Method for constructing threat target confrontation strategy behavior tree for block defense scene

Abstract

The invention relates to a block defense scene-oriented threatening target confrontation strategy behavior tree construction method, and solves the problem of lack of a behavior tree construction method capable of accurately mapping threatening features and fusing multi-source battlefield data at present. The confrontation strategy behavior tree is divided into a strategy layer, a task layer and an action layer from top to bottom, all the layers cooperatively operate to finally output decision behaviors, and in the operation process of the three strategy layers, preset rules constrain operation of all the layers; the intelligent agents in the scene are divided into a red party and a blue party, each of the red party and the blue party is provided with a plurality of attack nodes, defense nodes and detection nodes, each party is provided with an important node, the rest nodes are general nodes, the red party and the blue party carry out confrontation training by depending on favorable terrains and hidden positions, and confrontation strategy behavior trees of the nodes are recombined; and forming a threat target confrontation strategy behavior tree facing the block defense scene. The method has high efficiency and flexibility, and the complexity of overall decision making is reduced.

Description

Technical Field

[0001] This invention belongs to the field of unmanned equipment testing and evaluation, specifically involving a method for constructing a threat target countermeasure strategy behavior tree for urban defense scenarios. Background Technology

[0002] In modern urban warfare scenarios, urban blocks, as complex three-dimensional combat spaces, are characterized by dense buildings, intricate pathways, and dynamically changing environments, facing diverse military threats such as enemy infiltration, surprise fire attacks, and equipment harassment. Traditional urban block defense command models rely on manual assessment of the battlefield situation, resulting in long decision-making chains, rigid response strategies, and difficulty in quickly adapting to dynamic enemy situations, severely restricting the response efficiency and tactical flexibility of defensive operations. With the deepening development of intelligent warfare concepts, building a dynamic and adaptive threat countermeasure decision-making system has become a core requirement for improving the effectiveness of urban warfare experiments. Behavior trees, with their modular structure, visual logic, and dynamic task scheduling capabilities, have demonstrated significant advantages in military simulation and unmanned equipment control. However, currently, there is a lack of behavior tree construction methods that can accurately map threat characteristics and integrate multi-source battlefield data, addressing the unique characteristics of urban block defense in urban warfare experiments. This makes it difficult to support real-time adversarial decision-making in complex tactical scenarios, and technological innovation is urgently needed to fill this gap. Summary of the Invention

[0003] In the complex decision-making scenario of urban warfare, an adversarial strategy behavior tree is used to model the hierarchical decision-making process. Knowledge engineering is then employed to build the underlying decision-making logic, thereby enabling reasoning capabilities for decision-making during the adversarial process. Specifically, the top-level decision-making mechanism is modeled using the adversarial strategy behavior tree to generate and allocate various tasks. At the lower level, modular design and manually constructed logical rules effectively reduce search overhead during the decision-making process. By modeling the top-level logic using the adversarial strategy behavior tree and flexibly invoking the underlying win / loss rule base, a decision-making method that combines efficiency and flexibility is ultimately formed, thereby reducing the overall complexity of decision-making.

[0004] In the adversarial strategy behavior tree architecture, agents can flexibly express diverse planning schemes and decision-making logic based on various conditions and real-time states in the operating environment, combined with different control nodes and tree structures. This allows the originally static tree structure to realize various dynamic reactive policy responses. The adversarial strategy behavior tree enhances the flexibility of policy representation with its rich node types, and achieves real-time monitoring of the policy execution process through the feedback mechanism of node states. Different adversarial strategy behavior trees can be called by other behavior trees as subtrees. This feature further strengthens its modularity advantage. Utilizing this modularity and reusability, agents for adversarial scenarios can be quickly built. In addition, to ensure the normal operation of the adversarial strategy behavior tree, an underlying win / loss rule library needs to be built as a calling support; and to reduce the search scope of the adversarial strategy behavior tree, the decision output needs to be constrained by rules.

[0005] The adversarial strategy behavior tree can be divided into three layers from top to bottom: strategy layer, task layer, and action layer. Each layer works together to ultimately output decision-making behavior. Furthermore, during the operation of these three strategy layers, preset rules are needed to constrain the operations of each layer.

