Evaluation testing and predictive modeling services for AI-powered autonomous vehicles

The deployment of autonomous and intelligent systems within the modern automotive digital supply chain requires robust validation methods that demystify the "black box" nature of AI. Building upon seven years of research and the recent CIFRE PhD thesis by Céline Serbouh in partnership with the Conservatoire National des Arts et Métiers (CNAM) and its CEDRIC laboratory, UTAC has launched a unified service for the evaluation testing and predictive modeling of vehicles equipped with artificial intelligence.

Decrypting AI behavior through predictive modeling
Since 2019, UTAC has anticipated the arrival of highly automated mobility systems by developing methodologies to understand how these vehicles react in complex environments. The new service utilizes predictive modeling to make AI systems more explainable. By identifying the specific factors that influence vehicle decision-making algorithms, engineers can gain deeper insights into dynamic responses across a multitude of driving situations. This capability is essential for ensuring the safety and reliability of robotaxis, autonomous shuttles, delivery robots, drones, and mobile industrial robots.

This comprehensive offering supports manufacturers throughout the entire development cycle, from the initial project scoping phase directly through to final safety analysis and validation.

Optimizing test campaigns with generative AI
A critical outcome of this research is the optimization of validation campaigns through the targeted identification of the most relevant and critical driving scenarios. The UTAC service employs generative AI techniques, utilizing data collected from real-world testing to create highly representative virtual scenarios.

This hybrid approach allows manufacturers to explore a broader spectrum of edge-case use cases while significantly reducing evaluation timelines and physical testing costs. By modeling these environments, development teams can effectively reproduce and anticipate the specific behaviors of intelligent systems to accurately assess both their performance capabilities and their operational limitations.

Additional Context: This section details technical specifications and competitive benchmarking not included in the original product announcement
In the rapidly evolving market of homologation and validation for autonomous driving (AD) and advanced driver assistance systems (ADAS), testing providers are increasingly shifting from purely physical track testing to hybrid physical-virtual validation. Competitors such as TÜV SÜD, Applus+ IDIADA, and AVL have heavily invested in scenario-based testing frameworks to address the regulatory requirements for Level 3 and Level 4 autonomous systems.

For example, AVL's simulation toolchains and IDIADA's connected and automated vehicle (CAV) testing facilities provide robust virtual and physical track capabilities. The technical distinction of UTAC’s new service lies in its specific integration of generative AI to decode the deep learning algorithms of perception and control systems. By offering predictive modeling as a core service, UTAC not only tests whether an autonomous vehicle safely navigates a scenario but explicitly identifies the underlying factors of the AI's decision-making process. This bridges a crucial gap in explainable AI (XAI), which is becoming a fundamental requirement for the future regulatory homologation of highly automated mobility solutions.

Edited by Maria Brueva, Induportals editor – adapted by AI.

www.utac.com