An Enhanced Performance Analysis of Evolutionary Algorithms for Automated Design of Autonomous Vehicles

Abstract

This paper presents an extensive performance analysis of evolutionary algorithms (EA) used for automated design of autonomous vehicles (AVs). This research explores the algorithms' abilities to generate AV designs that exhibit safe driving behavior and also meet requirements concerning comfort, efficiency and reliability. The assessment of the EA performance is based on the analysis of two design scenarios-corridor driving and intersection crossing. On each of these scenarios, the performance of two distinct evolutionary algorithms was compared against several baselines, including a hand-crafted controller and several other methods from the literature. The results showed that the EA-generated designs outperform the other methods in terms of safety, efficiency and general performance on both of the test scenarios. Furthermore, the assessment revealed some interesting distinctions between both the tested evolutionary algorithms, which could be useful for practitioners and developers of autonomous vehicles. Overall, the results support the conclusion that evolutionary algorithms can be a reliable and effective tool for automated generation of safe and efficient AV designs