Original scientific paper

https://doi.org/10.7906/indecs.24.6.10

Critical Artificial Intelligence Validation in Autonomous Vehicles – Reconstruction of Decision Logic and Risk Analysis

Laszlo Ady ; Óbuda University, Doctoral School on Safety and Security Sciences, Budapest, Hungary *
Péter János Varga ; Óbuda University, Kandó Kálmán Faculty of Electrical Engineering, Budapest, Hungary
György Schuster ; Óbuda University, Kandó Kálmán Faculty of Electrical Engineering, Budapest, Hungary
Dániel Tokody ; Óbuda University, Doctoral School on Safety and Security Sciences, Budapest, Hungary & NextTechnologies Ltd., Maglód, Hungary

* Corresponding author.

Abstract

The integration of Artificial Intelligence into Autonomous Vehicles presents a paradigm shift in transportation, promising enhanced safety and efficiency. However, the inherent complexity and “black-box” nature of many Artificial Intelligence models, particularly deep learning systems, pose unprecedented challenges for safety validation and certification. This article addresses the critical need for reconstructing the decision logic of Autonomous Vehicle Artificial Intelligence systems to enable rigorous quantitative risk analysis. We review the current landscape of Autonomous Vehicle incidents and regulatory standards, highlighting the gap between Artificial Intelligence capabilities and established safety integrity requirements. The core contribution of this article is the introduction of a novel, comprehensive risk metric RAI, and a corresponding validation criterion, the Risk Balance Rate. These formulae integrate logic reconstruction, environmental context, and system criticality into a quantifiable framework for safety assessment. We further provide a qualitative comparison of different Artificial Intelligence model types – Traditional Machine Learning, Deep Learning, and Bayesian Networks – against the requirements for logic transparency, reliability, and risk. Our findings suggest that a hybrid approach, combining the strengths of different model paradigms with a robust reconstruction methodology, is essential for achieving the levels of validation required for public trust and regulatory approval. This article concludes that the proposed quantitative framework is a vital step toward standards-compliant, safe, and justifiable deployment of autonomous driving technology.

Keywords

autonomous vehicles; artificial intelligence; safety validation; risk analysis; functional safety

Hrčak ID:

350263

URI

https://hrcak.srce.hr/350263

Publication date:

30.12.2026.

Visits: 0 *