Towards the Formal Semantics of Scenario Tests for Autonomous Vehicles

Testing self-driving components in autonomous vehicles is a challenging task. Such components need to interact with a complex and continuously changing environment, making traditional software testing approaches ineffective or impractical. Scenario-based testing approaches aim to define various traffic situations to support this testing of autonomous components by providing sensor inputs for them while monitoring the output of the actuators. However, we need a formal description of traffic scenarios to measure some coverage metrics or synthesize traffic scenarios. This paper proposes mathematically precise behavior formalization to achieve this by using graph transformation rules. We show that our formalization can cover existing scenario specification implementation, such as Scenic.

Paper

Full text

PDF

Towards the Formal Semantics of Scenario Tests for Autonomous Vehicles

Semantic Scholar · Computer Science · 2021

Abstract

Testing self-driving components in autonomous vehicles is a challenging task. Such components need to interact with a complex and continuously changing environment, making traditional software testing approaches ineffective or impractical. Scenario-based testing approaches aim to define various traffic situations to support this testing of autonomous components by providing sensor inputs for them while monitoring the output of the actuators. However, we need a formal description of traffic scenarios to measure some coverage metrics or synthesize traffic scenarios. This paper proposes mathematically precise behavior formalization to achieve this by using graph transformation rules. We show that our formalization can cover existing scenario specification implementation, such as Scenic.

Similar papers

© 2026 NYSGPT2525 LLC