Simulation-Based Validation for Autonomous Driving Systems

Li, Changwen; Sifakis, Joseph; Wang, Qiang; Yan, Rongjie; Zhang, Jian · 2023 · OpenAlex-citations

DOI: 10.1145/3597926.3598100

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the critical challenge of rigorously validating autonomous driving systems (ADS), arguing that current industrial simulation practices are insufficient. While companies often cite high volumes of simulated miles as evidence of safety, the authors contend that random exploration of scenarios fails to adequately cover relevant real-world situations, particularly risky configurations. The motivation is to establish a well-founded validation method that ensures simulation effectively maps to real-world safety, moving beyond simple realism to semantic awareness and systematic coverage of dangerous states. The proposed solution is a framework named RvADS, which integrates the existing LGSVL industrial simulator with a formally defined testing environment. This environment comprises a Scenario Generator and a Monitor. The core technical contribution is the extraction of a semantic model from the simulator, represented as a metric graph. This mathematical model formalizes the environment’s topology and geometry, defining vehicle states by their distribution on the map along with kinetic and time attributes. The authors assume single-lane roads and junctions equipped with traffic signals, transforming HD map annotations into this formal structure. The Scenario Generator uses this model to systematically generate test cases that drive the system toward specific high-risk configurations, such as complex junction maneuvers, rather than relying on random exploration. The Monitor then observes the system’s behavior against properties specified in first-order linear temporal logic, verifying safety constraints and limited reachability goals. The study demonstrates that this systematic approach to scenario generation is significantly more effective than random exploration. By targeting specific risky situations and traffic rule violations, the method uncovered numerous flaws in the real simulator that would have been difficult to detect otherwise. The framework successfully validates that the ADS satisfies short-term safety goals (avoiding collisions), medium-term maneuvering goals, and long-term trip completion goals. The runtime verification component provides diagnostics for property violations, offering concrete evidence of system trustworthiness. The significance of this work lies in bridging the gap between industrial simulation tools and rigorous formal methods. By providing a semantic model that respects fundamental properties of time and space, the approach enables precise validation of ADS behavior. It offers a viable path for global validation through simulation, ensuring that testing covers the necessary fraction of real-world situations. This method enhances the reliability of ADS validation by replacing ad-hoc, random testing with a structured, logic-based verification process that can identify critical safety flaws before deployment.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• simulator validity fidelity

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Methodological Resource: validation psychometrics, tool software
• Theoretical Contribution: computational model