Onboard Sensor Systems for Automatic Train Operation

This paper introduces the specific requirements of the domain of train operation and its regulatory framework to the AI community. It assesses sensor sets for driverless and unattended train operation. It lists functionally justified ranges of technical specifications for sensors of different types, which will generate input for AI perception algorithms (i.e. for signal and obstacle detection). Since an optimal sensor set is the subject of research, this paper provides the specification of a generic data acquisition platform as a crucial step. Some particular results are recommendations for the minimal resolution and shutter type for image sensors, as well as beam steering methods and resolutions for LiDARs.