Automated Vehicles Using Occupancy Grid Map and Belief Theory

Reza Soltani, Purdue University

Abstract

Decades ago, talking about the vehicle that can drive without human being intervention or self-driving cars seemed funny, unrealistic and unlikely compare to the airplanes because of the obstacles. Nowadays, researchers, have made it a reality by utilizing of sciences such as computer, electronic, mechanic etc. despite of massive progress, it seems that we are still at the beginning of the road. Huge number of accidents that occur due to human errors is a proof of its correctness. Lack of sufficient preventative driving aid such as collision avoidance system is one of the significant reason. Most of the collision happens when human drivers make mistakes. On the other hand, autonomous vehicles (AVs) will play an important role in future travels, and more accident preventatives are needed.For driving assistant system (DAS), an ego vehicle extremely needs a detailed environment representation especially in crowded urban area for static and dynamic object detection and free space determination. Due to inaccurate pose estimation and noisy measurements by using Lidar sensor scanning, some uncertainty is generated because of the perception of the new areas and errors. When a lidar generate multi-layer data and eco it is possible to get complexity increase. Different sources of uncertainty need to be managed. otherwise, collision is happening. For that reason, a beneficial representation of drivable area is offered to navigate autonomous vehicles. That would be occupancy grid map (OGM) which plays major role for the obstacle avoidance.A collision avoidance system (CAS) is the best way to estimate the risks associated with the surrounding obstacles and guides the AV into a safer path when the ego car encounter with a dangerous situation. CAS provides safe region information and consequently collisionfree area in the presence of other obstacles such as cars, pedestrian, signs etc. There are some common CAS functions forward and rear collision warning and emergency braking that can help to prevent some major and minor accident. These techniques have been developed lately. However, they are not sufficient and reliable preventative for limiting or avoiding unexpected potential crashes with the surrounding obstacles.This thesis discusses OGM, CAS and belief theory, and propose some of the latest and the most effective method such as predictive occupancy grid map (POGM), risk evaluation model and OGM role in the belief function theory with the approach of decision uncertainty according to the environment perception with the degree of belief in the driving command acceptability. Finally, how the proposed models mitigate or prevent the occurrence of the collision.

Degree

M.Sc.

Advisors

Li, Purdue University.

Subject Area

Aerospace engineering

COinS