Busy urban tra c brings more and more problems for residents of large cities. The crossroads of two or more carriageways in the city are one of the main problems of tra c jams and delays on the road. The introduction of unmanned vehicles and systems capable of driving them safely and e ectively can contribute to solving this problem. The authors propose to consider the problem of intersections in the context of a mobile robotic system, since the organization of a group of mobile autonomous robotic devices most closely approaches the organization of the movement of unmanned vehicles. The authors considered the main strategies for organizing a group of unmanned vehicles, proposed and described a model of the system that controls the tra c at the intersection. To assess the feasibility of the developed model, a software simulator was developed, which allows to compare the e ciency of the developed model with the tra c management system using tra c lights. The results of the experiments performed allow us to say that the model proposed by the authors makes it possible to increase the capacity of the intersection within the framework of this study.
The development of the intelligent transport infrastructure (ITI) is one of the
most important areas of the internet of things (IoT). Automobile tra c in
modern large cities is a source of emission of harmful substances into the environment,
tra c jams and urban noise. In 2008, a study was conducted [
One of the ways to combat the increasing harmful e ects that car tra c has on the environment is to introduce unmanned vehicles (UVs) and systems that can organize their movement e ciently and safely. Driverless cars combine different technologies, devices, detectors and sensors for interaction with the environment and to obtain information from it, such as radar, computer vision, lidar, navigation technologies, odometry, etc. Tra c control systems receive information from these devices and make route planning taking into account obstacles, other road users and road terrain features.
Many companies are successfully engaged in the development and design of
UVs. For example, Waymo company [
Obviously, the use of todays methods of managing road tra c in cities, such as tra c lights, will be inappropriate in the context of the ITI of a smart city using autonomous UV groups. Applying a multi-agent approach, one can imagine a multitude of cars as a multitude of agents interacting with each other and moving according to certain rules. The speed of reaction and the calculations of the systems that control UVs, require a rethinking of the existing tra c rules developed for vehicles operated by humans.
To solve the problem of intersection management, the authors of this paper proposed a model of the system for safe and con ict-free travel of intersections by UVs. The model assumes that there is a transport infrastructure object (TIO) at each of the city intersections, which is responsible for organizing tra c at this intersection. At the same time, all such objects of transport infrastructure are able to exchange data with each other and optimize movement in such a way as to reduce the total time spent on overcoming the intersection by all UVs of the system. To determine the feasibility of using the presented model, the authors developed a software simulator and made a comparison with the intersection controlled by a tra c lights. This paper is organized as follows. Section 2 reviewed the scienti c literature in the eld of research into the problem of intersection management and the organization of the movement of UVs. Section 3 provides a classi cation of strategies for organizing UV group control. Section 4 describes the proposed model for the functioning of a tra c control system at an intersection and describes the main criteria for its functioning. The results of the experiments using the developed software simulator and their comparison with the intersection, the movement of which is organized using tra c lights are given in section 5. Section 6 presents the main conclusions of the study and describes plans for further research. 2
The idea of tra c control is being developed using the ant swarm system [
The article [
Researchers propose an alternative mechanism for coordinating the
movement of autonomous vehicles when overcoming intersections, based on the method
of representing vehicles as autonomous agents in a multi-agent system. [
In [
Au and Stone in [
A good overview of current research in the eld of intersection management
is given in [
On a at road, UVs can move autonomously and safely. Existing
technologies allow UVs to safely travel within the boundaries of the lane, change lanes
and avoid obstacles. With intersections, everything is much more complicated
vehicles need to cross several trajectories of other vehicles at once, which can
cause collisions in route planning systems for UVs. With the use of technical
means available to UVs and allowing communication between road users, it is
possible to implement a safe and e ective system for controlling the passage of
intersections. Applying a multi-agent approach, one can imagine a multitude of
cars as a multitude of agents interacting with each other and moving according
to certain rules. [
The speed of reaction and the calculations of the systems that control UVs, require a rethinking of the existing tra c rules developed for vehicles operated by humans. 3
In this paper, the authors consider a group of mobile UVs as a mobile robotic
system. Group management strategies are classi ed as centralized and
decentralized (see Fig. 1). [
