SMET: Scenario-based Metamorphic Testing for Autonomous Driving Models

TL;DR

SMET is a scenario-based metamorphic testing tool designed to enhance the security and robustness of autonomous driving models by detecting potential defects through complex scene analysis.

Contribution

The paper introduces SMET, a novel testing framework that utilizes scenario dimensions to identify defects in autonomous driving models.

Findings

SMET effectively detects potential defects in autonomous driving models.

Complex scenes are more effective for testing than simple scenes.

The tool demonstrates success across different autonomous driving models.

Abstract

To improve the security and robustness of autonomous driving models, this paper presents SMET, a scenariobased metamorphic testing tool for autonomous driving models. The metamorphic relationship is divided into three dimensions (time, space, and event) and demonstrates its effectiveness through case studies in two types of autonomous driving models with different outputs.Experimental results show that this tool can well detect potential defects of the autonomous driving model, and complex scenes are more effective than simple scenes.