Mathematical Problems in Engineering Improved CNP-Method-Based Local Real-Time Cooperative Task Allocation of Heterogeneous Multi-UAV in Communication-Constrained Environment
UAVs are widely employed in military and civilian fields because of their inherent advantages. The cooperative task allocation of multi-UAV is more in conformity with the requirements of UAV application scenarios, which has become a hot research topic. However, resource consumption allocation and the application scenarios of communication constraints are often ignored. This paper proposes a distributed multi-UAV task allocation method based on improved CNP to solve the local cooperative task allocation problem of heterogeneous multi-UAV in the communication-constrained environment. The improved CNP-based method can be divided into four stages: task release, bid application, coalition formation, and signing contracts. In the task release stage, we proposed the adaptive maximum number setting method of information transfer times and the information consistency method to solve conflicts in the local communication network. In the process of coalition formation, the resource consumption allocation algorithm based on the Gini coefficient is proposed to keep the resource difference between UAVs in the coalition within a reasonable range. The simulation results demonstrate that improved CNP-method-based cooperative task allocation can handle the local real-time task allocation problem of heterogeneous multi-UAV under communication constraints; it obtains greater task rewards and spends less time on task completion than the resource-welfare-based method, PTCFA. Simultaneously, the resource consumption algorithm makes the UAV swarm maintain a more reasonable resource difference to maximize the number of missions completed.
Unmanned aerial vehicle (UAV) has major benefits in terms of survivability, cost-effectiveness radio, and specifications [1, 2]; it plays a wide role in the military and civilian fields, including fire detection, target reconnaissance, fire strikes, information countermeasures, and a variety of other mission situations [3, 4]. Due to the limitations of a single UAV’s autonomy and functional unit loads, such as sensors and weapons, a single UAV cannot meet the increasingly complex application environment and mission requirements . Unlike a single UAV, multiple UAVs can share resources and cooperate to enhance the ability to perform tasks. The coordinated operation of multiple UAVs can better cope with the risks that may arise in the execution of tasks, shorten the task time, and improve the quality of task completion. Therefore, the UAV collaboration problem is an increasingly important area in UAV application research.
The collaborative target allocation problem is establishing an association and mapping link between the multi-UAV and the tasks, which is a combinatorial optimization problem with numerous complex constraints . Reasonable task allocation can help UAVs better complete tasks based on satisfying various constraints. The task allocation problem has gradually developed from a simple, small-scale ideal environment to a complex, large-scale real environment. The solution method has developed from a centralized method with low precision and strong global search ability to a distributed method with strong real-time performance. Faced with different environments, the requirements for solving algorithms are also different.
This paper aims to solve the collaborative task allocation problem of heterogeneous multi-UAV in a communication-constrained environment. The main contributions of this paper can be summarized as follows:(1)A method for information consistency and the information transfer strategy in a communication-constrained environment is proposed; it makes sure that the CNP algorithm runs smoothly by setting up a stable communication network between UAVs.(2)A local real-time task allocation method based on CNP is proposed, which is applied in a distributed architecture. It considers various constraints in the communication-constrained task environment. Simultaneously, the Dubins curve is applied to the path planning of UAVs to meet the constraints of the minimum turning radius.(3)The allocation method based on the Gini coefficient is designed for the resource consumption of UAVs during task allocation so that UAVs of the coalition can consume resources more rationally and hence execute subsequent tasks better.
The remaining sections of this paper are organized as follows. Section 2 introduces the current research on the collaborative task allocation problem. Section 3 describes the collaborative task allocation problem of heterogeneous multi-UAV and establishes mathematical models. In Section 4, a local real-time task allocation method based on CNP is proposed to solve the CMTAP problem with various constraints, and the allocation method based on the Gini coefficient is designed to reasonably control resource consumption. Section 5 illustrates the algorithm’s performance through simulation experiments. Section 6 gives corresponding conclusions and briefly takes over future work.
2. Related Work
Over the past decades, the task allocation of multi-UAV has become one of the focuses of research in UAVs, which is mainly because the existing tasks put forward higher requirements for UAVs, requiring larger-scale multi-UAV to work together. As scale increases, so does the heterogeneity of drones and missions. This section mainly reviews related research on mathematical models and solutions for multi-UAV task allocation.