[0006] 1) Countermeasure Strategy Behavior Tree Strategy Layer: This layer first collects situational data, then analyzes the data to formulate expected operational objectives, and decomposes the overall task in the confrontation process into several more specific sub-tasks. Specifically, the design of the top-level strategy layer covers sub-behavior tree modules such as situational analysis, task planning, doctrine design, and task updates.

[0007] 2) Countermeasure Strategy Behavior Tree Task Layer: This layer primarily focuses on decision-making and planning based on the output of the strategy layer. The sub-nodes of the strategy layer correspond to a series of tasks in the task layer. The task layer refines the deployment plan of the strategy layer, expanding its content into a specific sequence of combat tasks. It also assigns corresponding combat tasks to each combat node and completes preliminary target allocation. This level clarifies the participating units, the specific tasks to be performed, and the target allocation, but it does not involve the planning of specific behaviors such as aircraft maneuvers and attacks in air combat.

[0008] 3) Adversarial Strategy Behavior Tree Action Layer: In this layer, the focus is on designing the leaf nodes of the behavior tree. When the execution flow reaches a node at this level, it will trigger the execution of a series of atomic actions.

[0009] For the scenario of urban defense operations, the adversarial strategy behavior tree nodes can be divided into attack nodes, defense nodes, and detection nodes. The strategy layer, task layer, and action layer of each node are set according to different node attributes and purposes.

[0010] Attack node adversarial strategy behavior tree such as Figure 1 As shown.

[0011] Attack Node Strategy Layer: Attack nodes destroy important enemy nodes and other nodes.

[0012] Attack Node Task Layer: Under normal circumstances, all attack nodes converge on key enemy nodes, aiming to destroy them; this also serves to protect themselves from being preemptively eliminated by the enemy agent. All attack nodes can attack enemy nodes, but prioritize key enemy nodes.

[0013] Attack node action layer: 1) Move towards key enemy nodes unless special circumstances occur; 2) When attacked, move towards your own defensive nodes to seek protection and fire support, with the primary goal of protecting yourself; 3) If currently attacking a key enemy node, remain stationary. Even under attack, it will not approach a defensive node; destroying key enemy nodes is a higher priority, even if it means mutual destruction.

[0014] Defense node adversarial strategy behavior tree such as Figure 2 As shown.

[0015] Defense Node Strategy Layer: Defense nodes protect friendly units.

[0016] Defense Node Task Layer: Unable to attack important enemy nodes, therefore not serving as the main damage dealer for the team. At the same time, when friendly units are detected, they actively protect the team's forces, providing shelter for friendly attack and detection nodes.

[0017] Defense node action layer: 1) Unless special circumstances occur, move towards important enemy nodes to escort our attack nodes closer to important enemy nodes; 2) When a friendly combat node is detected to be low on health and requires protection, actively move to that node to protect friendly forces. Due to the importance of attack nodes in the overall mission, prioritize the protection of protected nodes, giving priority to important nodes, followed by attack and detection nodes. Defensive nodes are not on the list of nodes requiring protection.

[0018] Detection node adversarial strategy behavior tree such as Figure 3 As shown.

[0019] Detection node strategy layer: detect enemy target attributes and location.

[0020] Detection Node Task Layer: Detection nodes only have detection functions and mainly cooperate with defense nodes. When an enemy node is detected, it notifies the attacking node to attack. At the same time, in order to protect itself, it actively keeps the detected enemy units within the entire detection range and maintains the distance from the enemy. When attacked, it moves to the nearest friendly defense node for protection.

[0021] Detection node action layer: 1) Unless there are special circumstances, move towards important enemy nodes to provide vision of friendly combat nodes.

[0022] 2) If an enemy node is detected, maintain a certain distance to keep the enemy node within your detection range, record the coordinates of the enemy combat node, and hand them over to the attacking node to attack. 3) When attacked, move towards your own defensive nodes to seek protection and fire support, with the primary goal of protecting yourself.

