基于自动驾驶场景语义的蜕变测试优化方法OA

Equipped with advanced sensors,autonomous driving systems utilize artificial intelligence technology to perform environmental perception,decision-making,planning,and control,thereby assisting and replacing humans in completing driving tasks.To ensure the safety and reliability of autonomous driving systems,comprehensive testing is imperative.To address the issues of excessive redundancy in test cases and the difficulty of automatically matching appropriate metamorphic relations for these cases in end-to-end autonomous driving systems under regression testing contexts,we propose a metamorphic testing optimization method based on autonomous driving scenario semantics(MetaOpt),aimed at improving the efficiency of metamorphic testing for autonomous driving systems.Specifically,the proposed method involves cross-modal semantic information extraction from test cases,transforming unstructured test cases into structured scene element semantic information data.Then,test cases are reduced based on the similarity of scene element categories and distances.Following that,clustering of metamorphic relations is performed based on element keywords,along with matching"test case-metamorphic relation pairs".Finally,test cases are prioritized based on the rarity of elements in the scenes.Experimental results indicate that MetaOpt effectively reduces test cases,matches metamorphic relations and test cases,and enhances the error detection rate of autonomous driving models as well as the efficiency of metamorphic testing.