The task allocation problem is represented using one of many classical optimization problem models or variations after considering the task requirements, constraints, and characteristics of the UAV in a particular situation. The commonly used classic models are multiple traveling salesman problem (MTSP) , vehicle routing problem (VRP) , and other models. Zhu et al.  considered the limited resources of UAVs and modeled the task assignment problem as a capacity-constrained vehicle routing model (CVRP). Recently, the target allocation model study has focused mostly on the collaborative target allocation field.
For modeling task allocation, many factors need to be considered, including UAV, task, and constraint, which contain resources, minimum/maximum speed, turn radius constraint, and so on. Moreover, the task model depends on the corresponding task scenario. Zhang et al.  proposed a UAV cooperative attack task allocation model and strategy for attacking enemy air defense systems in a specific area. The method meets the goal of real-time requirements in a dynamic environment. Wu et al.  modeled the task allocation problem of multiple heterogeneous fixed-wing UAVs on multiple targets performing confirmation, attack, and damage assessment as a CMTAP problem and used an improved genetic algorithm to solve the problem.
Numerous centralized and distributed methods have been proposed [12, 13] to solve the multi-UAV task allocation problem. Centralized solutions, such as ant colony optimization , genetic algorithm [15, 16], and particle swarm optimization , rely on UAVs to communicate continually with the central station that creates a plan for the whole UAV swarm. Amorim et al.  proposed an improved Swarm-GAP method that can be applied to task assignment in dynamic military operations environments where UAVs are randomly destroyed. However, in the reallocation process of this method, more communication resources are required, and it is not suitable for a communication-constrained environment. Acevedo et al.  proposed an improved genetic algorithm, which considers more constraints for real task scenarios, including kinematic constraints, resource constraints (weapons and fuel), and time constraints. However, as a centralized algorithm, communication constraint is not considered. In the communication-constrained environment, centralized solutions respond slowly to dynamic changes [4, 19]. Then the distributed solution, such as contract net protocol (CNP)  and the consensus algorithm , is offered to resolve the issue. In the distributed solution, each UAV can directly process task information as a relatively independent decision-processing unit. It has strong fault tolerance, flexibility, and reliability. Additionally, each UAV should transmit bid information to the auctioneer, and this requires a stable connected topology in the CNP algorithm . Xiang et al.  regarded the missile as an agent and proposed an improved CNP algorithm to ensure that the missile could be assigned to a better target by updating the contract. The consensus algorithm necessitates massive information transmission to achieve consistent convergence . Duan et al.  proposed a new hybrid auction algorithm, which greatly improves the task allocation effect through hierarchical decision-making.
However, numerous studies predicate that UAVs are homogeneous, targets’ information is known in advance, and communication is unrestricted. In the complicated military environment, these assumptions are often not established. For example, when UAVs perform missions, they only know the rough information of the mission before taking off and need to perform real-time task allocation reasonably, which needs to meet limits such as resources, simultaneous attacks, and so on. In a war scenario, the tasks faced by the UAV swarm are unpredictable, including time, position, and resource requirements. It requires resource consumption allocation to be reasonable enough. Kim et al.  proposed a coalition formation method that effectively uses resource welfare as a consumption strategy to use attack resources and enables the UAV swarm to execute follow-up missions more successfully.
3. Mission Description and Model
In this section, the mission scenario, UAV, and target are modeled, and the more scientific objective function is proposed.
3.1. Mission Scenario
According to the set path, heterogeneous UAVs conduct reconnaissance in the mission area. When they reach the mission area’s boundary, they take the shortest path back to the mission area. Furthermore, they enter the strike mission phase when they find the target. The relevant information about the target is unknown before the UAVs take off, but when the reconnaissance UAV finds it, specific target information is collected by the UAV that finds it. Based on the actual requirements of cooperative operations, the UAV that finds the target creates an attack coalition with the surrounding UAVs to carry out strike missions together in the communication-constrained environment. To simplify the end-strike model of the attack mission, the condition for the completion of the attack mission is that the UAV coalition carries the resources required to destroy the target and reach the target within a certain range.
In this paper, various constraints are considered simultaneously in the study of cooperative heterogeneous multi-UAV task allocation, primarily the constraints of the UAV’s performance, the constraints of attack resources, the time constraints of simultaneous attacks, and the communication constraints between UAVs. The paper focuses on the local real-time heterogeneous multi-UAV task allocation method with limited communication.
