Multi-agent cooperative confrontation modeling method and system based on communication behavior tree

Through the multi-agent collaborative adversarial modeling method based on the communication behavior tree, the multi-machine communication problem in the multi-agent collaborative adversarial simulation simulation is solved, and the collaborative decision-making between the agents is realized, which improves the realistic and effective simulation simulation. .

CN118940642BActive Publication Date: 2024-12-06NAT UNIV OF DEFENSE TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411422447.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-10-12
Publication Date
2024-12-06
Estimated Expiration
2044-10-12

AI Technical Summary

Technical Problem

The existing multi-agent collaborative adversarial simulation simulation is difficult to achieve multi-machine communication and information sharing, reducing the realistic and effective adversarial simulation simulation.

Method used

Using a multi-agent collaborative adversarial modeling method based on the communication behavior tree, the single-agent confrontation decision-making behavior tree and the communication collaborative decision-making behavior tree are implemented to realize the sending and receiving of messages between agents and collaborative decision-making, including the long-term and wingmen. between commands and actions are issued.

Benefits of technology

It improves the realistic and effective nature of the adversarial simulation simulation, enhances the reusability of the model and manual modeling efficiency, and improves the intelligence in typical scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940642B_ABST
    Figure CN118940642B_ABST
Patent Text Reader

Abstract

The present invention relates to a multi-agent collaborative confrontation modeling method and system based on a communication behavior tree. The method comprises: constructing a drone confrontation strategy according to a drone confrontation task model; constructing a single-agent confrontation decision behavior tree and a communication collaborative decision behavior tree according to a single-agent confrontation behavior strategy and a multi-agent collaborative confrontation behavior strategy respectively; the communication collaborative decision behavior tree comprises a leader communication collaborative decision behavior tree and a wingman communication collaborative decision behavior tree; the leader communication collaborative decision behavior tree comprises a communication node for sending messages; the wingman communication collaborative decision behavior tree comprises a communication subtree for receiving messages and performing corresponding actions; the agents participating in the drone confrontation task make behavioral decisions according to the single-agent confrontation decision behavior tree or the communication collaborative decision behavior tree. The use of this method can improve the realism and effectiveness of confrontation simulation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of simulation technology, and in particular to a multi-agent collaborative confrontation modeling method and system based on a communication behavior tree. Background Art

[0002] The modeling of the confrontation process of a typical UAV confrontation mission involves the accurate simulation of the physical characteristics of equipment such as aircraft and projectiles, while the simulation of the confrontation process of a typical UAV confrontation mission involves the simulation and analysis of various decision-making processes. The complexity of UAV confrontation actions requires comprehensive consideration of multiple factors such as aircraft performance, projectile performance, radar technology, etc., in order to more realistically simulate the air confrontation environment. However, decision modeling is the core of UAV confrontation simulation, because in the actual environment, the accuracy and speed of decision-making directly affect the direction of the situation, and the ability to make autonomous decisions is the key to the effectiveness of unmanned equipment in confrontation. Through decision modeling, it is possible to evaluate the impact of different decisions on the confrontation results, optimize the confrontation strategy, and improve the effectiveness of air confrontation.

[0003] The aircraft decision modeling methods in typical UAV confrontation tasks mainly include game theory-based methods, knowledge encoding-based methods, and machine learning-based methods. Although the game theory-based methods perform well in some aspects, they have many limitations and perform relatively poorly when dealing with complex and uncertain confrontation environments. On the other hand, although the machine learning-based methods can generate effective strategies, due to their black box nature, the interpretation and understanding of the strategies become quite difficult.

[0004] In this case, the knowledge encoding method is particularly desirable. Among them, the behavior tree-based method is a common knowledge encoding strategy. The behavior tree-based method has significant advantages over other methods and shows strong interpretability and modularity. By establishing a clear behavior tree structure, the decision-making process can be decomposed into simple and manageable parts, making the system easier to understand and adjust. Therefore, the behavior tree-based knowledge encoding method has become a reliable choice in the modeling of UAV confrontation decision-making. At present, two-aircraft formation confrontation has become the main mode of modern confrontation. Compared with the traditional 1v1 confrontation, the two-aircraft formation confrontation requires not only the effective confrontation decision of a single aircraft, but also the coordination of the two aircraft to achieve an overall advantage over the non-self-side. However, the existing multi-agent collaborative confrontation has limitations, and it is difficult to achieve multi-aircraft communication and information sharing, which reduces the realism and effectiveness of confrontation simulation. Summary of the invention

[0005] Based on this, it is necessary to provide a multi-agent collaborative confrontation modeling method and system based on communication behavior trees to address the above technical problems.

