Collision avoidance in on-road environment for autonomous driving

Collision avoidance is an essential part in autonomous navigation. This paper proposes a collision avoidance method in on-road environment for autonomous driving. The proposed method divides a road map into six lane-level regions, assigns risk observers to corresponding regions, evaluates collision risks of situations based on risk observer distribution, and determines behaviors to deal with various collision-risky situations. Risk observers, designed to recognize and analyze situations at the level of lanes with regards to safety, return collision risks of situations. Based on the results of each risk observer, the system determines collision-free maneuvers to follow a global path. The method was implemented in robot operating system (ROS) for a reusable and hardware-independent software platform and tested in a two-phase procedure; simulation tests through a visualization tool called RViz in ROS with logged data and in-vehicle tests on the closed road with other vehicles to verify that our system operates properly on the road for autonomous driving without collisions. It successfully performed in several scenarios which can be happened in real road environment. The demonstration videos for in-vehicle tests are linked at fhttp://youtu.be/cjLcUdGg1j0, http://youtu.be/tB2rC6gMhcM.