Real-time self-driving car navigation and obstacle avoidance using mobile 3D laser scanner and GNSS

Abstract

Self-driving car navigation is currently attracting considerable research interests. The key problem is to guide the car to the destination in real-time with a safe and obstacle free path in real-world environments. We propose an innovative self-driving car navigation approach that incorporates a VFH (Vector Field Histogram) local path planner adapted for modern 3D laser scanners and a global path planner using satellite positioning and digital map along with a custom built PID controller (Proportional Integral Derivative controller). Since classical path planning methods such as VFH are often used on small robots with ultrasonic rangefinders or in simulation based environments, we applied the VFH method to a real self-driving car with two different LiDAR (Light Detection And Ranging) configurations. The quantitative results from extensive experiments indicate that the developed VFH method with the modern real-time 3D LiDAR generally outperform the conventional LIDAR in terms of efficiency, accuracy and reliability. In addition, the tracks produced by 3D LIDAR are more convergent, smooth and consistent than the other configuration. The maximum position deviation for the VFH with 3D LiDAR is 0.28 m and ź0.16 m while the deviation for the other low-cost solution is 0.88 m and ź0.49 m respectively. The global path planner can provide an accuracy of within 1 meter most of the time. The proposed approach is successfully implemented and tested on our self-driving car which took part in the national self-driving car competitions in recent years, and ranked No.3 and 4. in the future challenge 2014 self-driving car competition in China.