3.2. UAV and Target Model
In this paper, the take-off and landing of the UAVs are not taken into account, and altitude layering is used to achieve collision avoidance between UAVs, with the height being constant during the flight.
In the two-dimensional plane, the flight motion model of the UAV can be expressed as follows:where is the current position coordinates of the UAV, is the velocity of the UAV, is the yaw angle of the UAV, and is the velocity of the yaw angle. This paper ignores the UAV’s aerodynamics and attitude, and striking missions do not necessitate a certain flight direction and angle.
There are a total of static targets on the set mission area. When the UAV finds the target , it can directly obtain the resource requirements, including types and respective numbers of ammunition required to destroy the target. Resource requirements can be represented as a set of vectors:where m is the types of ammunition required for the target to be destroyed and is the number of ammunition required for the to be destroyed. For example, represents that destroying the target requires two resources and three resources.
There are UAVs in the mission area before the reconnaissance mission begins, and the attack resources carried are not time-sensitive and do not alter as time passes. The heterogeneity of UAVs is set as the difference in attack resources. And attack resources of can be represented as follows:where is the number of type resource of and is the number of types for the entire UAV swarm.
In the communication-constrained environment, each UAV can only communicate with UAVs within its communication range. To avoid two UAVs being unable to communicate with each other after discovering the same target, two coalitions are established to attack the target, and the communication range and reconnaissance radius are configured as , where is the communication range of UAV and is the reconnaissance radius of UAV.
3.3. Objective Function and Constraints
3.3.1. Cost Function of the Assault Coalition
where is the number of members for the attack coalition, is the fuel consumption coefficient, is the flight distance of , is the self-value of , and is the threat level of the target. is the attack coalition for .
3.3.2. Task Completion Revenue Function
where is the target value when it’s discovered, is the time decay function, and is the time interval between the discovery of the target and its destruction. And target time sensitivity is represented by . The higher the value, the quicker the goal value declines.
3.3.3. Total Revenue Function
where is weight coefficient; . It is simple to change the orientation of the objective function by modifying the value. For example, raising can hasten the completion of the mission, and raising can form coalitions with fewer UAV losses and shorter total flight distance.
The main constraints examined in the task allocation problem of heterogeneous UAVs performing real-time strike missions in a communication-constrained environment are as follows:(1)Resource constraints The target destruction necessitates a certain amount of resources; thus, the combined resources of the attack coalition must satisfy the requirement: and the quantity of is limited.(2)UAV self-performance constraints The fuel resources must be sufficient to ensure that the UAV can complete the range required; otherwise, it will be unable to join the attack coalition. It is set that the fuel is proportional to the flight distance. The flying distance constraint can be converted from a fuel constraint. The UAV’s flying distance must match the following requirement: where and are the positions of the and the , respectively, and is the remaining flight distance of . At the same time, the turning radius of a UAV is limited.(3)Time constraints To ensure that the UAV coalition cooperative combat, all of the coalition’s UAVs must reach the target point simultaneously. When a UAV has a short flight distance and a short flight time, it might hover or slow down its speed to increase the time it takes to reach the target. It can be expressed as follows: where is the flight time of . is the distance from to . is the velocity of , and .(4)Communication constraints The geometry of the air-to-air radio path is shown in Figure 1; its communication range is limited by terrain height; however, when the radio path crosses the horizon between UAVs, the communication range between UAVs is unaffected by terrain height. In this paper, the UAV communication is unaffected by terrain height.
In a communication-constrained environment, the two primary constraints are the communication range constraint and the time delay. The communication range constraint mainly means UAVs that are far away cannot communicate directly; out-of-range information transfer must be accomplished through a communication relay station. Communication time delay mainly refers to the time required to transmit information between UAVs.
4. The Improved CNP-Based Collaborative Task Allocation Method for Heterogeneous Multi-UAV
In this section, the improved contract net protocol (CNP) based collaborative task allocation method for heterogeneous multi-UAV is proposed for local real-time tasks of UAV swarms in a communication-constrained environment. The process of the target processing by the UAV is introduced in detail. This method can meet target resource requirements, time coordination, optimal range, UAV performance, communication constraints, and realize local real-time task allocation.
