Mainline Automatic Train Horn and Brake Performance Metric

TL;DR

This paper addresses the lack of a standardized performance metric linking rail-safety requirements with AI perception for mainline trains. It proposes a preliminary obstacle-detection submetric to evaluate perception systems, distinguishing collision-detection and non-contact obstacle detection, and grounds safety argumentation in CSM-RA risk assessment and reference-system comparisons. The core contribution is a percentile-distance based obstacle-detection metric for moving trains that trades off distance of detection against false-positive rates, with explicit consideration of the structure gauge, RoI, LROD, and ODD-data. The work aims to standardize benchmarking, guide data collection, and enable prediction of accident rates for particular perception systems within a chosen ODD, ultimately supporting safer, driver-replacing automation in mainline railways.

Abstract

This paper argues for the introduction of a mainline rail-oriented performance metric for driver-replacing on-board perception systems. Perception at the head of a train is divided into several subfunctions. This article presents a preliminary submetric for the obstacle detection subfunction. To the best of the author's knowledge, no other such proposal for obstacle detection exists. A set of submetrics for the subfunctions should facilitate the comparison of perception systems among each other and guide the measurement of human driver performance. It should also be useful for a standardized prediction of the number of accidents for a given perception system in a given operational design domain. In particular, for the proposal of the obstacle detection submetric, the professional readership is invited to provide their feedback and quantitative information to the author. The analysis results of the feedback will be published separately later.

Figures (4)

• Figure 1: Rough categorization of system's functions for driverless mainline rail traffic, which in comparison to driverless metros, require additional technological effort railwayvisonreview.
• Figure 2: Data required for two currently available approaches of safety argumentation for European mainline railway systems riskanal. The grey frame denotes the explicit risk assessment with resulting hourly fatality rates and the maximal values of harmonized design goals. The orange frame denotes the comparison with the a human train driver as reference system.
• Figure 3: Estimated consequences for a frontal collision of a train going at $130\km/h$ with a stationary passenger car depending on braking distance. The braking deceleration is set to be $1m/s^2$. The driver can hear the warning horn at a distance of $350m$ or less and may be able to escape. Negative distances mean the onward movement of an unbraked collided train. Warning horn and emergency braking start simultaneously. The solid kinked curve shows the number of seconds between hearing the warning horn by the car driver and the collision. The formula for this curve is added below the graph and provides an explanation for the kink. The dashed zigzag line depicts the size of the risk area at the collision site. For the sake of simplicity, it is assumed in this that the derailment risk in this example is only present in collisions at speeds of $130\km/h$ and above.
• Figure 4: Performance submetric for obstacle detection with results of two hypothetical systems A and B. $X$ can be replaced by a positive number up to $100$. A detection on contact with an obstacle and a non-detection are counted as the same.