Metamorphic Testing Harness for the Baidu Apollo Perception-Camera Module

<- Publications

Metamorphic Testing Harness for the Baidu Apollo Perception-Camera Module

Yifan Zhang, Dave Towey, Matthew Pike, Jia Cheng Han, George Zhou, Chenghao Yin, Qian Wang, Chen Xie

2023 IEEE/ACM 8th International Workshop on Metamorphic Testing (MET) | 2023 | View on Publisher's Website

Abstract

As autonomous driving systems (ADSs) grow in complexity, efficient and reliable testing methodologies become increasingly necessary. This study investigates the use of metamorphic testing (MT) to evaluate the perception-camera module of Baidu Apollo, an open-source ADS platform. Experiments reveal inconsistencies in obstacle detection under variations in brightness within specific regions of driving scenarios, demonstrating the utility of MT in addressing the oracle problem in ADS perception validation. Additionally, this paper presents a metamorphic testing harness designed to streamline ADS testing workflows, improving testing efficiency and organization. An industry case study illustrates the framework’s practical application in production environments, highlighting its potential benefits for real-world ADS testing.