The development of unmanned technologies on the railway began quite a long time ago, already in 1957, when the first experimental autopilot complex for suburban trains was created. To understand the difference between the levels of automation for railway transport, a gradation is introduced, defined in the IEC-62290-1 standard. Unlike road transport, rail transport has 4 degrees of automation, shown in Figure 1.
Figure 1. Degrees of automation according to IEC-62290
Almost all trains operating on the Russian Railways network are equipped with a safety device corresponding to automation level 1. Trains with automation level 2 have been successfully operated on the Russian railway network for more than 20 years, several thousand locomotives are equipped. This level is implemented by traction control and braking algorithms for energy-optimal train guidance along a given route, taking into account the schedule and indications of automatic locomotive signaling systems received via an inductive channel from track circuits. The use of level 2 reduces the fatigue of the driver and gives a gain in energy consumption and accuracy in the execution of the traffic schedule.
Level 3 assumes the possible absence of the driver in the cab, which requires the implementation of a vision system.
Level 4 implies the complete absence of a driver on board, which requires a significant change in the design of the locomotive (electric train). For example, automatic switches are installed on board, which will not be possible to cock again if they are triggered without the presence of a person on board.
Currently, projects to achieve levels 3 and 4 are being implemented by the world's leading companies, such as Siemens, Alstom, Thales, SNCF, SBB and others.
Siemens presented its project in the field of unmanned trams in September 2018 at the Innotrans exhibition. This tram has been in operation in Potsdam with GoA3 automation level since 2018.
Figure 2 Siemens tram
In 2019, Siemens more than doubled the length of its unmanned route.
Russian Railways is one of the first companies in the world to start developing unmanned railway vehicles. Thus, in 2015, at the Luzhskaya station, a project was launched to automate the movement of 3 shunting locomotives, where NIIAS JSC acted as the project integrator and developer of basic technologies.
The creation of an unmanned locomotive is a complex complex process that is impossible without cooperation with other companies. Therefore, at the Luzhskaya station, together with JSC NIIAS, such companies participate as:
- JSC "VNIKTI" in terms of the development of the onboard control system;
- Siemens - in terms of automating the operation of the marshalling yard (MSR-32 system) and automating the operation of pushing cars;
- JSC "Radioavionika" in terms of microprocessor interlocking systems that control arrows, traffic lights;
- PKB TsT - creation of a simulator;
- Russian Railways as project coordinator.
At the first stage, the task was to achieve level 2 of traffic automation, when the driver, under normal conditions for organizing shunting work, does not use the locomotive controls.
During the operation of conventional shunting locomotives, traffic control is carried out by transmitting voice commands from the dispatcher to the driver with setting the appropriate routes (turning arrows, turning on traffic lights).
When moving to level 2 of automation, all voice communication was replaced by a system of commands transmitted over a digital secure radio channel. Technically, the management of shunting locomotives at the Luzhskaya station was built on the basis of:
- unified digital station model;
- protocol for controlling the movement of shunting locomotives (for sending commands and monitoring their execution);
- interaction with the electrical interlocking system to obtain information about the specified routes, the position of the arrows and signals;
- positioning systems for shunting locomotives;
- reliable digital radio.
By 2017, 3 TEM-7A shunting locomotives operated 95% of the time at the Luzhskaya station in a fully automatic mode, performing the following operations:
- Automatic movement along a given route;
- Automatic access to wagons;
- Automatic coupling with wagons;
- Pushing wagons onto a marshalling yard.
In 2017, a project was launched to create a vision system for shunting locomotives and introduce remote control in case of emergency.
In November 2017, JSC NIIAS specialists installed the first prototype of a vision system for shunting locomotives, consisting of radars, lidar and cameras (Figure 3).
Figure 3 First versions of vision systems
During the tests at the station of the Luga vision system in 2017-2018, the following conclusions were drawn:
- The use of radars for detecting obstacles is impractical, since the railway has a significant number of metal objects with good reflectivity. The detection range of people against their background does not exceed 60-70 meters, in addition, radars have insufficient angular resolution and is about 1 Β°. Our findings were subsequently confirmed by the test results of colleagues from SNCF (French railway operator).
- Lidars give very good results with minimal noise. In the case of snowfall, rain, fog, there is a non-critical decrease in the detection range of objects. However, in 2017, lidars were quite expensive, which significantly affected the economic performance of the project.