The centralized management strategy can be divided into two classes - strategy using the principle of uni ed management and a strategy using the principle of hierarchical management. In the rst case, the group of objects contains a central control unit (CCU), which has a powerful computing center and carries out route planning, control and management of the activities of all the group Group management
strategies Centralized management
Decentralized management Unified management
Hierarchical management
Collective management
Swarm management Fig. 2: Centralized group management strategies: (a) centralized uni ed management strategy, r1; r2; : : : ; rn - system elements subordinate to the central control unit; (b) centralized hierarchical management strategy, r11; r12; : : : ; r1n; rm1; rm2; : : : ; rmk - groups of elements subordinate to the central nodes of the second level. objects. The objects of the group receive information from the environment with the help of sensors and detectors, transmit it to the CCU which, in turn, processes them and transmits various commands to the objects through which the group seeks to achieve a common goal. Diagram of the ows of information between the elements of the system is shown in Fig. 2(a). The advantages of such a system include the simplicity of its organization and algorithmization: only one central element is responsible for the formation of management tasks and their distribution. However, this strategy has a number of signi cant drawbacks. The CCU must have a powerful computing center, since it is entrusted with the task of optimizing the actions of all elements of the system. As the number of group members increases, the complexity of the optimization problem increases exponentially depending on their number, which may result in delays in making
Information exchange
channel r1 r2 rn r1 r2 rn Environment
Environment (b) Fig. 3: Decentralized group management strategies: (a) decentralized collective management strategy, r1; r2; : : : ; rn - elements of the system that communicate with each other to achieve the goal; (b) decentralized swarm management strategy, r1; r2; : : : ; rn - elements of the system. decisions that may be unacceptable when using systems with a similar strategy in the organization of transport infrastructure.
When using a hierarchical control strategy, there are several levels of control elements in the system. The rst level CCU controls a certain number of subordinates of the second level CCU, each of which is in charge of a group of robots. In such a scheme, second-level CCU receive information from agents subordinate to them, process it and transmit to CUU of the rst level, which, in turn, processes it and forms tasks that are transferred back to the second level and distributed among agents of each group. The advantage of such a scheme is that, compared with the uni ed management strategy, each individual CCU solves simpler tasks, which increases the overall speed of decision making. However, the complication of the management structure, due to the peculiarities of this scheme, can lead to signi cant delays or failures when transmitting commands between levels.The scheme of information ows between elements of such a system is shown in Fig. 2(b).
Systems with a centralized organization have such a common disadvantage as low fault tolerance. This is due to the presence of the CCU, the malfunction of which leads to disruption of the functioning of several agents or the entire system. Despite the possibility of using backup systems to maintain the performance of such systems, the costs of implementing such protective measures can be incomparably high. In systems with decentralized group management, there is no such drawback.
When using a decentralized management strategy, there is no CCU in the system, each agent has a computing center that has enough power to make decisions. When using such a scheme, the time spent on making decisions is minimized, as well as errors arising in the process of information exchange between agents of the group. One of the most important advantages of such a system is high fault tolerance - when one or more agents fail, the rest will continue to perform task.
Decentralized management strategy has more complex algorithms, each group member must make a decision that should ensure an approximation to the achievement of a common goal within the group. This implies a high intellectual level of all agents of the group, which implies the more complex task of optimizing the achievement of a goal within the group. Decentralized management approach can be divided into swarm and collective strategies.
The collective strategy is di erent from the swarm presence of the ability of group members to exchange information with each other. The organization of information ows in such a system is depicted in Fig. 3(a). The advantage of this approach is the possibility of increasing the e ciency of the group through the collective collaboration. However, this approach needs to ensure the protection of the information exchange channel, the violation of its work can lead to a violation of the data exchange between the participants.
The advantage of swarm strategy is high resiliency due to the lack of information exchange channel. The organization of information ows in a system with decentralized swarm control is shown in Fig. 3(b). When using this strategy, agents exist separately from each other and do not have the ability to communicate with each other, but they are able to analyze the state of the environment, and based on this, make decisions about further actions, Such actions should lead to achievement of a common goal, by changing the state of the agent and the in uence of other agents on the environment.