[0006] A multi-agent collaborative confrontation modeling method based on a communication behavior tree, the method comprising:

[0007] Constructing a drone confrontation strategy based on a pre-constructed drone confrontation task model; the drone confrontation strategy includes a single-agent confrontation behavior strategy and a multi-agent collaborative confrontation behavior strategy;

[0008] A single-agent confrontation decision behavior tree and a communication collaborative decision behavior tree are constructed according to the single-agent confrontation behavior strategy and the multi-agent collaborative confrontation behavior strategy respectively; the communication collaborative decision behavior tree includes a leader communication collaborative decision behavior tree and a wingman communication collaborative decision behavior tree; the leader communication collaborative decision behavior tree includes a communication node for sending messages; the wingman communication collaborative decision behavior tree includes a communication subtree for receiving messages and performing corresponding actions;

[0009] The intelligent agents participating in the UAV confrontation task make behavioral decisions according to the single-agent confrontation decision behavior tree or the communication and collaborative decision behavior tree.

[0010] In one embodiment, the multi-agent collaborative confrontation behavior strategy includes a collaborative strategy; the collaborative strategy includes the leader sending instructions to instruct the wingman to assist in completing the confrontation.

[0011] In one embodiment, making a behavior decision according to the communication collaborative decision behavior tree includes the leader and the wingman making a behavior decision according to the leader communication collaborative decision behavior tree and the wingman communication collaborative decision behavior tree, respectively.

[0012] In one of the embodiments, the message sent by the communication node includes a Hot command, a Cold command and a cover command sent by the leader to the wingman, and a decision line crossing report sent by the wingman to the leader.

[0013] In one of the embodiments, the search subtree of the leader's communication collaborative decision-making behavior tree creates a confrontation target direction and searches for the enemy. If attacked by a non-friendly UAV, the avoidance subtree is entered. The avoidance subtree makes an avoidance strategy based on the message of crossing the decision line reported by the wingman. If the distance of the enemy's projectile is less than a threshold, a backward maneuver is performed to avoid; if the distance of the enemy's projectile is greater than the threshold, a cover command is sent to the wingman, and a three-nine maneuver is performed to avoid; if the enemy is found, the attack subtree is entered. The attack subtree makes an attack strategy based on the message of crossing the decision line reported by the wingman; the attack strategy includes: if the distance of our UAV is less than the decision distance, the leader enters the Cold state, turns backward, and orders the wingman to enter the Hot state and face the enemy head-on; if the leader is in the Hot state, and the distance between the wingman and the non-friendly UAV is less than the decision distance, the wingman will enter the Cold state and turn backward. In any state, when the conditions for launching projectiles are met, the UAV launches projectiles and avoids non-friendly projectiles.

[0014] In one of the embodiments, the wingman reports crossing the decision line to the leader; the wingman receives the message from the leader, and if a Cold command is received, the wingman maneuvers backward; if a cover command is received, the wingman searches for the target and carries out a fire strike.

[0015] In one embodiment, the drone confrontation mission model includes mission objectives, action space and state space.

[0016] In one embodiment, the single-agent adversarial behavior strategy includes a search strategy, an attack strategy, and an avoidance strategy.

[0017] In one of the embodiments, it also includes: modeling the top-level adversarial decision-making behavior of the intelligent agent to obtain a corresponding decision-making behavior tree root tree; the root tree is used to describe the abstract adversarial behavior of the intelligent agent; modeling the bottom-level adversarial decision-making behavior of the intelligent agent to obtain a corresponding decision-making behavior tree subtree; the subtree is used to describe a single specific adversarial behavior of the intelligent agent; the root tree and the subtree of the intelligent agent are combined to obtain a single-agent adversarial decision-making behavior tree.

[0018] A multi-agent collaborative confrontation modeling system based on a communication behavior tree, the system comprising:

[0019] A strategy modeling module is used to construct a drone confrontation strategy based on a pre-built drone confrontation task model; the drone confrontation strategy includes a single-agent confrontation behavior strategy and a multi-agent collaborative confrontation behavior strategy;

[0020] A behavior tree construction module, used to construct a single-agent confrontation decision behavior tree and a communication collaborative decision behavior tree according to the single-agent confrontation behavior strategy and the multi-agent collaborative confrontation behavior strategy respectively; the communication collaborative decision behavior tree includes a leader communication collaborative decision behavior tree and a wingman communication collaborative decision behavior tree; the leader communication collaborative decision behavior tree includes a communication node for sending messages; the wingman communication collaborative decision behavior tree includes a communication subtree for receiving messages and performing corresponding actions;

[0021] The confrontation decision module is used for the intelligent agent participating in the UAV confrontation task to make behavior decisions according to the single-agent confrontation decision behavior tree or the communication and collaborative decision behavior tree.