- Cameras are an indispensable element of the technical vision system and are necessary for the tasks of detection, object classification, and remote control. For operation at night and difficult weather conditions, it is necessary to have infrared cameras or cameras with an extended wavelength range capable of operating in the near infrared range.
The main task of technical vision is to detect obstacles and other objects in the direction of travel, and since the movement is carried out along the track, it is necessary to detect it.
Figure 4. An example of multi-class segmentation (track, wagons) and determination of the track axis using a binary mask
Figure 4 shows an example of track detection. In order to unambiguously determine the route of movement along the arrows, a priori information is used about the position of the arrow, the readings of traffic lights, transmitted via a digital radio channel from the electrical interlocking system. At the moment, there is a trend on the world's railways to abandon traffic lights and switch to control systems via a digital radio channel. This is especially true for high-speed traffic, since at speeds of more than 200 km / h it becomes difficult to notice and recognize the indications of traffic lights. In Russia, there are two sections operated without the use of traffic lights - this is the Moscow Central Ring and the Alpika-Service - Adler line.
In winter, situations may arise when the track is completely covered with snow and the recognition of the track becomes almost impossible, as shown in Figure 5.
Figure 5 Example of a track covered with snow
In this case, it becomes unclear whether the detected objects interfere with the movement of the locomotive, that is, whether they are on the way or not. At Luzhskaya station, in this case, a high-precision digital model of the station and a high-precision onboard navigation system are used.
Moreover, the digital model of the station was created on the basis of geodetic measurements of base points. Then, based on the processing of many passages of locomotives with a high-precision positioning system, a map was completed along all the tracks.
Figure 6 Digital model of track development of Luzhskoy station
One of the most important parameters for the onboard positioning system is the error in calculating the orientation (azimuth) of the locomotive. The orientation of the locomotive is necessary for the correct orientation of the sensors and objects detected by them. With an orientation angle error of 1Β°, the error of the object's coordinates relative to the path axis at a distance of 100 meters will be 1,7 meters.
Figure 7 Influence of the orientation error on the transverse coordinate error
Therefore, the maximum allowable error in measuring the orientation of the locomotive in terms of angle should not exceed 0,1Β°. The onboard positioning system itself consists of two dual-frequency navigation receivers in RTK mode, the antennas of which are spaced along the entire length of the locomotive to create a long base, strapdown inertial navigation system and connection to wheel sensors (odometers). The standard deviation of determining the coordinates of the shunting locomotive is no more than 5 cm.
Additionally, studies were conducted at Luzhskaya station on the use of SLAM technologies (lidar and visual) to obtain additional position data.
As a result, the determination of the railway gauge for shunting locomotives at the Luzhskaya station is carried out by combining the results of gauge recognition and digital track model data based on positioning.
Obstacle detection is also carried out in several ways based on:
- lidar data;
- stereo vision data;
- work of neural networks.
One of the main sources of data are lidars, which produce a cloud of points from laser scanning. In the algorithms that are in operation, classical data clustering algorithms are mainly used. As part of the research, the effectiveness of using neural networks for the task of clustering lidar points, as well as for joint processing of lidar data and data from video cameras, is checked. Figure 8 shows an example of lidar data (a cloud of points with different reflectivity) showing a human dummy against the background of a carriage at the Luzhskaya station.
Figure 8. Example of data from lidar at Luzhskaya station
Figure 9 shows an example of extracting a cluster from a car with a complex shape according to the data of two different lidars.
Figure 9. An example of lidar data interpretation as a cluster from a hopper car
Separately, it is worth noting that recently the cost of lidars has fallen by almost an order of magnitude, and their technical characteristics have grown. There is no doubt that this trend will continue. The detection range of objects by lidars used at the Luzhskaya station is about 150 meters.
A stereo camera using a different physical principle is also used to detect obstacles.
Figure 10. Disparity map from a stereopair and detected clusters
Figure 10 shows an example of stereo camera data with the detection of poles, wayboxes and a wagon.
In order to obtain sufficient accuracy of the point cloud at a distance sufficient for braking, it is necessary to use high-resolution cameras. Increasing the image size increases the computational cost of obtaining the disparity map. Due to the necessary conditions for the resources occupied and the system response time, it is necessary to constantly develop and test algorithms and approaches for extracting useful data from video cameras.
Part of the testing and verification of the algorithms is carried out using a railway simulator, which is being developed by Design Bureau TsT together with JSC NIIAS. For example, Figure 11 shows the use of a simulator to test the operation of stereo camera algorithms.
