Safe AI in Autonomous Vehicles

Abstract

Today, the most prominent application of AI technology in the automotive domain is in the realm of environment perception. The diversity of the traffic environment and the complexity of sensor readings make it impossible to specify and implement perception functionality manually. Deep learning technology, on the other hand, has proven itself capable of solving the task very well. However, it is important to note that effectiveness alone does not guarantee a comprehensive solution, and the issue of validation currently remains unsatisfactorily resolved. The track provided different perspectives on the challenges pertaining to the use of AI/ML technology in highly automated driving functions, including considerations on safety verification and validation techniques for AI-based autonomous vehicles, formal methods and their application in assuring the safety of AI-based autonomous systems robustness and resilience of AI algorithms in uncertain and open environments, system architectures for AI-based autonomous vehicles, data-driven approaches for safety assurance and risk analysis in autonomous driving, safety standards, regulations, and certification processes for AI-based autonomous vehicles, as well as testing, simulation, and validation methodologies for autonomous vehicle systems.