[0023] The execution logic of the adversarial strategy behavior tree is as follows: At fixed time intervals, the execution process starts from the root node, sequentially traversing its child nodes according to the operating rules of the adversarial strategy behavior tree nodes. This execution signal continues to propagate according to the specific propagation rules of the node type it reaches, until it reaches a leaf node. When the execution process reaches a leaf node, the corresponding atomic behavior is triggered and acts on the environment, while simultaneously feeding back the execution status of that behavior to its parent node. By searching and traversing the tree structure based on the type of each node and its return status, the adversarial strategy behavior tree can execute different actions according to the structural characteristics and state changes of the tree.

[0024] In the entire execution process of the adversarial strategy behavior tree, communication and data sharing between nodes are crucial, meaning each node can access all the data in the system. The adversarial strategy behavior tree utilizes an external "blackboard" mechanism to achieve this communication and data sharing. This "blackboard" functions like a short-term data storage center. On one hand, it stores the specific model data required for the abstract task, making the construction of the adversarial strategy behavior tree more concise and powerful. On the other hand, the "blackboard" can save intermediate data generated by the operating nodes during operation, providing a communication path for other nodes to access and use this data. Therefore, the "blackboard" is an indispensable component in building the adversarial strategy behavior tree controller. The traversal process of the adversarial strategy behavior tree is essentially the process of forming decision-making behavior. Its ultimate goal is to determine the executable actions based on the node state feedback and the execution status of the behavior tree, thereby completing a complete decision.

[0025] This invention designs a method for constructing a threat target adversarial strategy behavior tree for urban defense scenarios, forming agent adversarial logic and behavior rules in urban defense scenarios, designing adversarial behavior trees for attack, defense, and detection nodes, and strictly executing them according to the adversarial strategy behavior tree method, thereby improving the efficiency of agent adversarial training. Attached Figure Description

[0026] Figure 1 A diagram illustrating the attack node's adversarial strategy behavior tree; Figure 2 Diagram of the behavior tree of the defense node's adversarial strategy; Figure 3 A schematic diagram of the behavior tree of the detection node's adversarial strategy; Figure 4 A schematic diagram of threat target countermeasure strategy behavior tree for street defense scenarios. Detailed Implementation

[0027] This invention proposes a method for constructing a threat target adversarial strategy behavior tree for urban street defense scenarios. Specifically, this method integrates behavior tree theory, reinforcement learning technology, intelligent decision-making algorithms, and data fusion and processing techniques. By constructing a behavior tree model, it enables intelligent and efficient adversarial decision-making for various threat targets in urban street defense scenarios, providing advanced technical support and decision-making basis for urban combat test adversarial systems.

[0028] A street defense combat scenario is designed. Within the scenario, agents are divided into red and blue teams. Each side has several attack nodes, defense nodes, and detection nodes, with one critical node assigned to each side and the rest being general nodes. The red and blue teams conduct adversarial training using advantageous terrain and concealed positions. The adversarial strategy behavior trees for both sides are arranged as follows: Figure 1-3 The execution shown reorganizes the adversarial strategy behavior trees of each node, forming a threat target adversarial strategy behavior tree for street defense scenarios, as follows: Figure 4 As shown.

[0029] The specific execution logic of the adversarial strategy behavior tree is as follows: the execution process is started from the root node at fixed time intervals. The child nodes of the adversarial strategy behavior tree are traversed from top to bottom and from left to right. The execution signal will be executed according to the priority of the node reached. When the node action cannot be completed or the execution condition is not met during the execution process, the next node is executed. When the entire behavior tree is executed or cannot continue to be executed, the execution stops.

[0030] The outcome is determined when one side's key nodes are destroyed or all enemies are wiped out. Before each round of combat begins, all agents are in their initial positions, and the maximum number of moves per round is set. Both the red and blue sides act according to a predetermined strategy described by the adversarial strategy behavior tree. The red side wins if it destroys a key blue node or eliminates all enemies before the maximum number of moves is reached; the blue side wins if the red side fails to destroy a key blue node within the maximum number of moves, or if all red attack nodes are eliminated before the maximum number of moves is reached.

Claims

1. A method for constructing a threat target adversarial strategy behavior tree for street defense scenarios, characterized by: The adversarial strategy behavior tree is divided into a strategy layer, a task layer, and an action layer from top to bottom. Each layer works together to ultimately output a decision behavior. During the operation of these three strategy layers, preset rules constrain the operations of each layer. The intelligent agents in the scenario are divided into a red team and a blue team. Each side sets up several attack nodes, defense nodes, and detection nodes. Each side also sets up one important node, and the rest are general nodes. The red team and the blue team conduct adversarial training by relying on favorable terrain and concealed positions. The adversarial strategy behavior trees of each node are reorganized to form a threat target adversarial strategy behavior tree for the street defense scenario.