Figure 11. A, B - left and right frames from the simulator; B β top view of the reconstruction of data from a stereo camera; D - reconstruction of stereo camera images from the simulator.
The main task of neural networks is the detection of people, wagons and their classification.
To work in severe weather conditions, JSC NIIAS specialists also carried out tests using infrared cameras.
Figure 12. Data from the IR camera
Data from all sensors are integrated based on association algorithms, where the probability of the existence of obstacles (objects) is estimated.
Moreover, not all objects on the way are obstacles; when performing shunting operations, the locomotive must automatically couple with the cars.
Figure 13. An example of visualization of the entrance to the car with the detection of obstacles by different sensors
When operating unmanned shunting locomotives, it is extremely important to quickly understand what is happening with the equipment, in what condition it is. There are also situations when an animal, such as a dog, appears in front of the locomotive. On-board algorithms will automatically stop the locomotive, but what to do next if the dog does not get out of the way?
To control the situation on board and make decisions in case of emergency situations, a stationary remote control and control panel has been developed, designed to work with all unmanned locomotives at the station. At Luzhskaya station, it is located at the EC post.
Figure 14 Remote control and management
At the Luzhskoy station, the control panel shown in Figure 14 controls the operation of three shunting locomotives. If necessary, using this remote control, you can control one of the connected locomotives by transmitting information in real time (the delay is not more than 300 ms, taking into account data transmission over the radio channel).
Functional safety issues
The most important issue in the implementation of unmanned locomotives is the issue of functional safety, defined by the standards IEC 61508 "Functional safety of electrical, electronic, programmable electronic systems related to safety" (EN50126, EN50128, EN50129), GOST 33435-2015 "Control, monitoring and safety devices of the railway rolling stock".
Safety Integrity Level 4 (SIL4) is required to comply with the requirements for on-board safety devices.
To comply with the SIL-4 level, all existing locomotive safety devices are built according to the majority logic, where calculations are performed in parallel in two channels (or more) with a comparison of the results to make a decision.
The computing unit for processing data from sensors on unmanned shunting locomotives is also built according to a two-channel scheme with a comparison of the final result.
The use of vision sensors, work under various weather conditions and in different environments requires a new approach to the issue of proving the safety of unmanned vehicles.
In 2019, the ISO/PAS 21448 standard βRoad vehicles. Security of Specified Functions (SOTIF). One of the main principles of this standard is the scenario approach, which considers the behavior of the system in various circumstances. The total number of scenarios is infinity. The primary design goal is to minimize areas 2 and 3 representing known unsafe scenarios and unknown unsafe scenarios.
Figure 15 Script transformation as a result of development
As part of the application of this approach, JSC NIIAS specialists analyzed all emerging situations (scenarios) since the start of operation in 2017. Some of the situations that are difficult to meet in real operation are worked out using the PKB TsT simulator.
Regulatory Issues
Regulatory issues must also be addressed in order to truly move to fully automatic control without the presence of the driver in the cab of the locomotive.
At the moment, Russian Railways has approved a schedule for the implementation of work on regulatory support for the implementation of measures to introduce automatic control systems for railway rolling stock. One of the most important issues is the updating of the Regulations on the procedure for internal investigation and accounting of transport accidents that caused harm to the life or health of citizens not related to production in railway transport. In accordance with this plan, in 2021 a package of documents regulating the operation of unmanned railway vehicles should be developed and approved.
Afterword
At the moment, there are no analogues of unmanned shunting locomotives in the world, which are operated at the Luzhskaya station. Specialists from France (SNCF company), Germany, Holland (Prorail company), Belgium (Lineas company) got acquainted with the developed control system in 2018-2019 and are interested in implementing such systems. One of the main tasks of JSC NIIAS is to expand the functionality and replicate the created management system both on Russian railways and for foreign companies.
At the moment, Russian Railways is also leading a project to develop Lastochka unmanned electric trains. Figure 16 shows a demonstration of a prototype of the automatic control system for the ES2G Lastochka electric train in August 2019 within the framework. International Railway Salon of Space 1520 "PRO//Dvizhenie.Expo".
Figure 16. Demonstration of the operation of an unmanned electric train at the MCC
Creating an unmanned electric train is a much more difficult task due to high speeds, significant braking distances, and ensuring the safe boarding / disembarking of passengers at stopping points. At the moment, tests are being actively conducted at the MCC. A story about this project is planned to be published in the near future.
Source: habr.com