[0022] The above-mentioned multi-agent collaborative confrontation modeling method and system based on the communication behavior tree realizes the behavior of agent task switching by building a decision behavior tree to complete the corresponding task, and realizes the function under the command by designing communication nodes and communication subtrees, and adding communication nodes to the single-agent confrontation decision behavior tree to obtain the leader communication collaborative decision behavior tree; adding the communication subtree to the single-agent confrontation decision behavior tree to obtain the wingman communication collaborative decision behavior tree to realize the command reception and action response, and multi-machine collaborative confrontation can be realized based on the above-mentioned communication behavior tree. The embodiment of the present invention can be analyzed and customized in a field-oriented manner, so that the model has strong reusability, high efficiency of manual modeling, high intelligence of modeling in typical scenarios, and improves the realism and effectiveness of confrontation simulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flowchart of a multi-agent collaborative confrontation modeling method based on a communication behavior tree in one embodiment;

[0024] Figure 2 A schematic diagram of the structure of a single-agent confrontation decision-making behavior tree in one embodiment;

[0025] Figure 3 It is a structural diagram of a leader communication collaborative decision behavior tree in an embodiment;

[0026] Figure 4 A diagram of the architecture of the communication part in one embodiment;

[0027] Figure 5 A schematic diagram of the structure of a communication subtree in one embodiment;

[0028] Figure 6 It is a structural diagram of a wingman communication collaborative decision-making behavior tree in one embodiment;

[0029] Figure 7 It is a structural block diagram of a multi-agent collaborative confrontation modeling system based on a communication behavior tree in one embodiment. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0031] In one embodiment, Figure 1 As shown, a multi-agent collaborative confrontation modeling method based on a communication behavior tree is provided, comprising the following steps:

[0032] Step 102: construct a drone confrontation strategy based on a pre-constructed drone confrontation mission model.

[0033] Drone confrontation strategies include single-agent confrontation behavior strategies and multi-agent collaborative confrontation behavior strategies.

[0034] Step 104, construct a single-agent confrontation decision behavior tree and a communication collaborative decision behavior tree according to the single-agent confrontation behavior strategy and the multi-agent collaborative confrontation behavior strategy respectively.

[0035] The communication collaborative decision behavior tree includes a leader communication collaborative decision behavior tree and a wingman communication collaborative decision behavior tree; the leader communication collaborative decision behavior tree includes a communication node for sending messages; the wingman communication collaborative decision behavior tree includes a communication subtree for receiving messages and performing corresponding actions.

[0036] Step 106, the intelligent agents participating in the drone confrontation task make behavioral decisions based on the single-agent confrontation decision behavior tree or the communication and collaboration decision behavior tree.

[0037] In the above-mentioned multi-agent collaborative confrontation modeling method based on the communication behavior tree, the behavior of agent task switching is realized by building a decision behavior tree to complete the corresponding task. The communication node and the communication subtree are designed, and the communication node is added to the single-agent confrontation decision behavior tree to obtain the leader communication collaborative decision behavior tree to realize the function under the command; the communication subtree is added to the single-agent confrontation decision behavior tree to obtain the wingman communication collaborative decision behavior tree to realize the command reception and action response. Based on the above-mentioned communication behavior tree, multi-machine collaborative confrontation can be realized. The embodiment of the present invention can be analyzed and customized in a field-oriented manner, so that the model has strong reusability, high efficiency of manual modeling, high intelligence of modeling in typical scenarios, and improves the realism and effectiveness of confrontation simulation.

[0038] In one embodiment, the UAV confrontation mission model includes mission objectives, action space and state space. In this embodiment, the mission objectives refer to the geographical area where the typical UAV confrontation mission is located, the confrontation parties, equipment performance, winning and losing conditions and initial state; the action space refers to the basic operations involved in the typical UAV confrontation mission, including moving, adjusting the flight altitude and starting / shutting down the radar. The state space refers to the basic states involved in the typical UAV confrontation mission, including parking, taking off, landing, normal flight, launching, being attacked and crashing.

[0039] Specifically, the drone confrontation mission is described as follows:

[0040] (1) Geographical area: a rectangular area extending from E112° in the west to E120° in the east, from N18° in the south to N24° in the north.

[0041] (2) The two opposing sides: The Red side uses Type A drones, and the Blue side uses Type B drones.

[0042] (3) Equipment performance: The radar and infrared search and tracking systems carried by both sides have basically the same performance. Type A drones have more jamming measures and more advanced electronic countermeasures systems. Type B drones are equipped with projectiles with a longer range.

[0043] (4) Winning and losing conditions

[0044] Victory condition: The Red side wins if its score is regular at the end of the game.