2. The method for constructing a threat target adversarial strategy behavior tree for a street defense scenario according to claim 1, characterized in that: Attack Node Strategy Layer: Attack nodes destroy critical enemy nodes and other nodes; Attack Node Task Layer: Under normal circumstances, all attack nodes will move towards important enemy nodes with the intention of destroying them; on the other hand, they will protect themselves from being eliminated by enemy agents in advance. All attack nodes can attack enemy nodes, and important enemy nodes will be attacked first. Attack node action layer: Move towards important enemy nodes unless special circumstances occur; When attacked, move towards your own defensive nodes to seek protection and fire support, prioritizing self-preservation; if currently attacking an important enemy node, remain stationary and do not approach the defensive node even when attacked, prioritizing the destruction of important enemy nodes, even if it means mutual destruction. Defense node strategy layer: Defense nodes protect friendly units; Defense Node Task Layer: Cannot attack important enemy nodes, does not act as the main output unit of the friendly forces, and actively protects friendly forces when friendly units are detected, providing shelter for friendly attack nodes and detection nodes. Defense node action layer: Unless special circumstances occur, move towards important enemy nodes to escort friendly attack nodes to important enemy nodes; when friendly combat nodes are detected to have low health and need protection, actively move towards the combat nodes to protect friendly forces. Distinguish the protection priority of protected nodes, prioritize the protection of important nodes, followed by attack nodes and detection nodes. Defense nodes are not on the list of nodes that need protection. Detection node strategy layer: Detects enemy target attributes and location; Detection Node Task Layer: Detection nodes only have detection functions. They cooperate with defense nodes to notify attacking nodes when enemy nodes are detected. At the same time, in order to protect themselves, they actively keep detected enemy units within the entire detection range and maintain distance from the enemy. When attacked, they move to the nearest friendly defense node for protection. Detection node action layer: If no special circumstances occur, move towards important enemy nodes to provide vision for friendly combat nodes; if an enemy node is detected, maintain a certain distance to keep the enemy node within your detection range, record the coordinates of the enemy combat node, and hand them over to the attacking node; when attacked, move towards friendly defensive nodes to seek protection and fire support, with the primary goal of protecting yourself.

3. The method for constructing a threat target adversarial strategy behavior tree for a street defense scenario according to claim 1, characterized in that: The specific execution logic of the adversarial strategy behavior tree is as follows: the execution process is started from the root node at fixed time intervals. The child nodes of the adversarial strategy behavior tree are traversed from top to bottom and from left to right. The execution signal will be executed according to the priority of the node reached. When the node action cannot be completed or the execution condition is not met during the execution process, the next node is executed. When the entire behavior tree is executed or cannot continue to be executed, the execution stops.

4. The method for constructing a threat target adversarial strategy behavior tree for a street defense scenario according to claim 1, characterized in that: The outcome is determined when one side's key nodes are destroyed or all enemies are wiped out. Before each round of combat begins, all agents are in their initial positions. The maximum number of steps for each round is set. Both the red and blue sides act according to the predetermined strategy described by the logic of the adversarial strategy behavior tree. If the red side destroys the blue side's key nodes or eliminates all enemies within the maximum number of steps, the red side wins. If the red side fails to destroy the blue side's key nodes within the maximum number of steps or if all of the red side's attack nodes are eliminated before reaching the maximum number of steps, the blue side wins.

5. The method for constructing a threat target adversarial strategy behavior tree for a street defense scenario according to claim 1, characterized in that: in Throughout the execution of the adversarial strategy behavior tree, each node can obtain all the data in the system. The adversarial strategy behavior tree uses an external "blackboard" mechanism to realize communication and data sharing between nodes. The "blackboard" is similar to a short-term data storage center. On the one hand, it stores the specific model data required for the abstract task. On the other hand, the "blackboard" can save the intermediate data generated by the operation node during the operation, providing a communication path for other nodes to access and use this data.

Images