LiDAR-based Obstacle Detection and Avoidance for Autonomous Vehicles using Raspberry Pi 3B

Abstract

Autonomous vehicles are redefining the transport industry -obstacle detection and avoidance are key to their operation. A number of sensor technologies have been developed and trialled. This paper presents the implementation of a Hokuyo URG-04LX Light Detection And Ranging (LiDAR) sensor on an autonomous vehicle developed with a Raspberry Pi 3B microcontroller and demonstrates its effectiveness for object detection and avoidance in varying conditions. The LiDAR sensor was integrated with the Raspberry Pi 3B using Python on LUbuntu (lightweight version of Ubuntu) and Robot Operating System (ROS). Various scenarios with low light conditions, reflective surfaces at multiple angles, simple stopping tests and different motion paths at varying speeds were tested. All tests were run at 3.2 and 4mph speed. It was found that the LiDAR sensor performed well for basic object detection but did not respond well to reflective or dark surfaces. We further compared the LiDAR's performance with ultrasonic sensors and found that it outperformed ultrasonic sensors for stopping distances. Overall, the LiDAR acts as an effective sensor for the autonomous vehicle, showing its viability in detecting objects and acting as a small scale representation of autonomous technology.

A Review on Sensor Technologies and Control Methods for Mobile Robot with Obstacle Detection System

“…According to the study by [14], they implemented obstacle detection by using a 2D Light Detection and Ranging (LiDAR) sensor on an autonomous vehicle with a robot operating system (ROS). They used the data collected by the LiDAR sensor to create a 2D mapping on Rviz and also reacting to the obstacle if the distance reading from the sensor reached a certain threshold.…”

“…They used the data collected by the LiDAR sensor to create a 2D mapping on Rviz and also reacting to the obstacle if the distance reading from the sensor reached a certain threshold. Tests were carried out for the LiDAR based obstacle detection to detect objects and moving pedestrians and identified that the LiDAR sensor does not respond well to reflective surfaces as the laser light from the LiDAR sensor may not deflected properly, resulting in varied measurement distance that only obtained an average accuracy of 85.92% from the Lidar sensor and actual distance, which also occurs in the research of [14], that utilised Braitenberg strategy, pedestrian detection shows the LiDAR data with higher fluctuation and performs poorer when compared to the ultrasonic sensor, which is likely due to the use of only single laser model for LiDAR measurements, and robot also cannot avoid transparent objects such as acrylic and polycarbonate bottles. The experiment result of [15] shows that the average accuracy during optimal conditions is 99.1%.…”