[0045] Failure condition: If the Red team's score is negative at the end of the game, the game fails.

[0046] End conditions: The simulation time exceeds 45 minutes or all drones of one side are destroyed.

[0047] (5) Initial situation

[0048] Location: 370 km away.

[0049] Heading: 0° (due north).

[0050] Speed: 350 knots (Mach 0.61).

[0051] Equipment: 6 medium-range air-to-air projectiles, 2 short-range air-to-air projectiles.

[0052] The action space is designed as follows:

[0053] (1) Mobile

[0054] Description: The drone can move horizontally in any direction.

[0055] Velocity range: v∈[350,f(h)] knots, where f(h) is a function of the altitude h of the drone.

[0056] (2) Adjust the flight altitude

[0057] Description: The drone can adjust its altitude between 20 and 45,000 feet.

[0058] Time: It takes time to adjust the altitude, and the rate of descent is greater than the rate of ascent.

[0059] (3) Launch equipment

[0060] Description: Drones can fire projectiles and machine guns at non-friendly drones.

[0061] Hit probability: depends on factors such as target speed, maneuverability, and the intersection posture of the projectiles.

[0062] (4) Emitting interference

[0063] Description: The drone can launch chaff, infrared countermeasures, or towed decoys.

[0064] Function: Towed decoys and chaff interfere with radar lock, and infrared jammers counter infrared-guided projectiles.

[0065] (5) Enable / disable radar

[0066] Description: Controls the startup status of the airborne radar.

[0067] Effect: Activate to improve search and attack efficiency, but may expose the position.

[0068] The state space design is as follows:

[0069] (1) On Ground: The drone is parked at a parking space for maintenance and refueling.

[0070] (2) Taking Off: The drone is in the process of taking off and is vulnerable to attack.

[0071] (3) Landing: The drone is in the process of landing and is vulnerable to attack.

[0072] (4) Normal flight (Cruising): The UAV performs routine flight operations.

[0073] (5) Firing: The drone is executing the launching action.

[0074] (6) Under Attack: The drone is under attack from a non-friendly party and countermeasures must be taken.

[0075] (7) Destroyed: The drone is destroyed and cannot be operated.

[0076] In one embodiment, the single-agent adversarial behavior strategy includes a search strategy, an attack strategy, and an avoidance strategy.

[0077] In this embodiment, the search strategy refers to the two-aircraft formation using radar to search in the direction of the confrontation target according to the instructions of the superior. If the drone finds that it is under attack through the radar warning receiver without finding any non-friendly drones, or finds non-friendly throwing objects through the radar, it will set the direction of the attack as the direction of the confrontation target, and search in that direction after completing the avoidance; the attack strategy refers to the drone pushing the throttle to the maximum, but not starting the afterburner, and trying to increase the flight altitude to increase its own energy. At the same time, the drone will launch throwing objects at the farthest distance possible, forcing the non-friendly drone to evade, thereby obtaining energy advantage and angle advantage; the avoidance strategy refers to the formation trying to be outside the attack envelope of the non-friendly drone at the beginning of the confrontation. After the medium-range missiles of the non-friendly drone are exhausted, the distance will be shortened, but the confrontation will not be carried out within the visual range as much as possible, so as to ensure the survival rate while improving the hit rate of the own throwing objects.

[0078] The purpose of the confrontation is to make non-friendly targets enter the area where the probability of being hit by our equipment system is higher, while avoiding being in the danger zone as much as possible.

[0079] Specifically, this paper selects a two-aircraft formation as the minimum execution unit, takes the Red side's penetration mission as the goal, and designs the Red two-aircraft confrontation strategy: First, set the decision distance, which can be modified according to the experimental needs. For example, setting the distance outside the enemy's projectile attack envelope can keep the friendly UAV away from non-friendly UAVs and improve the range advantage of our projectiles. Setting the distance to within the line of sight may allow the UAV to enter the line of sight confrontation with non-friendly UAVs.

[0080] During the search phase, the formation will search in the direction of the counter target indicated by the superior. If the drone detects that it is under attack through the radar warning receiver without detecting any non-friendly drones, or detects non-friendly projectiles through the radar, it will set the direction of the attack as the direction of the counter target and search in that direction after completing the evasion.

[0081] After discovering a non-friendly drone, it will enter the attack phase. The red drone will push the throttle to the maximum, but will not start the afterburner, and will try to increase the flight altitude to increase its own energy. At the same time, the drone will launch projectiles at the farthest distance possible, forcing non-friendly drones to evade, thereby gaining energy and angle advantages. If the leader's distance from the non-friendly drone is less than the decision distance, the leader will switch to the "cold" state, turn back, and order the wingman to switch to the "hot" state and face the enemy head-on. If the leader is in the "hot" state and the wingman's distance from the non-friendly drone is less than the decision distance, the wingman will switch to the "cold" state and turn back. In any state, as long as the projectile launch conditions are met, the drone will launch projectiles and will try to avoid non-friendly projectiles.

