Artificial intelligence (AI) can make an important contribution to the mobilty revolution. As part of smart grid mobility, people will move from one place to another and transport goods will be moved in a resource-efficient way. Experts in platform learning systems describe in a recent report the safest, most flexible and economical way to get on the road, on the rails, on the water or in the air.
Learning systems offer great potential for meeting the challenges of today’s mobility. « Less congestion, accidents and emissions – AI can help us achieve these goals for the mobility of the future. To get closer to this stage, transport means and transport infrastructure must be intelligently networked, » explains Christoph Peylo, head of the Bosch Center for Artificial Intelligence and co-leader of the Mobility and Intelligent Systems working Group from the ‘Plattform Lernende Systeme’. The idea: sensors, cameras and intelligent infrastructures and platforms that collect, manage and share data as traffic data. More powerful machine learning (ML) techniques are used to process the collected data.
In the « Mobility and Intelligent Transport Systems » working group of the Plattform Lernende Systeme, representatives of science and industry discussed the possibilities and challenges of the learning system for each transport mode. For its first report, the working group has identified five areas of action that should be addressed by science, industry, politics and society to promote artificial intelligence-based mobility and intelligent transportation systems, sustainable and demand-driven: networking and system interaction, fleet availability and transport, infrastructure status, human-machine interaction (MMI) in the mobility space, safety in intelligent transport systems and social aspects.
For the members of this working group, learning systems are developed for all modes of transport, including the railways. Their applications range from unmanned delivery drones and network rail systems with autonomous trains, intelligent forms of mobility in public transport and carpooling, to autonomous cars. These sometimes competing developments are used in parallel in the same environment – for example in the existing road and rail network – and can have significant effects on our society. What does this report say about railways?
What exactly does the learning system mean?
Despite confusion about its definition and its business value, AI continues to find its way forward. Machine learning (ML) is a method of data analysis that automates the building of analytical models. These algorithm-based models are primarily built from statistical techniques and theoretical computer science and leverage large datasets to continuously learn and improve.
Learning systems and artificial intelligence (AI) have long been in the private and professional life: we stay in touch with each other at any time via a smartphone, we have access to a wide range of information in real time via intelligent support systems, which also help us at work, at home or to guide us (Google maps). But the key question is how to handle the huge amounts of data we leave behind.
Today, emerging huge amount of data with advanced algorithms become the interest of most AI research by using learning system to create patterns among the data. « We are able to make very good forecasts using huge databases, » explains Emma Frejinger, Associate Professor in the Department of Computer Science and Operations Research at the Université de Montréal. « It was not possible before.»
There are several types of learning system:
- By 2019, the majority of apprenticeship programs were still supervised, that is, they needed to be fed by developers annotated data. Supervised learning brings a prebuilt database to the system. Developers specify the value of information, for example, if they belong to category A or B. It is tedious and demanding work, which involves a significant risk of errors.
- Learning by reinforcement. The process by which AI improves from experience – learning – without recourse to human programming. Most often, artificial intelligence will have access to data and use it to learn.
- Unsupervised learning helps find patterns in data set without pre-existing labels. Two of the main methods used in unsupervised learning are principal component and cluster analysis.
Deep learning is the most advanced evolution of AI to date. He is part of a broader family of machine learning methods based on artificial neural networks. Learning can be supervised, semi-supervised or unsupervised. It requires a large amount of computer processing capacity as well as a large amount of tagged data to understand the task at hand. It can thus achieve the highest degree of accuracy in data.
At the technical level, these learnings can be greatly accelerated by the deployment of 5G, which is debated in several countries, as well as high-capacity servers.
Which mobility potential ?
Artificial intelligence already exists in various forms in the field of mobility: on the one hand with the assistance and automation systems that are already implemented in each transport system. And secondly through the use of AI for operations, planning and asset management. But we must go much so far.
The aim is to make multimodal mobility efficient and user-friendly, exploiting the enormous potential for optimization between each operator. This requires a much stronger coupling of each other’s systems, with all that this implies at the technical and regulatory level, in particulary at the level of data protection.
Large amounts of data
Mobility is certainly a sector that produces the most data. Volumes are increasing from year to year. When you turn on the « position » button on your smartphone, it records all your displacements, even your movements with a sports application, for example. All datas goes to servers and are analyzed. In many cities, we find, for example, the metro, the bus, the various taxi services, the bicycle and the shared automobile. But they are all related to each other since they all use the same routes and streets. Your smartphone records all the transports you take, the time you spend there and the frequency each week or each month. Hence the need to look at the situation in a global way. It is these services as a whole that we want to better understand and plan better. All modes of transport are linked. The system is however more complex. This is where the challenge of artificial intelligence lies. Consequently, AI methods can be presented as a smart solution for such complex systems that can’t be managed using traditional methods. Many researchers have demonstrated the advantages of AI in transport.
One of the big challenges is to avoid having to use too much annotated data (supervised learning). Systems will be able to improve their performance through continuous learning from data captured automatically (non-supervised learning), without the need for annotation. This more autonomous learning mode will require a quasi-permanent control of the knowledge of the machine. And it is very expensive. In the future, with data complexity in ITS (Intelligent Transport Systems), hence, deep learning techniques will be essential to find patterns to achieve a more connected transportation systems.
