Extensive R&D resources have been directed to improve current electrified railway operations on the EU level and beyond, so that the future solutions would be safer and more efficient. To achieve the efficiency and safety related objectives, the level of autonomy needs to be raised to reduce the possibility of human error in some critical operations and to increase the maximum usage of the rolling stock. To do that, you need to be able to localize the train, in any conditions, and to understand what is happening around, and especially, in front of it. One could say that those two functionalities are absolutely crucial when the railways, as a business, moves constantly towards more eco-efficient way of working.
Autonomous railway-based systems have been existing already for some time, especially in closed environments, which have been controlled by external infrastructure-based solutions. These include for example metros and airport shuttles. The first real world implementations of autonomous trains are reaching large-scale piloting stages and initial results have been promising. There are some impressive projects running currently for example in Russia and Germany and new projects are announced almost weekly all over the world – also in Finland, where Proxion together with VTT are going towards autonomous train.
The topic has been actively researched also in Finland for many years. GIM Robotics has been working with railway based systems publicly since 2019, when VR (the state owned national railway company in Finland) selected GIM Robotics to pilot modern technology solutions for improving railway safety issues. After that project, GIM Robotics has participated in a large national Digirail project aiming “Towards Digital and Intelligent Rail Transport”. We have been part of a large consortium basically turning a traditional locomotive into a kind of autonomous mobile robot, which is obviously bound to tracks and still fully controlled by an experienced driver/operator. It has been clear from the beginning, that there are certain areas, which are causing extra challenges for the development crew. Those can be divided into two main groups: areas where the localization of the vehicle is extraordinary difficult and areas where there are a lot of interactions between the vehicle and humans. To achieve the long-term goals, the machine must be able to localize itself in real-time and it must be able to perceive the environment in all-weather conditions. Read more about Digirail testing phases from here.
The current train localization system is often relying on identifiable electronic beacons or transponder markers, called balise in this context, placed between the rails. They are an elementary part of the Automatic Train Protection (ATP) system and have a big role also in the European Train Control System (ETCS), because they provide the exact position information every time a train passes a particular balise. If all routes are fully equipped with balises, the system provides the backbone for the ETCS. However, one must remember that any electrical system, be it passive or active, will have occasional problems. Even though some safety features have been integrated to the balise system, to make sure that the ETCS knows all the time near real-time positions of every train, some additional supportive systems are required.
These systems include Global Navigation Satellite System (GNSS) based solutions, which unfortunately have their well-known shortcomings, mainly related to some challenging (e.g., forestry, downtown and mountains) or even denied (e.g., tunnels) areas. GNSS system’s vulnerability is further highlighted when somebody (individual or state-level operator) intentionally disturbs satellite signals through spoofing and jamming actions. The more turbulent the world is, the more frequent that type of activity will become. Additionally, some traditional wheel rotation measurement sensors (odometry information) are used to calculate the traveled distances. This approach is neither without problems, because wet or snowy conditions on steep slopes can result in occasional slipping, which will create errors for the distance estimations. To secure the full localization coverage, one additional system should be included. This localization system can be based on a detailed constantly updated feature-based map of the working environment, sensory system capable of detecting those selected features in all conditions (including various point cloud producing sensors like LiDARs, radars or depth-cameras), and the algorithms to produce probabilistic real-time 6D position estimates.
Our 3D localisation (GIM–Locator) and environmental modelling (GIM-Mapper) software systems provide the foundations for our situational awareness software (GIM-Observer). Situational awareness means here that the machine should not only monitor its internal state but also sense, perceive, classify and model what is currently happening around it and even predict what will most likely happen in the near future. However, the most important single objective at the moment, is to provide reliable obstacle detection capabilities for all rail operations on their continuous journey towards the development of autonomous rail systems. The train operator must be confident that the train, even with a rather low level of autonomy, will not collide with anything and it is still capable of performing the given everyday tasks.
If one wants to increase the level of autonomy in railway operations and operate those machines mostly in a complex urban environment, the requirements are exceptionally tight. One must consider humans, who can cross tracks randomly wherever they want. To be able to do that, autonomous or even semi-autonomous railway operations in the urban areas require highly sophisticated sensory systems for continuous environmental monitoring. Usually, at least different cameras, LiDAR and radar are included in the sensor suite for real-time assessment of the situation to provide the machine enough time to react in all weather conditions. The localization integration to the obstacle detection development is crucial for the long-term objectives when moving towards human supervised automatic functions and finally to autonomous operations.
We have started a new light railway project with Škoda Group, which puts our expertise in a real test in dynamic urban areas. You can read more about that project from here.