[0082] In one embodiment, the steps of constructing a single-agent adversarial decision-making behavior tree according to a single-agent adversarial behavior strategy include: modeling the top-level adversarial decision-making behavior of the agent to obtain a corresponding decision-making behavior tree root tree; the root tree is used to describe the abstract adversarial behavior of the agent; modeling the bottom-level adversarial decision-making behavior of the agent to obtain a corresponding decision-making behavior tree subtree; the subtree is used to describe a single specific adversarial behavior of the agent; and the root tree and subtree of the agent are combined to obtain a single-agent adversarial decision-making behavior tree.

[0083] In this embodiment, if Figure 2As shown, a schematic diagram of the structure of a single-agent confrontation decision-making behavior tree is provided. The single-agent confrontation decision-making behavior tree constructs subtrees through sequence nodes, selection nodes, condition nodes, and action nodes to achieve confrontation decision-making behavior switching and complete enemy search, approaching non-friendly attacks, and avoiding attacks. In the basic structure of this single-agent confrontation decision-making behavior tree, the root node of the behavior tree is a sequence node, and its subtrees represent search actions, attack actions, and avoidance actions, respectively. They are executed in order from left to right. When a subtree fails to run, other subtrees on the right are no longer executed, and the left subtree has a higher priority than the right.

[0084] In the search subtree, the Red UAV determines whether to set the confrontation target direction by executing the conditional node. If there is no enemy, a random angle is used as the confrontation target direction. The Red UAV searches in the confrontation target direction and maintains a high energy state during the search to occupy energy and angle advantages as soon as it approaches the non-friendly party. When the Red UAV is attacked during the search process, it enters the avoidance subtree and determines whether the distance of the enemy's projectile is less than the threshold by executing the conditional node. If the distance of the enemy's projectile is less than the threshold, the confrontation target direction is set and the backward maneuver is used to avoid; if the distance of the enemy's projectile is greater than the threshold, the confrontation target direction is set and the three-nine maneuver is used to avoid. When the Red UAV is found during the search process, it enters the attack subtree and determines whether the target non-friendly drone has been attacked by executing the node. If the target non-friendly drone has not been attacked, the projectile is launched; if the target non-friendly drone has been attacked, the projectile is continuously guided.

[0085] Specifically, in terms of the operation process of the Red Team UAV behavior tree, the selector first selects the behavior through the conditional node to realize the state switching and selection of the behavior tree. The various parameters of the UAV are input into the behavior tree, and after the corresponding nodes are executed and the state parameters are changed, the corresponding decision is made according to the latest state of the current UAV and the confrontation environment information to execute the corresponding action, and the state information of the UAV (position, direction, number of thrown objects, etc.) is updated, and then the selection-execution process is cyclically carried out until the end.

[0086] In one embodiment, the multi-agent collaborative confrontation behavior strategy includes a collaborative strategy; the collaborative strategy includes the leader sending instructions to instruct the wingman to assist in completing the confrontation.

[0087] In one embodiment, making behavioral decisions according to the communication collaborative decision behavior tree includes the leader and the wingman making behavioral decisions according to the leader communication collaborative decision behavior tree and the wingman communication collaborative decision behavior tree, respectively. In this embodiment, compared with the single-agent confrontation decision behavior tree, the communication collaborative decision behavior tree realizes the collaborative confrontation of different agents by adding communication action nodes and communication subtrees. In the leader communication collaborative decision behavior tree, the command to the wingman can be issued by specially customizing the communication node to complete the multi-machine collaborative confrontation.

[0088] In one embodiment, the messages sent by the communication node include Hot commands, Cold commands and cover commands sent by the leader to the wingman, and a report of crossing the decision line sent by the wingman to the leader. In this embodiment, Hot command: the leader sends a Hot command to instruct the wingman to switch to the "Hot" state, so that the red team formation will confront according to the UAV countermeasure strategy. Cold command: the leader sends a Cold command to instruct the wingman to switch to the "Clod" state, so that the red team formation will confront according to the UAV countermeasure strategy. Cover command: the leader sends a cover command to instruct the wingman to cover himself to avoid projectiles, or guide himself to launch projectiles. Report of crossing the decision line: the wingman reports that he has crossed the decision line to assist the leader in understanding the status of the wingman, making correct decisions, and executing the UAV countermeasure strategy.