Thus, after analyzing the existing approaches to managing of a group of UVs, it can be concluded that when using a centralized management strategy to organize safe passage for a group of intersection vehicles, there are certain risks. With an unexpected sharp increase in the number of UVs approaching the intersection, the CCU must quickly solve the problem of optimizing the passage, while taking into account the speeds and trajectories of approaching cars, to form optimal routes for them. Due to the above-described speci c features of systems with centralized control, the simultaneous processing of large amounts of data can cause delays or malfunctions in the system, which can lead to a collision of UVs. This is unacceptable in the organization of a safe transport infrastructure, therefore, organizing travel using a decentralized management strategy for a group of UVs is seen by the authors of this work as more promising.
A fundamental factor in ensuring security during the intersection of intersections by a group of UVs is communication between group members. Based on the circumstances described, a decentralized collective management strategy was chosen to implement the model in the framework of this work. It is assumed that there are a number of intersections in the city, on which there is one TIO. TIOs carry out information exchange among themselves, jointly and decentralized solving the problem of tra c optimization at all intersections simultaneously. In the system there is no central object responsible for managing all intersections, each of the intersections is managed individually.
Denote C = fc1; c2; : : : ; cng - the set of UVs in the system. I = fi1; i2; : : : ; img - set of objects of transport infrastructure. The total square of travel sites S = s1 + s2 + : : : + sk; where si - square of i's place (see Fig. 4). At the same time there are two options for dividing the total area of the city into elementary areas, they are shown in (1) and (2). (1) (2) (3) (4) si \ sj = si \ sj 6=
Each TIO interacts with all UVs, at the same time, the car cj can interact with only one infrastructure object ih. The main objective of this work is to set the task of organizing the movement of UV within the city and to build the optimal route, in which the actual time tc of the vehicles movement is as low as possible tc ! min. The reference time of movement t0c is the time of movement of the vehicle in ideal conditions in the absence of interference and other vehicles that impede its movement tc ! t0c. Let tk - k-th instant of time, Pci = fpci1; pc2i2; : : : ; plcig - the distance traveled by the vehicle during the actual 1 travel time tc. Then, it can be illustrated in (3) and (4): :9 tk : Pctik \ Pctjk 6=
tc ! t0c 8c 2 C
In this case, the infrastructure object ii generates a route for the vehicle ci, taking into account the data received from the neighboring infrastructure facility ij . The authors of the article identi ed the following criteria for overcoming a group of intersections for vehicles: 1. The number of cars ready to cross the intersection tends to zero at each intersection. 2. The actual speed of the car is close to the expected when driving in the city and when overcoming the intersection: Speedavg ! Speedneed. 3. The area of occupied space at the intersection is minimal, Socc ! min. 4. The time for a vehicle to cross an intersection at real speed is minimal: tcarvogss ! min. 5. The total time spent on overcoming the constructed route with real speed also tends to minimum: Ptcross ! min.
For clarity of the principles of the system and the possibility of developing a software simulator, the following simpli cations are introduced: { UVs strictly follow directions at the intersection; { UV's computing devices know in advance their size, acceleration and deceleration dynamics, maximum speed; { UVs strictly follow the instructions for choosing the trajectory, speed, place to stop, received from the TIOs; { UVs and TIOs are equipped with communication devices, and it is understood that such devices provide ideal communication conditions, without delays, interference and data loss.
Thus, UVs and TIOs perform a number of inherent functions. Vehicles collect data on their technical condition and movement, data on the trajectory of other cars and transfer them to infrastructure facilities, store a plan of the city zone along which the route passes. Infrastructure objects accumulate information about the system, develop a plan for locally and globally optimal plans for the movement of vehicles in the city, monitor the performance of tasks by means of transport, and control the activities of other infrastructure objects if the city or automobiles they control are common. 5
To evaluate the e ectiveness of the proposed model, a software simulator was developed that imitates the movement of UVs within the city, in particular, their interaction at the intersections. The simulator involved creation of an intersection model and a UV model. Intersection model requirements: { the allocation and state of all the elementary road sections are known to all the tra c participants; { the beginning of the road should be situated on one of the borders of the simulated intersection, meanwhile its end should be on the opposite side, i.e. it is assumed that there are only straight roads which coordinates coincide with the coordinates of the elementary sections located in the same row (column); { each road must belong to either vertical or horizontal type.
