Measuring Train Driver Performance as Key to Approval of Driverless Trains

TL;DR

This paper addresses how to establish safety targets for driverless trains by anchoring them to human driver performance within the CSM-RA risk-acceptance framework. It presents ATOSenseData, a public, simulator-based dataset of 711 human-driver decisions (RT, time-to-arrival, and distance-to-obstacle) collected across speeds, obstacle sizes, protection systems, and color contrasts from two labs, to inform regulation and standardization. The dataset supports the development of standards (e.g., DIN SPEC 91516) and provides a basis for safety analyses and risk acceptance when evaluating computer-vision-based obstacle detection in DTO/UTO systems. By offering raw and post-processed statistics and emphasizing non-parametric analysis due to non-normal RT distributions, the work establishes a practical foundation for future probabilistic studies and real-world baselines in driverless train certification.

Abstract

Points 2.1.4(b), 2.4.2(b) and 2.4.3(b) in Annex I of Implementing Regulation (EU) No. 402/2013 allow a simplified approach for the safety approval of computer vision systems for driverless trains, if they have 'similar' functions and interfaces as the replaced human driver. The human driver is not replaced one-to-one by a technical system - only a limited set of cognitive functions are replaced. However, performance in the most challenging function, obstacle detection, is difficult to quantify due to the deficiency of published measurement results. This article summarizes the data published so far. This article also goes a long way to remedy this situation by providing a new public and anonymized dataset of 711 train driver performance measurements from controlled experiments. The measurements are made for different speeds, obstacle sizes, train protection systems and obstacle color contrasts respectively. The measured values are reaction time and distance to the obstacle. The goal of this paper is an unbiased and exhaustive description of the presented dataset for research, standardization and regulation. The dataset with supplementing information and literature is published on https://data.fid-move.de/de/dataset/atosensedata

Figures (3)

• Figure 1: Simulation environments of ATOSenseData atosenseeiart. On the left side, the simulation environment of BBI resembles a driver's cab of the 193/248 series (Vectron Dual Mode). The software used is Zusi 3 Professional with the display being a 32-inch UHD screen. On the right side, DLR used the simulation environment RailSET® johne2016railsite. An original control panel of a traction unit together with the VIRES software were used. In addition to the front screen, views from the side windows and ambient sounds were provided.
• Figure 2: Contrast of stimuli in the BBI simulation environment in ATOSenseData atosenseeiart, high on the left and low on the right side. Right picture shows a curved track, which did not appear in experiments.
• Figure 3: Shapiro-Wilk normality test of logarithmic RT for every configuration of four PSFs. The PSFs in a configuration are denoted by a comma-separated list of their values as visual angle, speed, train protection system and contrast. The x-axis plots the logarithmic RT in seconds and y axis the absolute number of samples. $n$ refers to the number of samples and $p$ denotes the p-value for normal distribution.