This is one of the flagship projects of the future of rail
transport. But running an autonomous train is more complex than
automating a metro train. Because the requirements for safety are very
high.
In the Lille metro, it has been a long time since we can watch the rails
pass between two stations, since there is no driver’s cab anymore.
These automatic trains, deployed in 1983 by Keolis, are piloted by a
control center. Operators simply optimize the frequency of passage and
the number of trains on the tracks, according to the flow of passengers.
When is such independence for trains? Not immediately, if we compare
these two modes of transport and if we consider the complexity of
driving a train without a driver.
Unlike metro trains, trains run on an open circuit.
Impossible, therefore, to implement, as on lines 1 and 14 of the Paris
metro, doors securing the entrances and exits of travelers. But these
play a major role in the proper functioning of the service, especially
during peak hours, said Edgar Sée, the director of automation of line 4
of the RATP. “They allow a better regulation of the line and the parking
time of the train. In Lyon, Line D has the highest level of security,
but no doors. As soon as an object is detected on the channel by its
sensors, the traffic stops. “
In fact, the driver of a train monitors his environment. “He observes
the catenaries to check their condition, keep an eye on the tracks,
there watching for the possible presence of passengers …,” says Luc
Laroche, the director of the train autonomous project at the SNCF. On
autonomous trains, it will automate this skill to integrate. And so
equip the train with sensors and teach it to use it.
The challenge is not thin. On average, a train takes between 3 and 4 km
to stop. And at a speed of around 300 km / h, this means that these
sensors must be efficient. “The autonomous train needs cameras and
radars with a large detection distance,” warns Alberto Parrondo,
Thales’s director of strategy for land transport activities. “For
example, you need a camera that can recognize signals over 2 km away to
see if the train has to brake or accelerate. To achieve this, we try to
capitalize on technologies already known, especially from the world of
defense, seeking to lower costs, “says the expert.
No right to the error
Another source of inspiration: the automobile. As part of the autonomous terrestrial transport project in safety in its environment (TAS), the IRT SystemX, Alstom and the Compiègne University of Technology are working with SNCF on the automation of the observation functions of a driver. These include recognizing signs, pedestrians on the platforms, objects on the tracks or anomalies on a train that we cross. The partners of this project use sensors (radars, lidars, cameras) from autonomous cars and combined with artificial intelligence. “We are testing the efficiency and the limits of these sensors in rail applications,” says Raphael Chosidow, project manager at SNCF.
Currently, the performance of these recognition systems is between 80 and 90%. But in the rail, security levels are very high and the system will not be allowed to make mistakes. “The recognition must therefore reach a validation rate of 100% – which is not the case at the moment – or be associated with the location of the train and mapping. Part of the data already available deserves to be corrected. If all the traffic signs on the tracks are listed, their position is not always accurate in the censuses available to the network operators. “We have therefore processed our own data for extremely precise positioning. ” security levels are very high and the system will not be allowed to make mistakes.
“The recognition must therefore reach a validation rate of 100% – which is not the case at the moment – or be associated with the location of the train and mapping. Part of the data already available deserves to be corrected. If all the traffic signs on the tracks are listed, their position is not always accurate in the censuses available to the network operators.
“We have therefore processed our own data for extremely precise positioning. ” security levels are very high and the system will not be allowed to make mistakes.
“The recognition must therefore reach a validation rate of 100% – which is not the case at the moment – or be associated with the location of the train and mapping. Part of the data already available deserves to be corrected. If all the traffic signs on the tracks are listed, their position is not always accurate in the censuses available to the network operators.
“We have therefore processed our own data for extremely precise positioning. ” or be associated with train location and mapping. Part of the data already available deserves to be corrected. If all the traffic signs on the tracks are listed, their position is not always accurate in the censuses available to the network operators. “We have therefore processed our own data for extremely precise positioning. ” or be associated with train location and mapping.
Part of the data already available deserves to be
corrected. If all the traffic signs on the tracks are listed, their
position is not always accurate in the censuses available to the network
operators. “We have therefore processed our own data for extremely
precise positioning. “
Eventually, the removal of all signage
With these elements, engineers then modeled the train and its environment in the laboratory to perform a test of embedded technologies, before launching tests in real conditions. “A VB 60 000 freight locomotive runs tests at 100 km / h twice a month on RER D and Transilien R test lines in Villeneuve-Saint-Georges and Montereau. We chose this locomotive because it has a corridor that allows to install sensors, “says Raphael Chosidow.
Working on an existing train is not trivial: it is indeed
on the trains in circulation that the partners hope to eventually deploy
their solutions. A retrofit approach. “We will have to equip trains
that have been running for ten to fifteen years with different
technologies,” insists Raphaël Chosidow.
Beyond the increase in detection capabilities of the train, the other
challenge is related to its degree of autonomy vis-à-vis its
environment. Driverless metros are automatic … but not autonomous.
“They receive a mission and a movement authorization from the ground
computer that oversees the movement of several trains,” says Edgar Sée.
They then manage to go from point A to point B, respecting
the limit of movement allowed. To achieve this, they need to know where
they are on the line, including an odometer, which indicates the
distance traveled, and thanks to the tags placed on the ground, for a
precise location. “
Communication between the control center and the train is continuous via
the automatic control system CBTC (communication based train control):
the train sends its position, the ground of the movement indications.
“The train communicates information to a control center that will tell it if it can go further, faster … The calculations that the automatic train carries out remain dependent on its pre-planned environment,” explains Alberto Parrondo. This type of permanent communication between the ground and the train is gradually being deployed on the rail lines, via GSM-R, with the ERTMS level 2 European signaling system [see page 38].
But the autonomous train should go one step further: as
part of the European ERTMS level 3 system, it is a suppression of all
signaling that is targeted.
In parallel, the experts are looking to lift another lock, which also
mobilizes autonomous car designers: that of demonstrable artificial
intelligence. There is no question of being satisfied with a system that
works, having found only a solution by machine learning. You also need
to know why and how it works.
“We can not let algorithms based on unprovable technologies take over responsibility, without verifying that they are safe,” said Alberto Parrondo.
Given the list of challenges to be met, there is no doubt
that it will be necessary to count on the seasoned eye of the train
drivers for a few more years …?
*This article was first published by lerail.com
