A Study on Testing Autonomous Driving Systems

In recent years, with the rapid development of artificial intelligence and other related technologies, the traditional automotive industry has begun to integrate information technology in an all-round way. Due to the contributions of computer vision, deep learning, and sensitive sensors, autonomous driving systems (ADS) has now achieved great progress. But as we all know, the primary requirement for autonomous driving is absolute safety. However, technology innovation has brought great challenges to the testing of ADS, and due to the high cost of field testing, industrial companies rarely open relevant test data for research. This paper aims to study existing testing methods for ADS. Our study shows that there are few published works focusing on testing aspects of ADS. However, there is an obvious trend on the record of published works on testing ADS. Also, we can find that most reviewed works focus on setting up virtual test environment including generating, synthesizing, or reconstructing test input data. They either treat ADS as a whole to conduct (sub) system level testing or limit ADS into certain scenarios. From this, we believe that testing of ADS has just begun to attract researchers' interest; great effort should be paid before ADS becomes maturer.