4.1. Local Real-Time Task Allocation Method with Limited Communication
Smith  firstly proposed the contract net protocol as a framework for solving distributed negotiation problems in 1980. In the market mechanism, the simulated bidding negotiation process is employed to find the optimal solution for the objective function. The CNP method has been shown that has superior scalability, robustness, and negotiation efficiency to other methods; it is suitable for local real-time task allocation.
In the local real-time task allocation of heterogeneous multi-UAV, the steps of the local real-time task allocation method based on CNP are divided into task release, bid application, coalition formation, and sign contracts. The process of the contract network is shown in Figure 2.
In the step of task release, a new communication method is designed according to the characteristics of communication constraints to ensure that the UAV information can reach an agreement within the scope of the local communication network. Moreover, a new resource consumption allocation method is designed in the bidding application information processing step, combining the Gini coefficient in economics, to ensure that the UAV swarms reserve reasonable attack resources to deal with unanticipated missions. The following are the specific details of the four steps.
4.1.1. Task Release
After finds the target , it judges whether is established. If the condition establishes, executes the task alone; if not, becomes the leader of the coalition and sends task bidding information to all UAVs in its local communication network for seeking coalition formation.
During the flight of the UAV swarm, the communication is formed into an ad hoc (point-to-point) network , and the location of UAVs is constantly changing. Depending on whether there is a communication node, the communication mode between UAVs can be divided into direct communication and indirect communication . UAV swarms often use flooding to transmit information to other UAVs within the communication range, ensuring that all UAVs can receive information. However, there are often problems with implosion and broadcast storms . Peng W et al.  proposed that limiting the number of broadcasts could help solve the issues mentioned above. The number of broadcasts can be limited in the task release and bid application stages. In the task release stage, the number of nodes between the single-link communication of the UAV can be limited, which is named as the maximum number of information transmission ; in the bid application stage, the bidders’ conditions are strict to avoid environmental changes leading to contract breach, which causes waste of communication resources.
To reduce the communication load between UAVs, a self-adaptive maximum number of information transmissions method based on the target level and resource requirement (AT-MIT) is proposed. AT-MIT can ensure that when attacking important targets, the coalition can be formed as successful as possible; when attacking secondary targets, it reasonably reserves communication resources and speeds up the formation of the coalition. It can be expressed as follows:where is the weight coefficient, is the level of , is the weight vector used to transform the target’s resource requirement into the difficulty level of destruction, and is the roundup function.
Before the coalition leader performs the information transfer between UAVs, the remaining information transfer time is first judged. Initial time, , and with each transmission, is reduced by one. When , information is no longer transmitted.
For example, when the finds the , it will get all information about , including the level and the resource requirement of . By calculating , then . The local communication network of is shown in Figure 3. The dotted circle around the coalition leader indicates its reconnaissance range and can communicate directly, bidding information is transferred from to , and minus 1. Then serves as a communication node and sends bidding information to within its communication range except . Although is within the communication range of , ; then the bidding information is not transmitted from to .
Before the coalition is formed successfully, the UAV swarm has been conducting reconnaissance according to the set reconnaissance route. Furthermore, its communication topology is in dynamic transformation, so it is critical to screen out coalition members to construct a stable communication topology. This paper selects the coalition members by judging the expected arrival position (EAP).
After finds , transmits bid information to all UAVs in the local communication network. In order to screen out potential coalition members that constitute a stable communication topology, UAVs receiving bid information send the EAP at the moment of the successful coalition formation as part of the bid application to the coalition leader . The moment of coalition formation successfully is expressed as follows:where , , and represent the moment of coalition formation, moment of finding , and time of coalition formation process, respectively. And coalition formation process includes coalition leader sending bid information, potential coalition members sending bid applications, bid application information processing, and signing contracts, which can be expressed as follows:where is a single information transfer time between UAVs, is the time to process bid information and plan the coalition’s offense, and is the time to calculate EAP.
After finds , gets information about the target. The information can be expressed as , where , , , and is the position, resource requirement, level, and threat of , respectively. Coalition leader sends bid information to UAVs within communication range. The information can be expressed as , where is serial number and EAP of UAV. is the estimated revenue of the . The bid information is transmitted to each UAV according to Algorithm 1 in the local communication network, and UAVs are consistent with the information.
