Collision avoidance testing for autonomous vehicles

• Main Author: Koesno, Margaret Claire
• Other Authors: Liu Yang
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/148467
• Institution: Nanyang Technological University

Autonomous Vehicles (AVs) have been one of the most anticipated technological breakthroughs for the past decade. Companies are investing billions into research and development with the hope that self-driving cars can be on the road by 2025 and even commercially available by 2030. Research has shown that having AVs on the road will revolutionize road transportation by reducing congestion, improving traffic speed, and even increasing road safety. As of right now, the industry is still far away from getting highly automated AVs on the road. Surprisingly, this is not due to the lack of technological capabilities, but rather due to the need for more research and work to be done on sufficiently testing these AV systems. Ensuring the safety of AV cars has been the most challenging and important issue in the development cycle. This project aimed to perform collision avoidance testing on an AV system and detect the underlying issues that may lead to dangerous scenarios. An AV Testing Platform written in JavaScript was used to drive the testing procedure and communicate with Apollo, the autonomous driving system, and the SVL Simulator, a testing simulator. This project will also look into investigating the effectiveness of different test case generation algorithms, namely the Genetic algorithm and the Random algorithm. The effectiveness of the algorithm will be judged based on speed and ability to generate collision test cases. Twelve batches of the test cases were run, and the results were analyzed accordingly. Eight different issues were identified to have been the reason for collision during the test case scenarios. Furthermore, it was proven that the Genetic algorithm, although superior to the Random algorithm, would benefit from being further optimized.