For a long time to come, road traffic's safety, comfort, and ecology will depend on the cooperation of purely human-driven and (to some degree) automated vehicles.
There is no question that automation promises progress in driving security and comfort. In today's cars, several driver support systems are standard (such as those for headway distance control lane-keeping control). Further research and development are underway towards partially or fully automated transport using Intelligent Transport System (ITS) technologies. On the roads, human-driven and autonomous vehicles can move together (what we call here mixed traffic). More importantly, for the foreseeable future, the behaviour of the people driving the vehicles will also be affected by what kind of intelligent features help the driver. Therefore, in addition to equipping a single vehicle with intelligent solutions, a central challenge is facilitating interaction between drivers supported by different levels of automation. To support the safe driving of increasingly automated road vehicles, forecasting the behaviour of other road users is extremely important.
By learning about the group behaviour of vehicles (and other road users such as pedestrians, cyclists, scooters) travelling in complex interactions, this heterogeneous transport will be safer, more efficiently regulated, and managed. Behavioural analysis of vehicles inroad transport that considers the interactions between members of a larger related group is a new area of research.
A consortium of BME, Magyar Közút Zrt., and Dmlab Ltd. is researching the swarm behavior of nearby cars (and othert ravelling entities) in the framework of the RAJ project.
Many road users participate in the traffic along relatively simple rules and metacommunication signals supplementing the formal rules and making local decisions. Even in today's traffic, the spontaneous interactions of road users are fundamental, which is taken into account by human intelligence based on a variety of (including many non-formalized, intuitive) perceptions.
In contrast to free individual decisions, the emerging transport systems must also exercise mutual influence in an organized manner. (For example, at overtaking, at intersections, in urban traffic.) Therefore, modelling the behaviour of a single vehicle or driver is not sufficient to understand and control mixed traffic situations.
Also, in developing vehicle automating features, it is essential to adopt approaches to improve the performance of the whole driver-vehicle cooperation. Understanding human drivers' intentions and recognizing driver behaviour is imperative to facilitate smooth and appropriate control mode transitions.
To illustrate the problem and the perspectives, we can take as an example the behaviour of swarms that occur in nature (e.g., the group movement of birds, bees, or ants). Although research has begun in recent decades to examine the behaviour of animal swarms, the disciplines involved have typically not sought formal analytical modelling. Applying the research approach to vehicle fleets is therefore still in its infancy. Some simple models depict the interaction between a given vehicle and its neighbours but do not consider types of human behaviour. For example, traffic congestion is a complex phenomenon that is difficult to predict accurately(sometimes even to find the cause in retrospect). To describe this, the so-called concept of agent-based modelling has emerged (the system consists of decentralized individual agents, each agent interacts with other agents according to localized knowledge). Interacting agents can be drivers, cyclists, scooters, pedestrians who have certain goals and decision-making powers. Local interactions between individual agents can lead to the development of group behaviour. That is a bottom-up approach.
The swarm behaviour-based approach provides a promising methodology for investigating hybrid road transport with an even larger volume of heterogeneous automation in the future. Describing the micro-level behaviour of entities in road traffic modelling as a swarm provides a promising and more efficient methodology for understanding previously unexplored anomalies in traffic and both long-term and short-term predictions. In the RAJ project, especially concerning hybrid transport, we use swarm behaviour methodologies and new models to investigate how road capacity, roaduser safety, and passenger comfort can be improved and predicted even inunexpected, extreme situations.
As a practical result of the research, we expect that the throughput and safety of road transport can be optimized: by informing the drivers expediently; in the case of higher-automated vehicles, even certain local coordination of control functions is feasible at the swarm level. We identify different typical and deviant behaviours, learn about their impact on transport. We are developing a method to validate the results of the impact assessment. That results in higher efficiency utilization of the oretical traffic capacity on a given road and minimization of critical situations. In the short term, the project results can be used to prepare CCAM living laboratories in Hungary for various research and development projects.
In mixed road transport, the emergence of highly automated individual vehicles may have few overall benefits. However, modelling the lessons of swarm-level behaviour can achieve optimization and security improvements interpreted for entire traffic control units. Machine learning models will be developed for data obtained through swarm intelligence, and the behaviour of elements and associated anomaly detection procedures can be determined. It can be used to identify an object that affects the traffic situation, a vehicle that behaves in some way special, strange (e.g., towed vehicle, slow car, slippery road section, etc.), or dangerous(e.g., due to malaise). When we recognize the vehicle group as a swarm, each vehicle may receive a signal or advice to deal with such anomalies.
We set up the swarm and individual behaviour models during the project by analyzing the data observed in real road traffic. To this end, we collect large amounts and high variety data sets on real road traffic, both on highways and at urban intersections. Developing data collection infrastructures are on the way in the RAJ project. One of the urban intersections (in Zalaegerszeg) is already equipped with sensors (a prototype, based primarily on a camera set and data management IT); the following type will be more sophisticated, including AI control and real-time features. A section of the M7motorway also is a data source for the project, using cameras, microphones, and a weather station. Data streams originated in future HungarianCCAM infrastructures also planned to be involved in the project. IT environment for the development of new methods and algorithms in scientific cooperation are also under development.
As a result of the project, we are on our way to building a monitoring system of self-driving vehicles to examine the cooperation of such vehicles independently of the manufacturers. It will be possible to analyze what changes these vehicles bring to traffic and how they change drivers' reactions and driving habits sitting in non-self-driving vehicles. During the evelopment of self-driving vehicles, only the data from the internal view of the development company is collected. Still, with the help of road network observation points, changes in the typical behaviour of vehicles and drivers can be examined in large quantities:
It is also possible to show the effect of weather phenomena or other traffic factors by examining a large mass of road users.
In a simulation (and perhaps emulation) environment, it becomes possible to validate these models and deepen the understanding and examination of traffic situations.
An interdisciplinary research group has been set up to bring together the scientific expertise needed for transport behaviour analysis (transport, automotive, communications, IT, artificial intelligence, mathematics, and psychology).
BME provides interdisciplinary scientific competencies and develops the necessary data management and artificial intelligence-based deep learning analysis R&D platform. Magyar Közút Zrt. has the broadest practical experience in Hungary in road transport planning, organization, and management tasks. Dmlab Kft. brought its solid and creative data analysis competence and experience to the consortium. The research group, the infrastructure, and the developed methods can be integrated into domestic and international intelligent transport projects, the domestic professional potential of the collaborations becomes more valuable. The collected real-life traffic data also helps to connect to the research and development bloodstream in an international environment. One of the crucial conditions for artificial intelligence research is traffic data from the real-life environment of the right size and quality.
(The RAJ project issupported by the Innovative Mobility Program of the Institute of Transport Sciences (KTI).)
Written by Data Management and Communication Team