As shown in Algorithm 1, UAVs in the local communication network transmit the task bidding information to the UAV, which can be directly connected (lines 1–5); then the UAV that receives the information judges whether it meets the communication conditions and resource requirements firstly and selects the optimal task bidding information according to the mission revenue (lines 6–14). The UAVs in the local communication network are consistent with the bid information through the information transmission.
4.1.2. Bid Application
After step 1, each UAV in the local communication network of retains the same bid information. The UAV swarm is in flight, and there is a communication time delay between UAVs. If each UAV calculates the estimated time of arrival (ETA) with the target according to the moment when it receives the bid, in the process of information exchange, the ETA will change and cause the performance of task allocation to be degraded. Therefore, each UAV needs to agree on a unified time. Because the bid information includes the moment of the target was discovered and the time of coalition forms, the UAV needs to calculate its EAP according to the moment of coalition formation. Then UAV uses EAP to calculate its ETA to the target and send the bid application to the coalition leader. The bid application of can be expressed as follows:where is the UAV for bidding, is the attack resource carried by , and are the EAP in the moment of coalition formation and ETA to the target, and .
4.1.3. Coalition Formation
After step 2, the coalition leader receives all bid applications and needs to deal with the application information. Firstly, the leader should screen out coalition members that can stably exist in the local communication network. Then under the expectation of the minimum number of coalition members, the leader screens out optimal coalition combination according to the objective function. After the coalition members are determined, members share the target’s attack resource requirements—the coalition leader of plans an attack scheme and resource consumption allocation scheme.
4.1.4. Signing Contracts
The coalition leader transmits the information of the winning bid to the UAVs in the local communication network. Then the UAV that wins the bid adds the task of hitting the target to its task sequence, and the members of carry out the task according to the plan formulated by the leader. The members of the coalition who received the information earlier continue to perform the original task until . In other words, members act simultaneously when the farthest UAV receives the winning bid information.
4.2. Resource Consumption Allocation Method Based on Gini Coefficient
After the coalition leader selects the combination of members , the UAVs in the coalition share the resources requirement. In the case of resource constraints, the greedy algorithm is often used to allocate the required attack resources. This algorithm leads to unscientific resource consumption in the UAV swarm; some UAVs are rendered incapacitated prematurely, resulting in lower success rates for subsequent coalition formations. It leads to a decline in the robustness of the UAV swarm in mission execution, the number of missions completed, and the dynamics of responding to unexpected missions. This paper proposes a resource consumption allocation method based on the Gini coefficient to solve the unreasonable allocation of resource consumption problem.
4.2.1. Gini Coefficient
In 1912, the famous Italian statistician Corrado Gini proposed a coefficient to measure the degree of inequality in the distribution of wealth. The characteristic of the Gini coefficient lies in its integrity. It is often used to evaluate the degree of equality of wealth distribution in a country. When there are fewer residents in the country, the conclusion of its judgment is more reliable. The UAV swarm can be regarded as a temporary country, and the attack resources carried can be equivalent to wealth income. The Gini coefficient can measure the equilibrium degree of UAV attack resource allocation.
The geometric meaning of the Gini coefficient is based on the Lorentz curve. When the Gini coefficient is used to measure the balance of the distribution of UAV attack resources, the definition of the Lorentz curve is to rank the UAV attack resources from low to high, and the cumulative percentage of the number of drones and the cumulative percentage of attack resources are calculated. The corresponding relationship is drawn on the graph; the Lorentz curve is obtained as shown in Figure 4.
In Figure 4, the calculation of the Gini coefficient can be considered as the ratio of the area formed by the connection of the absolute equilibrium curve and the Lorentz curve ( area in Figure 3) to the area formed by the absolute equilibrium curve and the horizontal axis ( area in Figure 3), which can be expressed as , but the Lorentz curve is irregular; the area of and cannot be calculated directly. In this paper, the method of reference  is used to calculate G; the area of is approximately divided into the area of several trapezoids , which can be shown as follows:where is the percentage of the total resources from the UAV with the least resources to the UAV.
4.3. Resource Consumption Allocation Method Based on Gini Coefficient
After the coalition to attack the target is formed, the sum of various resources in the coalition satisfies task resource requirements, which can be expressed as . For the -type attack resources in the , firstly, the coalition leader screen out the set of UAVs that contain -type resources. Then the leader traverses all resource allocation combinations that meet the target resource requirements and calculates the Gini coefficient of the remaining resources for each combination separately. The combination of the optimal Gini coefficient is used as the allocation scheme of the -type resource. The leader calculates the allocation scheme of other types of resources in turn and finally summarizes into the resource consumption allocation plan. The detail is described as Algorithm 2.
4.3.1. Optimal Gini Coefficient Value
For a series of tasks on the battlefield, the resource consumption allocation method based on the Gini coefficient can effectively control the difference in resources among UAVs. Moreover, it improves the UAV swarm’s robustness, the number of tasks to be completed, and the dynamics of responding to sudden tasks improved. But the performance improvement is based on selecting the appropriate Gini coefficient (G) value. Therefore, the G value needs to set an appropriate evaluation function and corresponding experiments to select according to the characteristics of the mission scene.
The selection of the G value is affected by many factors, such as the total number of tasks, the size of the UAV swarm, the location of a single UAV, and the number of resources initially carried by the UAV. In the communication-constrained environment, UAVs in the center of the UAV swarm are requested by more coalition and undertake more tasks. Because of its mission-critical status, it should carry more attack resources than UAVs at the edge of the UAV swarm. Therefore, the selection of the optimal G value should mainly be affected by the mission status of the UAV.
5. Simulation and Analysis
In this section, four simulation experiments have been performed to verify the performance of the improved CNP-method-based communication-constrained task allocation. In the first group, the method proposed is used in the example scenario to illustrate the whole process of heterogeneous multi-UAV performing cooperative tasks. The second group contrasts the polynomial time coalition formation algorithm (PTCFA), resource welfare-based algorithm, and the improved CNP-based algorithm to analyze the algorithm performance. In the third group, experiments compare the performance between the AD-MIT and the fixed maximum number of information transfers in the example scene by the Monte Carlo method. And, in the fourth group, the greedy algorithm, balance consumption algorithm, and resource consumption based on the Gini coefficient algorithm are compared.
All simulation experiments in this paper are implemented on the MATLAB 2019b version of the 16 GB Intel(R) Core(TM) i7-10875H. The relevant parameter settings are described in detail in the experimental settings of each section.
5.1. Heterogeneous Multi-UAV Collaborative Task Scenario
The mission area is set to 3200 m. The UAV swarm comprises six observation strike integrated UAVs; each UAV’s type and the number of attack resources are different. There are two task objectives in the mission area, and the information is shown in Table 1. Before the UAV swarm takes off, there is no prior information about the target until a UAV finds it and starts the local real-time task allocation. The information on UAVs at the initial moment are shown in Table 2. The relevant parameter settings are as follows: UAV reconnaissance range, communication range, and flight velocity are 250 m, 1,000 m, and 50 m/s, respectively.
At the initial moment, the UAV swarm performs the reconnaissance mission according to the set path. The finds the in 33.0 seconds. becomes the coalition leader after assessing that its resource is insufficient to meet the requirements of . Then it sends bid information within its communication network. can be calculated according to equation (10), and the local communication network of is shown in Figure 5; the solid line is the track of the UAV when it performs the reconnaissance mission; and the end of the solid line is the position of each UAV when was discovered by . Then the UAVs connected by the dotted line express that communication is possible at present. According to the bid application submitted by the UAVs in the communication network, is formed to attack . The Dubins track of U3, U4 attacking T1 and the actual distance to the T1 are shown in Figures 6 and 7, respectively. According to Figure 6, it can be seen that the planned track meets the minimum turning radius constraint, and it can be seen from Figure 6 that UAVs in the coalition can reach the target at the same time.
The finds the in 41.3 seconds. After judges that its resources cannot meet the requirements, it sends the task bid information to other UAVs in the local communication network. And is calculated based on target information. The local communication network of and the path of execution of the task is shown in Figure 8. But are in the execution stage and do not participate in the bid of the . In the same way, is screened out to attack . The Dubins track of the UAV in the coalition and the actual distance to are shown in Figures 9 and 10, respectively. It can also be seen that the UAVs in the coalition can reach at the same time and track satisfies minimum turning radius constraint.
When all targets have been hit, the track of all UAVs is shown in Figure 11. After the UAV reaches the boundary of the mission area, it returns to the area by the shortest path. After completing the mission, the UAV continues to perform the reconnaissance mission to find the next target such as the and .