Learning systems makes it possible to technically implement high automation of new rail functions and make existing functions safer and more efficient. Existing functions can be improved, such as the anticipated perception of the environment, the coordination of traffic or the automatic monitoring of vehicle components, which relates to predictive maintenance. Learning systems can more accurately predict the maintenance of rail assets as the mass of data accumulates. We can measured the challenge to collect usable datas and the enormous quantity which to be analysed.
Systems based on unsupervised learning are implemented in network security: an automatic learning system detects abnormal behavior. For example, a section occupied by another train. Intelligent algorithms, in the form of artificial neural networks, can analyze traffic and develop more efficient traffic management systems, while guaranteeing the safety of the trains. This greatly interest to the railway sector, which must provide a very high degree of safety while improving its performance.
Knowledge of flows
The rapid development of intelligent transport systems has increased the need to propose advanced methods to Predict traffic information. These methods play an important role in the success of ITS subsystems such as advanced traveller information systems, advanced traffic management systems and advanced public transportation systems. Intelligent predictive systems are developed using historical data extracted from track sections occupancy and trains. Then, these data becomes an input to machine learning and AI algorithms for a real time predictions.
Traffic simulation studies, which are important when deciding whether to add a track or modify a station, can be enriched by learning systems. This is a considerable potential to control traffic flow with greater precision and to be able to manage traffic more efficiently. However, linking individual and collective data poses great challenges, both technically and legally. An intelligent compromise between individual and collective data is essential for the systems not to be delivered to itself. Learning systems can play a key role in this regard. The prerequisite is the choice of data editing algorithms that support their management and the decision making. They thus make it possible to control the complexity of key performance indicators (KPIs).
Another important application of learning systems in the field of rail transport is driverless automated driving. Automated trams, metros and automated trains would not only provide greater capacity on the rail network, but would also make it possible to adapt the transport system to the needs of users by optimizing public transport schedules and transport capacity. Highly automated or driverless driving in mixed rail traffic conditions or on routes with less stringent safety infrastructure requires information about vehicles first and foremost. The automated train must be able to detect its environment as a human driver does today. In practice this would amount to granting analytical capabilities close to an autonomous car, although the train is guided on rails. Currently, trains do not communicate with each other and are dependent on a traffic control center. It’s not 100% autonomy.
A special case of rail automation is the Driver Advisory Systems. These intelligent support systems, based on new algorithms, enable more efficient driving, more energy efficiency, which improves the network capacity, operational stability and system reliability. Intelligent automation would make operations more flexible – like road traffic – and rail transport could be served in a demand-driven way by reactivating closed routes with smaller, automated rail vehicles. In addition, modern AI methods could provide more reliable information about the availability of the transport fleet or the state of the infrastructure.
>>> To read: Autonomous trains, a brief review
Automation of rail transport must take into account the interaction of vehicles with the infrastructure and the traffic and safety control systems. Current regulations, which govern safety in rail traffic, for example, still require the use of proven technologies. Technological changes should therefore require regulatory adjustments for railway operations. This is a real challenge when we know how much transportation safety is now being submitted by government agencies. In particular, networking with global traffic management can present a potential security risk: if several trains wait for each other because they are dependent on one or the other, waiting can trigger engorged knots or temporary overloads.
Modern AI methods can play a key role as elementary components of the entire automated driving process chain. They can be used, for example, in interpreting environmental perceptions, locating, forecasting, planning and implementing optimal driving strategies, as well as other related areas of infrastructure, such as: building and managing a large database or mapping the exact environment of the rail network.
The mechanical components of both the infrastructure and the rolling stock suffer from wear and the current techniques temporarily put them out of service at the least breakdown. Intelligent operational monitoring, predictive maintenance and intelligent and adaptive troubleshooting are able to increase asset availability, put fewer trains in the workshops, which is crucial for customer focus.
To do this, learning systems continually need data from different sources at all times and from which they can expect typical failure scenarios but also proactively intervene in maintenance. In the railway sector, the possibilities of intelligent surveillance are innumerable. In particular, continuous monitoring and evaluation of recorded sensor and telemetry data shall be performed, for example with regard to rail temperatures or other electrical components, time measurements, image recordings, etc. The infrastructure can be constantly monitored by drones, through a greater flexibility of the very strict regulation of these flying devices. The goal is to ensure nearly 98% availability of trains and 100% from infrastructure.
>>> To read: Digitalization in the rail sector, Infrabel at the top
Man / Machine Interface
« A crucial point of intelligent network mobility is well-designed interfaces for man-machine interaction (MMI). No matter where you are in a mobility space, people and intelligent systems will meet: whether it’s booking tickets, carpooling or bicycles, driving automated vehicles or simply crossing the street », says Tobias Hesse, Head of Vehicle and System Development at the German Aerospace Center and co-leader of the working group. The interaction must be user-friendly, intuitive and secure. « For example, a highly automated vehicle must be able to communicate intelligently with its occupants and with other drivers, cyclists or pedestrians at all times. Learning systems bring both new challenges and solutions. Innovative », he explains.
The report’s authors propose a global mobility platform that gathers, orchestrates and makes available to heterogeneous user groups the offers of various mobility service providers, as well as information on traffic and infrastructure. « In the next step, the Plattform Lernende Systeme working group wants to design a complete mobility platform, which should be the focal point for information from mobility providers, participants and infrastructure. Many stakeholders will use data bases and develop sustainable mobility products and offers », said Christoph Peylo. We are looking forward…
The report « Towards a smart mobility space » is available at this link, only in German
Vous devez être connecté pour poster un commentaire.