Author(s)

Aleš Filip

Abstract

In recent years, artificial intelligence (ai) has found numerous applications in medicine, energy, industry and various transport sectors, including rail and road. the use of ai for autonomous train operation is listed as one of the research challenges in the new master plan of the european railway Joint Undertaking (October 2021). Nowadays, AI and machine learning (ML) algorithms are also widely used in connected self-driving cars (SDCs) for detection, classification and localization of objects on roads. naturally, the rail industry also wants to benefit from recent advances in SDCs. While the current level of safety on the railways is acceptable to society, mass deployment of SDCs is expected to significantly reduce the number of accidents caused by human driver behaviour. Safety is thus currently a major challenge in the development of driverless cars. in contrast, various driverless automatic train operation (ATO) systems supported by automatic train protection with guaranteed high safety integrity level (SIL 4) have been introduced in the last decades, but mainly on segregated networks such as the metro. Therefore, the aim of SDC technology transfer is to go beyond segregated lines and develop fully autonomous driverless trains for open rail networks. In this paper, a comparative analysis was used to show how the required safety is assured in automated driving of trains and cars. The results of the analysis describe the differences, intersections and synergies in these two different application areas, in particular in terms of the basic pillars of safety, the safety standards and regulations used, interoperability requirements, safety demonstration, certification and independent assessment. Finally, the paper summarises how the rail experience in safety could be used to improve SDC safety, or conversely, how the ATO could benefit from transferring the latest AI and ML technologies developed specifically for SDCs.

Keywords

automatic train operation, autonomous vehicles, machine learning, self-driving cars.