[0089] In one embodiment, Figure 3 The structure diagram of the leader communication collaborative decision behavior tree shown in FIG. The behavior decision made according to the leader communication collaborative decision behavior tree includes: the search subtree of the leader communication collaborative decision behavior tree creates a confrontation target direction and searches for the enemy situation. If attacked by a non-friendly drone, the avoidance subtree is entered. The avoidance subtree makes an avoidance strategy based on the message of crossing the decision line reported by the wingman. If the distance of the enemy's projectile is less than the threshold, the wingman performs a backward maneuver to avoid; if the distance of the enemy's projectile is greater than the threshold, a cover command is sent to the wingman, and a three-nine maneuver is performed to avoid; if the enemy situation is found, the wingman performs a three-nine maneuver to avoid. Enter the attack subtree, and the attack subtree makes an attack strategy according to the message of crossing the decision line reported by the wingman; the attack strategy includes: if the distance of our drone is less than the decision distance, the leader aircraft will enter the Cold state, turn back, and order the wingman to enter the Hot state and face the enemy head-on; if the leader aircraft is in the Hot state, and the distance between the wingman and the non-our drone is less than the decision distance, the wingman will enter the Cold state and turn back; in any state, when the conditions for launching projectiles are met, the drone launches projectiles and avoids non-our projectiles. In this embodiment, in the leader aircraft communication collaborative decision behavior tree, the default tree structure is executed from left to right, the root node is a sequential node, the rightmost subtree creates a confrontation target direction, searches for the enemy, and enters the avoidance subtree if attacked by a non-our drone; if the enemy is found, enter the attack subtree.

[0090] In one embodiment, the behavior decision making based on the wingman communication collaborative decision behavior tree includes: the wingman reports crossing the decision line to the leader; the wingman receives the message from the leader, and if a Cold command is received, the wingman maneuvers backwards; if a cover command is received, the wingman searches for the target and conducts a fire strike.

[0091] like Figure 4As shown in the communication part architecture diagram, the wingman communication collaborative decision behavior tree realizes the reception and implementation of the leader's instructions by adding a communication subtree. The communication subtree consists of conditional nodes and action nodes, which can receive the message sent by the leader, parse and judge the message, and take corresponding actions. Figure 5 The structure diagram of the communication subtree is shown in the figure. The root node of the communication subtree is a sequential node, and the subnodes are executed sequentially from left to right. If a Cold command is received, the wingman will maneuver backwards; if a cover command is received, the target will be searched for and a fire attack will be carried out; and the message of crossing the decision line will be reported according to the current status. Compared with the single-machine confrontation behavior tree, the wingman communication collaborative decision-making behavior tree can complete the command reception by specially customizing the communication subtree.

[0092] like Figure 6 The structure diagram of the wingman communication collaborative decision-making behavior tree shown in the figure, in which the wingman will first communicate, receive the leader's instructions, report its own situation, and then cooperate with the leader to confront according to the leader's instructions and actual situation. The tree structure of the wingman communication collaborative decision-making behavior tree is executed from left to right. The root node is the sequential node, and the rightmost subtree is the communication subtree, which receives and parses messages and reports its own situation to the leader. The second subtree is the search subtree, which searches for targets in the direction of the confrontation target; the third subtree is the attack subtree to complete the enemy attack task; the fourth subtree is the avoidance subtree, which avoids projectiles. After the wingman communication collaborative decision-making behavior tree is constructed, the wingman can independently decide more actions during the confrontation, and no longer rely solely on the orders issued by the leader, such as avoiding non-friendly projectiles. In actual confrontation, the drone will obviously not avoid non-friendly attacks because it has not received an avoidance order. Instead, it needs to issue an order to achieve the confrontation purpose without avoiding non-friendly attacks.

[0093] When the lead aircraft receives a report of crossing the decision line, it switches to "Cold" status and issues corresponding instructions; when the wingman receives a Cold instruction, it executes a backward maneuver; when the wingman receives a Hot instruction, it executes a head-on maneuver to meet the enemy; when the wingman receives a cover instruction, it searches for the enemy and opens fire to cover the leader aircraft.

[0094] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0095] In one embodiment, Figure 7 As shown, a multi-agent collaborative confrontation modeling system based on a communication behavior tree is provided, comprising:

[0096] The strategy modeling module 702 is used to construct a drone confrontation strategy based on a pre-built drone confrontation task model; the drone confrontation strategy includes a single-agent confrontation behavior strategy and a multi-agent collaborative confrontation behavior strategy;

[0097] The behavior tree construction module 704 is used to construct a single-agent confrontation decision behavior tree and a communication collaborative decision behavior tree according to the single-agent confrontation behavior strategy and the multi-agent collaborative confrontation behavior strategy respectively; the communication collaborative decision behavior tree includes a leader communication collaborative decision behavior tree and a wingman communication collaborative decision behavior tree; the leader communication collaborative decision behavior tree includes a communication node for sending messages; the wingman communication collaborative decision behavior tree includes a communication subtree for receiving messages and executing corresponding actions;

[0098] The confrontation decision module 706 is used for the intelligent agent participating in the drone confrontation task to make behavior decisions according to the single-agent confrontation decision behavior tree or the communication and collaborative decision behavior tree.

