IRT SystemX and its partners involved in the EPI project (AI-based Performance Evaluation of Decision Systems – Évaluation des Performances de systèmes de décision à base d’IA) filed a patent with the French National Institute for Industrial Property (INPI). The aim: offer promising leads to optimize the evaluation of an autonomous pilot.
Today, the evaluation techniques of an autonomous vehicle are exploited to validate all its decisions and, in fine, to enable its industrialisation. This validation process must undergo a set of standards and norms in connection with artificial intelligence and operational safety, in order to better manage the decisions of the autonomous vehicle and contribute to its certification process.
This patent, filed by IRT SystemX, Airbus Project (ex. APSYS), Expleo France, Stellantis and Naval Group, aims to offer a methodology to evaluate autonomous systems (relying on models based on reinforcement learning), and thus accelerate their certification. By pairing these AI-reinforcement learning methods with the constraints inherent to the operational safety of the vehicle (Safety/SOTIF), the team drafted relevant and secure driving scenarios depending on the environment in which the autonomous pilot is manoeuvring. This method does not require field data, which is almost non-existent – a true asset of this patent. In due course, it will be possible to integrate the method in the norms and standards necessary to validate the industrialization of AI-based systems for an autonomous vehicle, with the possibility to verify and validate industrial specifications and requirements.
Find out more:
