Content uploaded by Ari Virtanen
Author content
All content in this area was uploaded by Ari Virtanen on Mar 27, 2020
Content may be subject to copyright.
13th ITS European Congress, Brainport, the Netherlands, 3-6 June 2019
Paper number ITS-TP1869
Railroad level crossings and an autonomous vehicle
Ari Virtanen*, Anne Silla, Maria Jokela & Kimmo Kauvo
Technical Research Centre of Finland VTT Ltd. Finland
ari.virtanen@vtt.fi
Abstract
Level crossing is an interface between road and rail environments. So far, very little attention is put on
investigating how an autonomous car can pass level crossings and especially the unprotected ones.
Today, the vast amount of level crossings are still unprotected. This paper discusses the problematics
of autonomous cars approaching and passing the level crossings, and focusses especially on the
behaviour of autonomous car, its information requirements and safety aspects related to both crossing
the protected and unprotected level crossings. Discussion covers the use of the vehicle-to-everything
(V2X) messaging, use of the autonomous car´s environment perception sensors and the use of a train
tracking solution to provide required data for the autonomous car to cross the railroad level crossing
safely. This paper proposes a solution on how the autonomous car could cross the LCs safely and
demonstrates this solution via simulations.
Keywords:
Autonomous vehicle, level crossing, connected,
1. Introduction
According to European Union Agency for Railroads (ERA, 2014), there are an average of five level
crossings (LC) per 10 kilometres of railway line in Europe, equating to over 118,000 level crossings
across the European Union, half of which are unprotected. In Finland in 2018 there were 3,058 level
crossings and 2,051 of them were without any warning or protection devices derived from open data
provided by The Finnish Transport Infrastructure Agency. (The Finnish Transport Infrastructure
Agency, 2018). Level crossings form an interface between road and rail networks. Therefore, the
safety management of level crossings is a shared responsibility between road and rail operators.
There is a continuous trend to actively decrease the amount of level crossings. According to the
Finnish Transport Infrastructure Agency, approximately 20−25 level crossings are removed yearly.
Building a bridge or an underpass costs up to 2 M€ and adding a protection devices up to 0.25 M€.
When comparing these numbers to total amount of level crossings, it is clear that level crossings will
exist to far future (The Finnish Transport Infrastructure Agency, 2018). Therefore, it is clear that also
autonomous cars need to pass both protected and unprotected level crossings.
Railroad level crossings and an autonomous vehicle
2
The objective of this paper was to discuss the problematics of autonomous cars approaching and
passing LCs. The discussion focussed especially on the behaviour of autonomous car, its information
requirements and safety aspects related to crossing both protected and unprotected level crossings.
After the discussion, this paper proposed a solution on how the autonomous car could cross the LCs
safely. At the end, this proposed solution was demonstrated via simulations.
SAE J3016 defines six levels (0−5) of vehicle automation (SAE 2018). In this paper it is assumed that
the automation level of the autonomous car is 4 or above, which means that the vehicle can drive
without a driver intervention and there can be passengers or the vehicle can also be completely empty.
If the vehicle has passengers inside one interesting question is how much information vehicle should
provide to its passengers when approaching the LC. Even though this topic is interesting it is excluded
from this study.
2. Discussion
2.1. Safe passing of level crossings
Level crossing is usually marked at least with Saint Andrews Cross (see Figure 1). Other protective
devices are warning sounds, barriers and/or traffic lights. Nevertheless, even though level crossings
have similarities among countries, the markings are not standardised worldwide.
Saint Andrew’s Cross
Traffic lights
and barrier
No stop zone
Virtual stop line
Figure 1. Level crossing
According to the Finnish traffic regulations (author’s translation):
• Train always has a priority in level crossings.
• Road user should follow specific care and despite of any protective devices observe if any
trains are approaching the level crossing.
• Vehicle speed should be such that it is possible to stop the vehicle before the tracks.
• It is not allowed to pass the level crossing if any train is approaching, traffic lights oblige to
stop, warning sound is heard or barrier is down or moving. One must stop at safe distance
Railroad level crossings and an autonomous vehicle
3
from tracks, before barrier or semaphore.
• When passing of level crossing is allowed it should be done without hesitation.
• If instructions above are followed, there are two virtual stop lines and stopping between them
is not allowed (no stop zone).
These kind of regulations are common in road traffic. Humans understand them easily but an
autonomous car is a machine. Machines are good at measuring things, therefore seconds and metres
are more practical for them. Thus, a term “safe distance” is meaningless; instead, one should give an
exact value, e.g. 10 metres before the level crossing. Another example is adjusting the vehicle speed to
such that it is always possible to stop the car before the tracks. The stopping distance depends on
multiple factors such as the amount of friction, type of road surface and inclination which all have an
effect on the braking distances. Of course these parameters can be measured but again, human is
superior compared to a machine on estimating these factors in advance.
2.2. Protected level crossings
Protected level crossing has barriers and traffic lights. The barrier type can vary from half barriers to
full barriers. It can be assumed that protective devices can be detected with environment perception
sensors of the autonomous car. Common to all level crossings is that the car is not allowed stop on
tracks (see Figure 2). Level crossing status is either “open” or “closed”, because signalling system is
electro-mechanical. When a train arrives to trigger point (1.2 km before the level crossing), level
crossing status changes to “closed”, traffic lights change to red and barriers start to close.
Figure 2. Protected level crossing (The Finnish Transport Agency, 2012).
Differences to road intersection with traffic lights is the absence of yellow light that informs when the
Railroad level crossings and an autonomous vehicle
4
status of traffic lights is going to change. In addition, in level crossing lights the green light is replaced
with blinking white light. The third difference is that in level crossing car always drives straight over,
and there are no right or left turns. After passing the barrier before the level crossing car enters “no
stop” zone. Special care must be taken so that the autonomous car does not misinterpret the closing
barrier after the level crossing as a stop command because otherwise the car will stop on the tracks.
2.3. C-ITS messages
Because protective devices require power, it is possible to install ITS-G5 DSRC radio roadside unit
(RSU) to the level crossing. Messages that can be used are:
·The Decentralized Environmental Notification Message (DENM) (EN 302 637-3.) Its main
purpose is to notify the road users for potentially dangerous road events.
·The Cooperative Awareness Message (CAM) (EN 302 637-2) is for the exchange of
information between road users and roadside infrastructure, providing each other's position,
dynamics and attributes. Road users may be cars, trucks, motorcycles, bicycles or even
pedestrians while roadside infrastructure equipment includes road signs, traffic lights or
barriers and gates.
·MAP (SAE J2735) - topological definition of lanes within an intersection, links between
segments, lane types and restrictions.
·SPaT (SAE J2735) - Traffic light signal phase and timing information and the status of traffic
controller. Prediction of duration and phases.
Relevance of each of the above options:
·DENM message can be used to inform the autonomous car about the presence of LC and
where the dangerous location starts and ends but because it does not contain information about
LC status, its usefulness for the autonomous car is limited. Autonomous cars use pre-planned
route and a routing algorithm can include LC data to the route. Thus DENM contain a
redundant information.
·CAM messaging from the approaching train could provide train location information to the
autonomous cars. The main problem is the communication range of the ITS-G5, which is only
few hundred meters (Gozalves et al. 2012). Therefore, a relay station is needed to achieve the
required communication range. The second challenge is the reliability of the communication.
The communication is not fail safe since one does not know if missing message means that no
train is not approaching or no message is transmitted.
·SPaT message could send information on LC status to the autonomous car and thus support
sensor based recognition. However, since the state of LC is either “open” or “closed”, LC
protection system cannot produce “time to green” or “remaining green time” values.
·MAP is very useful and contains required features to describe LC geometry precisely.
As a conclusion, using RSU that sends DENM, SPaT and MAP messages together with sensor based
recognition provides enough data for safe passing for the level 4 or level 5 autonomous cars. In case of
Railroad level crossings and an autonomous vehicle
5
level 2 autonomous cars the control can be passed to the driver who performs the crossing of LC
manually.
2.4. Passive level crossings
Passive level crossing can be considered as a challenging point to the autonomous car. Often passive
level crossings are only equipped with Saint Andreas Cross. Since the locations of level crossings are
known, an autonomous car can be aware of them via route planning.
2.5. Detection with sensors
Road design regulations related to level crossings set some requirements for the visibility of LC from
road to track. Visibility area definition is shown in Figure 3. The visibility should be measured from
1.1 metres above ground. Required minimum visibility depends on speed limit of the track (Table 1).
Maximum speed at railway lines with level crossings is 140 km/h in Finland, and therefore the
minimum detection range of 840 metres is required.
Figure 3. Visibility area definition (The Finnish Transport Agency , 2012).
Table 1. Minimum visibility requirements in Finland (The Finnish Transport Agency, 2012).
Distance from the track LS [m]
8
Train speed [km/h] LR[m]
40 240
60 360
80 480
100 600
120 720
140 840
Commercial lidar sensors used in autonomous cars have a typical detection range of 120 meters
(Velodyne, 2019). Range is limited due technical factors such as power requirements and target
Railroad level crossings and an autonomous vehicle
6
reflectivity. Infrared beam intensity is limited also by eye safety regulations. Therefore, currently
available automotive lidar cannot detect the train early enough. Sensors that have the capability to
detect objects up to 1 km are cameras and radars. Radars aimed to adaptive cruise control (ACC)
applications have detection range up to 250 metres and have a narrow beam, typically ±6-9° (Bosch,
2019). Radar technology has less limiting factors to increase the detection range.
Narrow field-of-view (FOV) makes detection along the track challenging. Using car heading and track
heading one can calculate an estimation of the direction angle where the train might be approaching,
but it might be out of sight because of the narrow FOV. One factor is the installation height of the
sensor, which is limited in small vehicles. Vegetation or snow banks might limit the visibility if the
sensor is installed too low. According to Kutila et al. (2018) adverse weather conditions like rain,
snow and fog can dramatically limit the available detection range. Therefore, relying only on sensors
do not produce reliable enough train detection data to ensure safe passing of unprotected LC.
3. Proposed solution
Proposed solution is based on train tracking and delivery of arrival estimates to the LC. Access to data
has to be via the cellular network because the level crossings may be located far from the power
sources. Proposed solution is very cost effective because it does not require any installations to the LC
itself. Development of the traffic management systems is going to such direction where train
positioning and communication with it is a standard feature.
4. Simulations
Train traffic used in the simulation was recorded during the Junavaro project (Öörni, 2011). Simulated
train use recorded GNSS data (Figure 4). Simulator outputs estimated the arrival time of train to each
level crossing along its route. Estimation is not restricted to level crossings only: one can add arbitrary
geolocation to the list and estimation can be calculated to it as well. LC can be closed either fixed
distance or fixed time before the train arrives to LC. Special attention has been put to communication
latencies. Junavaro system can provide accurate estimation of the train arrival time to LC despite
varying communication latencies and positioning accuracy (Virtanen, 2015). It should be mentioned
that GNSS positioning alone is not reliable enough for SIL4 (safety integrity level) applications.
Railroad level crossings and an autonomous vehicle
7
Figure 4. Simulation system principle.
System is very flexible for information dissemination. The estimated train arrival time and the state of
the level crossing can be delivered via information displays, mobile phones, ITS-G5 DSRC radio
connection or using LTE connectivity. A client contacts to service interface and request data for
specific LC. In this study ITS-G5 unit is used to simulate protected level crossing RSU and LTE
connection is used for data requests at unprotected level crossing. The system is very cost effective,
because it does not need any additional installations to the LC.
As mentioned earlier, SPaT and MAP messages is the best choice for autonomous cars. By using SPaT
messages one can provide compatibility to road intersections data messaging. Therefore, the Junavaro
system was modified to provide necessary phase and time information. The principle of the SPaT
calculation is shown in Figure 5.
GNSS
Level crossing
Location
Remaining open timeTime to open
LC openLC closed
Train length
Trigger point
Figure 5. Principle of the SPaT information determination.
“Remaining open time” is the time what takes the train to arrive to the trigger point where the LC is
set to “closed” state. “Time to open” is the time from the trigger point to the moment where the last
railway car passes the LC. Thus the train length needs to be known. Train composition is known in the
backend systems, therefore this value is also known. In the case where multiple trains passes the level
crossing at the almost the same time, SPaT messaging integrate all trains behind one message. The
Railroad level crossings and an autonomous vehicle
8
Junavaro system can show arrival times and directions for all trains approaching the LC.
Figure 6. Example of the SPaT data produced by Junavaro.
In Figure 6 shows an example of the SPaT data from LC Kirkkotie at Rail section 142 in Finland
(59,864443°, 23,075697°, WGS84) produced by Junavaro after modification. Data is produced using
distance based trigger point 1.2 km before the LC. If the adaptive closing scheme is used, remaining
open time is “arrival time to LC” minus 30 seconds. Remaining open time is set to infinity value after
the train has passed the LC. Red line represent LC states, open and closed. Study shows that it is
possible to produce equivalent SPaT messages to railroad level crossing as road intersections.
An autonomous car behaviour is shown in Figure 7. User at first sets the destination. Route planning
module at then produces a coarse route plan. The next step is to search static events along the route
such as zebra crossings, intersections, bus stops and level crossings and divide the route to the sections.
Then a set of pre-defined behavioural rules are adapted to each section. During the execution of the
route plan, a trajectory planner continuously create a new trajectory to the car. In the case of the LC, it
send requests using LC identification (LD ID) to server and receive SPaT messages as a response.
After receiving the message, it checks its relevance and ignore irrelevant messages. Then it estimates
car arrival time to the LC. A virtual obstacle is set, if the analysis show that LC is closed when the car
arrives to the LC. After the LC is opened again, the virtual obstacle is removed and the car continues
the journey following the behavioural rules set for LC until it leave the section.
Railroad level crossings and an autonomous vehicle
9
Figure 7. Autonomous car behavior when approaching the LC.