[0099] In one embodiment, the multi-agent collaborative confrontation behavior strategy includes a collaborative strategy; the collaborative strategy includes the leader sending instructions to instruct the wingman to assist in completing the confrontation.

[0100] In one embodiment, making a behavior decision according to the communication collaborative decision behavior tree includes the leader and the wingman making a behavior decision according to the leader communication collaborative decision behavior tree and the wingman communication collaborative decision behavior tree, respectively.

[0101] In one embodiment, the message sent by the communication node includes a Hot command, a Cold command, and a Cover command sent by the leader to the wingman, and a Decision Line Crossing Report sent by the wingman to the leader.

[0102] In one of the embodiments, the search subtree of the leader's communication collaborative decision-making behavior tree creates a confrontation target direction and searches for the enemy. If attacked by a non-friendly UAV, the avoidance subtree is entered. The avoidance subtree makes an avoidance strategy based on the message of crossing the decision line reported by the wingman. If the distance of the enemy's projectile is less than a threshold, a backward maneuver is performed to avoid; if the distance of the enemy's projectile is greater than the threshold, a cover command is sent to the wingman, and a three-nine maneuver is performed to avoid; if the enemy is found, the attack subtree is entered. The attack subtree makes an attack strategy based on the message of crossing the decision line reported by the wingman; the attack strategy includes: if the distance of our UAV is less than the decision distance, the leader switches to the Cold state, turns backward, and orders the wingman to switch to the Hot state to meet the enemy head-on; if the leader is in the Hot state and the distance between the wingman and the non-friendly UAV is less than the decision distance, the wingman switches to the Cold state and turns backward; in any state, when the conditions for launching projectiles are met, the UAV launches projectiles and avoids non-friendly projectiles.

[0103] In one of the embodiments, the wingman reports crossing the decision line to the leader; the wingman receives the message from the leader, and if a Cold command is received, the wingman maneuvers backward; if a cover command is received, the wingman searches for the target and carries out a fire strike.

[0104] In one embodiment, the drone confrontation mission model includes a mission goal, an action space, and a state space.

[0105] In one embodiment, the single-agent adversarial behavior strategy includes a search strategy, an attack strategy, and an avoidance strategy.

[0106] In one of the embodiments, it is also used to model the top-level adversarial decision-making behavior of the intelligent agent to obtain the corresponding decision-making behavior tree root tree; the root tree is used to describe the abstract adversarial behavior of the intelligent agent; the bottom-level adversarial decision-making behavior of the intelligent agent is modeled to obtain the corresponding decision-making behavior tree subtree; the subtree is used to describe the single specific adversarial behavior of the intelligent agent; the root tree and subtree of the intelligent agent are combined to obtain a single-agent adversarial decision-making behavior tree.

[0107] For the specific limitations of the multi-agent collaborative confrontation modeling system based on the communication behavior tree, please refer to the limitations of the multi-agent collaborative confrontation modeling method based on the communication behavior tree above, which will not be repeated here. Each module in the above-mentioned multi-agent collaborative confrontation modeling system based on the communication behavior tree can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0108] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0109] The above-mentioned embodiments only express several implementation modes of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the attached claims.

Claims

1. A multi-agent collaborative confrontation modeling method based on communication behavior tree, characterized in that: The method comprises: Constructing a drone confrontation strategy based on a pre-constructed drone confrontation task model; the drone confrontation strategy includes a single-agent confrontation behavior strategy and a multi-agent collaborative confrontation behavior strategy; A single-agent confrontation decision behavior tree is constructed according to the single-agent confrontation behavior strategy, and a communication collaborative decision behavior tree is constructed according to the single-agent confrontation decision behavior tree and the multi-agent collaborative confrontation behavior strategy; the multi-agent collaborative confrontation behavior strategy includes a collaborative strategy; the collaborative strategy includes the leader sending instructions to instruct the wingman to assist in completing the confrontation; the communication collaborative decision behavior tree includes a leader communication collaborative decision behavior tree and a wingman communication collaborative decision behavior tree; the leader communication collaborative decision behavior tree includes a communication node for sending messages; the wingman communication collaborative decision behavior tree includes a communication subtree for receiving messages and performing corresponding actions; The intelligent agents participating in the UAV confrontation task make behavioral decisions according to the single-agent confrontation decision behavior tree and the communication and coordination decision behavior tree; The step of constructing a single-agent adversarial decision-making behavior tree according to the single-agent adversarial behavior strategy comprises: Modeling the top-level confrontation decision-making behavior of the intelligent agent to obtain a corresponding decision-making behavior tree root tree; the root tree is used to describe the abstract confrontation behavior of the intelligent agent; Modeling the underlying confrontation decision-making behavior of the intelligent agent to obtain a corresponding decision-making behavior tree subtree; the subtree is used to describe a single specific confrontation behavior of the intelligent agent; The root tree and subtrees of the agent are combined to obtain a single-agent adversarial decision-making behavior tree.

