Tree-Based Scenario Classification: A Formal Framework for Coverage Analysis on Test Drives of Autonomous Vehicles

TL;DR

This paper introduces a formal framework using logic-based classifiers and feature trees to classify and measure coverage of scenario sets in autonomous vehicle test drives, enhancing safety analysis.

Contribution

It presents a novel formal framework for classifying scenario sets and measuring coverage in autonomous vehicle testing using logic-based classifiers and feature trees.

Findings

Effective classification of urban driving scenarios

Demonstrated approach on simulation data

Enhanced coverage measurement for scenario sets

Abstract

Scenario-based testing is envisioned as a key approach for the safety assurance of autonomous vehicles. In scenario-based testing, relevant (driving) scenarios are the basis of tests. Many recent works focus on specification, variation, generation and execution of individual scenarios. In this work, we address the open challenges of classifying sets of scenarios and measuring coverage of theses scenarios in recorded test drives. Technically, we define logic-based classifiers that compute features of scenarios on complex data streams and combine these classifiers into feature trees that describe sets of scenarios. We demonstrate the expressiveness and effectiveness of our approach by defining a scenario classifier for urban driving and evaluating it on data recorded from simulations.