A Unified Framework for Scenario CoverageEstimation and Representative Selection inSimulation-Based Testing of Automated Vehicles
Year: 2026
Type: preprint
Abstract: Scenario-based testing is now central to automated vehicle validation, yet a key challenge remains: how to select a limited set of scenarios that faithfully represents a given Operational Design Domain (ODD). The combinatorial explosion of scenario types, continuous parameters, and categorical factors makes exhaustive testing infeasible and current approaches fragmented across scenario classes. This paper proposes a unified smart sampling framework for multi-scenario selection: Driving situations are encoded as time series in a scenarioagnostic representation and projected into a common latent space, enabling the selection of a minimal, statistically representative subset that preserves both inter-scenario proportions and intra-scenario distributions, including categorical conditions. Applied to more than 280,000 real-world scenarios, the method reduces the test set by up to 97% while maintaining statistical fidelity. It provides quantifiable guarantees on representativeness, directly linking sample size to uncertainty, and identifies missing conditions for targeted data completion. This approach enables manufacturers to drastically reduce testing cost while preserving coverage and reliability.
Source: HAL
DOI: 10.82157/dsa/2026/20
Language: English
Subfield: Automotive Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7222736280