2. The method according to claim 1, characterized in that Making behavior decisions according to the communication collaborative decision behavior tree includes the leader and the wingman making behavior decisions according to the leader communication collaborative decision behavior tree and the wingman communication collaborative decision behavior tree, respectively.

3. The method according to claim 2, characterized in that The messages sent by the communication node include Hot commands, Cold commands and cover commands sent by the leader to the wingman, and decision line crossing reports sent by the wingman to the leader.

4. The method according to claim 3, characterized in that Behavior decisions based on the leader communication collaborative decision-making behavior tree include: The search subtree of the leader's communication collaborative decision-making behavior tree creates a confrontation target direction and conducts enemy search. If attacked by a non-friendly drone, it enters the avoidance subtree. The avoidance subtree makes an avoidance strategy based on the message of crossing the decision line reported by the wingman. If the distance of the enemy's projectile is less than the threshold, it will maneuver backward to avoid; if the distance of the enemy's projectile is greater than the threshold, it will send a cover command to the wingman and perform a three-nine maneuver to avoid. If the enemy is found, the attack subtree is entered, and the attack subtree makes an attack strategy according to the message of crossing the decision line reported by the wingman; the attack strategy includes: if the distance of our drone is less than the decision distance, the leader aircraft switches to the Cold state, turns back, and orders the wingman to switch to the Hot state to meet the enemy head-on; if the leader aircraft is in the Hot state and the distance between the wingman and the non-friend drone is less than the decision distance, the wingman aircraft switches to the Cold state and turns back; in any state, when the conditions for launching projectiles are met, the drone launches projectiles and avoids non-friend drones.

5. The method according to claim 3, characterized in that: Behavior decisions based on the wingman communication collaborative decision-making behavior tree include: The wingman reports crossing the decision line to the leader; The wingman receives messages from the leader. If it receives a Cold command, it will maneuver backwards. If it receives a cover command, it will search for the target and launch a fire attack.

6. The method according to claim 1, characterized in that The UAV confrontation mission model includes mission objectives, action space and state space.

7. The method according to claim 1, characterized in that The single-agent adversarial behavior strategy includes a search strategy, an attack strategy and an avoidance strategy.

8. A multi-agent collaborative confrontation modeling system based on communication behavior tree, characterized in that: The system comprises: A strategy modeling module is used to construct a drone confrontation strategy based on a pre-built drone confrontation task model; the drone confrontation strategy includes a single-agent confrontation behavior strategy and a multi-agent collaborative confrontation behavior strategy; A behavior tree construction module is used to construct a single-agent confrontation decision behavior tree according to the single-agent confrontation behavior strategy, and to construct a communication collaborative decision behavior tree according to the single-agent confrontation decision behavior tree and the multi-agent collaborative confrontation behavior strategy; the multi-agent collaborative confrontation behavior strategy includes a collaborative strategy; the collaborative strategy includes the leader sending instructions to instruct the wingman to assist in completing the confrontation; the communication collaborative decision behavior tree includes a leader communication collaborative decision behavior tree and a wingman communication collaborative decision behavior tree; the leader communication collaborative decision behavior tree includes a communication node for sending messages; the wingman communication collaborative decision behavior tree includes a communication subtree for receiving messages and performing corresponding actions; A confrontation decision module, used for an agent participating in a drone confrontation task to make a behavior decision according to the single agent confrontation decision behavior tree and the communication and coordination decision behavior tree; The behavior tree construction module is also used to model the top-level confrontation decision-making behavior of the intelligent agent to obtain the corresponding decision-making behavior tree root tree; the root tree is used to describe the abstract confrontation behavior of the intelligent agent; the bottom-level confrontation decision-making behavior of the intelligent agent is modeled to obtain the corresponding decision-making behavior tree subtree; the subtree is used to describe a single specific confrontation behavior of the intelligent agent; the root tree and the subtree of the intelligent agent are combined to obtain a single-agent confrontation decision-making behavior tree.