The intersection model includes the following set of characteristics: fc; r; Rg, where c - number of columns that de ne the intersection; r - row amount de ning the intersection; R = fR1; R2; : : : ; Rng - set of roads where the vehicle can move. In its turn, each road is characterized by a set of parameters ft; d; Eg, where t - road type (vertical or horizontal); d - road direction (passing or oncoming); E - set of elementary sections de ning the roadway. For the experiment, it was decided to limit to 4 lanes: two vertical (oncoming and passing) and two horizontal (oncoming and passing). Intersection size: 10 10 elementary areas. The general model of the simulated intersection is presented in Fig.4.
UV model assumes presence of the following characteristics: { E - set of elementary sections based on a road map and the planned start and nal positions of the UV; { s - initial (maximum) speed. Speed is understood as amount of the elementary areas crossed by a UV per one conditional time discrete. In the conducted experiments maximum UV's speed is considered equal to 2, also, as the UV approached to the intersection, it reduced speed smoothly and passed the intersection on the minimum speed equal to 1; { c - turning point if the nal position of the UV is on a di erent road (taking into account the direction of road movement); { ST - sequence of steps for the UV to go through the planned path (calculated on the basis of E; s; c; one step is passed in one conditional time discrete).
Conditions of the experiment: { UVs can move in any direction within the roadway, according to the direction of the roads; { the intersection model is spatially limited; { the number of UVs simultaneously on the observed eld is limited by the capacity of the current section; { in case of a con ict (more than one UV pretend to the same elementary section contemporaneously), the UVs give way to each other, taking into account the maximization of the intersection capacity (see (5)): >>8 PL PN M >>>Y = l=1 j=1 iP=1 nlji > > <>> 8>1; iMfjth UV is situated at the ith time discrete on >>>>>>>>nlij = >:<t0h;eotlhtherweliesme entary section, nlji 1 6= nlji > :Y ! max ; (5) where N - number of UVs, passing the intersection; L - number of elementary sections at the intersection; M - amount of time discretes, for which N UVs passed the intersection; { on the intersection the probability of the appearance of new UVs is given, the number of appearing vehicles is determined randomly.
For comparative analysis, a simulator based on conventional tra c lights control was prepared. Its properties are the following: { the presence of tra c lights at the intersection;
8,272 6,197 6,543 7,149 8,905 0,2
0,5 Appearance probability of a new UV 1 Time based on the developed model
Time based on the traffic lights { calculation of ST is performed in the same order as in the model described above, but during the entrance into road intersection, the signal of the tra c lights is checked.
For the tra c lights control system at the intersection, the following cycles are taken: green light - 30 seconds, 4 seconds yellow, and 1 second all red.
The average time of the crossing of the intersection by UV was chosen as an indicator for comparing the quality of the algorithms. Three groups of experiments were conducted, di ering in the appearance probability of new UVs on the intersection: P = 1 (at least one new UV appears), P = 0:5 and P = 0:2. The duration of each experiment is 1000 discrete time increments. After a series of independent launches for each group of experiments, the results took the form presented in Fig. 5.
According to the represented data, it is possible to note that the proposed model of the intersection control system allows to reduce the average intersection passing time per UV by an average of 20,5% in comparison with an intersection regulated by tra c lights. 6
As a result, in this work some approaches to the tasks of intersection management using objects of transport infrastructure were considered. The system was considered as a mobile robotic system, a model of interaction between unmanned vehicles and objects of transport infrastructure was developed. To test the e ectiveness of the model, a software simulator was developed that allows simulation of the control of car tra c at the intersection using the developed model and using tra c lights. The results of the experiments showed that the developed model is more e cient than the use of tra c lights. In future studies, it is planned to model automobile tra c at several intersections within the framework of a smart city and to create a physical model for carrying out real experiments.