5. Conclusions
Level crossing is an interface between road and rail environments. If an autonomous car needs to pass
the LC and its automation level is less than 4 (SAE J3016), control of the car can be passed to the
driver, which then manually operate the car over the LC. If the automation level is 4 or above this is
not possible, because the driver is not present.
An autonomous car follows the planned route and therefore the presence of the LC along the route is
known. Protected LC equipment can be detected by using environment perception sensors. Protected
LC has also electrical power available, therefore the use of road side unit sending C-ITS messages is
possible. Useful C-ITS messages are SPaT and MAP, where former provide status information and the
latter precise geographic presentation of the LC environment.
Today, the vast amount of level crossings are still unprotected. Train speeds can be up to 140 km/h,
which require visibility range of 840 metres to both directions. Today the detection range of the
sensors used in autonomous cars today reach up to 250 metres in a good conditions. Adverse weather
conditions dramatically limit the available detection range. Therefore the use environment perception
sensors alone cannot guarantee the safe passing.
Proposed solution is based on train tracking and delivery of the arrival estimates to the LC. Our
Junavaro-system was modified to produce SPaT message content. Simulation show that it can
precisely produce “remaining open time” and “time to open” values. Mobile broadband connection
can be used for delivery of the SPaT messages. Therefore proposed solution does not need any
installations to the LC and therefore it is very cost effective.
An autonomous car behaviour is based on setting a virtual obstacle before the LC. An algorithm
continuously estimates car arrival time to LC. If the LC is expected to be closed when the car arrives,
a virtual obstacle is set true and cleared when the LC is open.
Acknowledgement:
This study was funded by The Safer-LC project. The Safer-LC project has received
funding from the European Union's Horizon 2020 research and innovation
programme under grant agreement No 72320
Railroad level crossings and an autonomous vehicle
10
References:
1. ERA (2014). Railway safety performance in the European Union. European Railway Agency.
http://www.era.europa.eu/Document-Register/Pages/Railway-Safety-Performance-Rep-2014.aspx
2. ETSI - EN 302 637-3 Intelligent Transport System (ITS); Vehicular Communications; Basic Set of
Applications; Part 3: Specification of Decentralized Environmental Notification Basic Service
3. ETSI - EN 302 637-2 Intelligent Transport Systems (ITS); Vehicular Communications; Basic Set of
Applications; Part 2: Specification of Cooperative Awareness Basic Service
4. SAE J2735. Dedicated Short Range Communications (DSRC) Message Set Dictionary
J2735_200911
5. Finnish Transport Agency. (2012). Tien suunnittelu tasoristeyksessä (Road planning guidelines at
level crossing). Liikenneviraston ohjeita 3.12. ISBN 978-952-255-110-8 30 p. (in Finnish)
6. Finnish Transport Infrastructure Agency (2018) Open Data.
https://vayla.fi/web/en/open-data#.XLmHEaRS9PY
7. Finnish Transport Infrastructure Agency. (2018). Removal of the level crossings (in Finnish)
https://vayla.fi/rataverkko/tasoristeykset/tasoristeysten-poistaminen#.XED9zWlS-Uk
8. Velodyne HDL-64E brochure. https://velodynelidar.com/hdl-64e.html
9. Gozalvez, J., Sepulcre, M. and Bauza, R. (2012). IEEE 802.11p vehicle to infrastructure
communications in urban environments, in IEEE Communications Magazine, vol. 50, no. 5, pp.
176−183, May 2012.
10. Öörni, R.,Hietikko, M., Kauvo, K., Lattunen, A. & Virtanen, A. (2011). Autossa toimiva
junavaroitusjärjestelmä..Kokeilu Hanko–Hyvinkää-rataosalla [In-Vehicle Warning System for
Railway Level Crossings. Pilot Tests on Hanko–Hyvinkää Railway Line]. Espoo 2011. VTT
Tiedotteita – Research Notes 2603. 106 s. + liitt. 30 s. (in Finnish)
11. Bosch (2019) Bosch LLR4 Technical data sheet.
https://www.bosch-mobility-solutions.com/en/products-and-services/passenger-cars-and-light-com
mercial-vehicles/driver-assistance-systems/left-turn-assist/long-range-radar-sensor
12. Virtanen, A., Kostiainen, J., Simola, N., & Iqbal, O. (2015). Improving level crossing safety with
satellite technology. In 22nd ITS World Congress ERTICO - ITS Europe.
13. Kutila, M., Pyykönen, P., Holzhuter, H., Colomb, M. & Duthon, P. (2018). Automotive LiDAR
performance verification in fog and rain. 1695-1701. 10.1109/ITSC.2018.8569624.
14. SAE J 3016-2018 Taxonomy and Definitions for Terms Related to Driving Automation Systems
for On-Road Motor Vehicles.
https://webstore.ansi.org/Standards/SAE/SAE30162018?source=blog