The formation of the coalition in the process of task allocation is an important link in determining the quality of task allocation. The experiment in this section compares the performance of the three algorithms resource welfare-based algorithms , PTCFA , and improved CNP-based algorithms.
The mission area is set to 5,000 m. There are 30 randomly distributed fixed targets in the mission area. Each target requires three types of attack resources to destroy, and the demand for each type of resource is a random integer between 1 and 10. The task level is a random integer between 1 and 3. The experiment set up five groups of UAVs with the numbers 5, 10, 15, 20, and 25. Each UAV’s number of various attack resources is a random integer between 1 and 5. Each group of experiments is carried out 100 times simulation. The reconnaissance range of the UAV is set to be 300 m; the communication range is 600 m; and the communication time delay between UAVs is 0.1 s. In the simulation experiment, the UAVs start from random positions on the west side of the mission edge with different heading angles.
When the number of UAVs changes, different coalition formation algorithms are used to perform tasks. The average time and average task revenue recorded are shown in Figures 12(a) and 12(b), respectively.
Figure 12(a) shows that as the number of UAVs increases, the average time to perform the task for the three construction methods gradually decreases. At the same time, PTCFA has the shortest average time to finish the mission among the three algorithms, regardless of the number of UAVs. It is due to the process of PTCFA, which aims to screen out members that take the shortest time to complete missions. The method proposed in this paper is better than resource welfare algorithms but slightly worse than PTCFA in task completion time.
Figure 12(b) shows that the average avenue obtained by the three coalition formation algorithms for performing missions increases as the number of UAVs increases. When the number of UAVs grows from 15 to 20, the resource welfare algorithm shows a lower growth trend than the other two algorithms. The main reason is that there is no minimum requirement for the number of UAVs during the establishment of the coalition. As shown in Figure 12(b), the algorithms proposed in this paper can obtain higher task revenue than the other two algorithms.
5.2. Performance Comparison of Different Information Transfer Times
The maximum number of information transfers affects the efficiency of the UAV in accomplishing the mission in the communication-constrained environment. In this group of experiments, the fixed maximum information transfer times and AT-MIT are compared.
The mission area, target, and UAV setting are consistent with the previous section. All targets can only be found by any UAV in the experiments once, and UAV resources do not decrease after executing the mission. When the number of UAVs rises, the experiment compares the average number of successful coalition formations and the average time using different information transfer times.
As Figure 13(a) illustrates, the number of completed missions rises as the number of UAVs in the mission area increases. However, when and numbers of UAVs are 5 and 25, there is no significant advantage compared to other information transformation times. Because when the number of UAVs is too small, the distance between UAVs is too far. There is no directly connected UAV, and the number of information transmissions is ineffective; when the number is too large, the number of directly connected UAVs is frequently sufficient to form the coalition successfully. At the same time, AT-MIT is better than but inferior to in the number of completions.
Figure 13(b) demonstrates that when AT-MIT is used for task allocation, the average benefit is greater than for and but less than for . The reason is that increasing the number of information transformations directly results in a longer time required to form the coalition, reducing the benefits of task completion. At the same time, due to the small number of information transmissions and UAVs in the local communication network, it is impossible to choose the optimal one during the coalition formation process.
The method of resource consumption among UAVs affects the robustness of the UAV swarm performing tasks. By varying the G value, experiments analyze the influence of the G value on the performance of the method. And the performance of three resource consumption methods is studied by experiments: the Gini-coefficient-based algorithm, the greedy algorithm, and the balance algorithm.
The 6 UAVs executed the 30 missions in the mission area, and when the UAVs could not complete the task, the experiment was terminated. The mission target needs 3 different types of resources to destroy. The demand for each resource type is a random integer between 1 and 50, and the mission level is a random integer between 1 and 3. Experiments run 100 times, and the tasks are reset for each experiment. When the G value varies from 0 to 1, the average number of tasks completed by the UAV fleet is shown in Figure 13.
As illustrated in Figure 14, when G equals 0.31, the average number of tasks completed reaches the maximum value of 27.8. It is primarily because UAVs in the middle of the mission area receive more coalition formation requests and are required to perform more tasks in the communication-constrained environment. Therefore, in the resource consumption allocation scheme, the UAV in the middle of the mission should maintain a resource advantage over the UAV in the edge position. However, if the resource difference is too big, the UAV’s resources in the edge position will be consumed prematurely, and the corresponding mission will not be completed.
