A Declarative Metamorphic Testing Framework for Autonomous Driving

Autonomous driving has gained much attention from both industry and academia. Currently, Deep Neural Networks (DNNs) are widely used for perception control in autonomous driving. However, several fatal accidents caused by vehicles have raised serious safety concerns about models. Some recent studies successfully the metamorphic testing technique to detect thousands of potential issues some popularly prior study is limited a small set relations, which do not reflect rich, real-world traffic scenarios also customizable. This paper presents novel declarative rule-based framework called RMT . provides rule template with natural language syntax, allowing users flexibly specify an enriched based on rules domain knowledge. automatically parses human-written relations using NLP-based parser referring ontology list generates test cases variety image transformation engines. We evaluated three With detected significant number abnormal model predictions that were work. Through large-scale human Amazon Mechanical Turk, we further confirmed authenticity generated validity predictions.