As illustrated in Figure 15, the resource consumption algorithm based on the Gini coefficient completes more tasks on average than the other two resource consumption algorithms. The reason is that the other two algorithms neglect to account that the quantity of tasks performed by UAVs in different positions varies significantly in a communication-constrained environment.
Figure 16(a) shows that the resource consumption algorithm based on balance maintains consistency in the coalition’s UAV resource allocation. UAV1and UAV2 of Figure 16(a) are incompatible with other UAV resources because they have fewer coalitions with other UAVs on missions. As Figure 16(b) show, the algorithm based on the greedy algorithm is similar to the algorithm based on the Gini coefficient (G value is too large), leading to the large difference in resources between UAVs. Simultaneously, we can notice that the algorithm based on the Gini coefficient can maintain a reasonable resource differential between UAVs, which is more beneficial to the execution of the following tasks.
This paper proposes an improved CNP-based method, which solves the heterogeneous multi-UAV task allocation with resource constraints, UAV self-performance constraints, time constraints, and communication constraints. This paper proposes AD-MIT, and the information consistency method to solve conflicts in the communication-constrained environment. Meanwhile, this paper proposes a resource consumption allocation method based on the Gini coefficient to keep the resource difference between UAVs and then improve the ability to deal with unplanned tasks. The experimental results demonstrate that the improved CNP-based method can obtain a higher performance in time and reward the accomplished task accomplished compared to the other two methods. Simultaneously, compared with the setting of a fixed number of information transformations, AD-MIT can rationally utilize communication resources and accelerate coalition formation to ensure the number of tasks completed in the communication-constrained environment. Under resource constraints, the Gini-coefficient-based resource consumption allocation method is better than the greedy method and the balanced method in the number of tasks accomplished. The method can reasonably regulate the degree of resource differentiation between UAVs to increase the swarm’s robustness in a variety of mission scenarios and deal with unplanned tasks.
In future studies, the relationship between the optimal G value and numerous task scenario factors will be further investigated, which is the premise of applying it to practical scenarios, and it is worth continuing to study.
Data can be obtained directly from the article.
Conflicts of Interest
The authors declare that there are no conflicts of interest regarding the publication of this paper.
X. Wang, Z. Liu, Y. Cong, L. Jie, and C. Hao, “Miniature fixed-wing UAV swarms: review and outlook,” Acta Aeronautica et Astronautica Sinica, vol. 41, no. 4, 023732 pages, 2020.View at: Google Scholar
S. Zhu, Y. Zhang, Y. Gao, and W. Fan, “A cooperative task assignment method of multi-UAV based on self organizing map,” in Proceedings of the 2018 International Conference on Cyber-Enabled Distributed Computing and Knowledge Discovery, pp. 4370–4375, Zhengzhou, China, October 2018.View at: Publisher Site | Google Scholar
J. M. Liu, Q. Wang, Y. J. Xu, and L. Cunzhuang, “A novel distributed method for time-critical task allocation problems in multi-UAV system,” in Proceedings of the IEEE International Conference on Communications (ICC), pp. 14–23, Montreal, QC, Canada, June 2021.View at: Publisher Site | Google Scholar
J. J. Acevedo, B. C. Arrue, I. Maza, and A. Ollero, “Distributed approach for coverage and patrolling missions with a team of heterogeneous aerial robots under communication constraints,” International Journal of Advanced Robotic Systems, vol. 10, no. 1, p. 28, 2013.View at: Publisher Site | Google Scholar
J. Xiang, X. Dong, Q. Li, and R. Zhang, “Cooperation target assignment of missiles based on multi-agent technique and improved contract net protocol,” in Proceedings of the 2018 IEEE CSAA Guidance, Navigation and Control Conference (CGNCC), pp. 1–6, Xiamen, China, August 2018.View at: Publisher Site | Google Scholar
W. Peng and X. C. Lu, “On the reduction of broadcast redundancy in mobile ad hoc networks,” in Proceedings of the Workshop on Mobile & Ad Hoc Networking & Computing, pp. 129-130, Boston, MA, USA, August 2000.View at: Google Scholar
Z. Jianhua, “An convenient method to calculate Gini coefficient,” Journal of Shanxi Agricultural University (Society Science Edition): Social Science Edition, vol. 6, no. 3, pp. 275–278, 2007.View at: Google Scholar