Ground unmanned system multi-agent simulation behavior model construction method based on BDI

Abstract

The invention discloses a BDI-based ground unmanned system multi-agent simulation behavior model construction method, and relates to the technical field of unmanned system simulation, and the method comprises the following steps: S1, constructing an internal BDI cognitive layer of an agent; s2, setting an intention-behavior converter; s3, constructing a Couzin model layer; and S4, establishing a sensing feedback loop. According to the method, the motion naturalness is remarkably improved while the task efficiency is guaranteed through a bidirectional coupling architecture, the trust and controllability of people to an unmanned cluster are enhanced by quantifying a mapping chain and establishing a transparent bridge from a human instruction to a cluster behavior, and the problem of disjunction of cognition and behaviors in a traditional model is effectively solved; according to the method, the intelligent agent cluster shows smooth and natural self-organizing motion characteristics during task execution, the established task target can be efficiently completed, the sense of reality of a simulation scene and the natural coordination of intelligent agent behaviors are remarkably improved, and a simulation result is closer to the operation performance of an actual ground unmanned cluster.

Description

A Method for Constructing Multi-Agent Simulation Behavioral Models for Ground-Based Unmanned Systems Based on BDI Technical Field

[0001] This invention relates to the field of unmanned system simulation technology, specifically to a method for constructing a multi-agent simulation behavior model of a ground-based unmanned system based on BDI, and in particular, a hybrid modeling method that integrates a cognitive architecture (BDI model) with a biologically inspired self-organizing cluster model (Couzin model). Background Technology

[0002] Ground-based unmanned systems are intelligent operational systems centered on unmanned ground equipment. They can replace manual labor in high-risk and complex environments to complete tasks such as emergency rescue, geological exploration, and patrol duty. This represents a core direction for the practical application of unmanned system technology. Multi-agent technology is the core support for the simulation and deduction of ground-based unmanned systems. After years of development, research on intelligent agent simulation behavior models has yielded substantial results. Currently, the mainstream behavior modeling methods mainly include four types, and the principles and inherent limitations of each method are as follows:

[0003] Finite state machine-based methods decompose agent behavior into a series of discrete states. At any given time, the agent is in only one state and performs the corresponding behavior. State transitions are triggered by preset conditions. However, this method requires a finite number of states and clear state transition relationships, has poor scalability, and is difficult to widely apply.

[0004] Behavior tree-based methods: As an improvement on finite state machines, this is a hierarchical and modular decision logic modeling framework that decomposes complex behaviors into simple task nodes and defines the execution order and logical relationships of nodes through a tree structure; however, the model maintenance of this method is more difficult.

[0005] Machine learning-based methods construct agent behavior models as approximations of knowledge acquisition and representation processes. By learning from historical experience and relevant data, they obtain behavioral performances that conform to experience and have strong scalability. However, this method has a long training time, requires a large amount of training data, and is difficult to adapt to dynamic external environments.

[0006] The Cousin model-based approach belongs to the biological cluster model and focuses on the study of cluster behavior. Based on the Boid cluster model, it uses mathematical modeling to elaborate on the aggregation of individuals among biological communities, the repulsion of surrounding individuals, and the variable speed of interacting individuals. It divides the system into three sensory regions: repulsion, attraction, and synchronization, and clearly explains the internal relationship mechanism between individuals, local and whole in the community. The core of this method is designed for cluster movement and lacks the ability to plan high-level tasks.

[0007] Based on the above description, the core problem with existing technologies is that the core defect of existing multi-agent simulation behavior model construction technologies for ground-based unmanned systems is the isolated use of various modeling techniques. This leads to a disconnect between the agent's "cognition" and "behavior," creating a significant performance bottleneck and becoming a key challenge for industry development. Specifically, the problems manifest as follows:

[0008] Performance bottleneck of pure cognitive model (BDI): Although agents using pure BDI model can complete complex task planning, their underlying motion is mostly based on simple point-to-point path tracking, resulting in scattered formation, stiff trajectory and unnatural collision avoidance during swarm maneuvering; specifically, in standard reconnaissance missions, the average formation maintenance error of the swarm is higher than that of the bio-inspired model, and the frequency of curvature change of the motion trajectory is significantly increased.

[0009] Performance bottlenecks of the pure reactive swarm model (Couzin model): Although agents using the pure Cousin model can exhibit smooth self-organized movement, they lack high-level task understanding and planning capabilities. When facing dynamic, multi-stage tasks (such as "search-identify-encircle"), the macro-level task completion rate is low and the response delay to sudden instructions is long. Specifically, in scenarios with dynamic target switching, the overall task completion time is much longer than that of models with planning capabilities, and the response time to new threats exceeds several times the planning cycle. Summary of the Invention

[0010] To address the shortcomings of existing technologies, this invention provides a method for constructing a multi-agent simulation behavior model for ground-based unmanned systems based on BDI, thus solving the problems mentioned in the background section.

[0011] To achieve the above objectives, this invention provides the following technical solution: a method for constructing a multi-agent simulation behavior model for ground-based unmanned systems based on BDI, which constructs a two-layer coupled "cognition-behavior" architecture within the agent and establishes a perception feedback loop to form a decision-action closed loop, specifically including the following steps:

[0012] S1: Construct the BDI cognitive layer inside the intelligent agent, and generate task-level planning intentions based on beliefs and desires. The beliefs include external land environment beliefs and internal state beliefs. The intentions are updated and generated through the BDI decision function.

[0013] S2: Set up an intent-behavior converter to parse and map the abstract intent output by the BDI cognitive layer into a dynamic parameter set of the Cousin model. The dynamic parameter set includes a role weight vector and an interaction area radius parameter. The mapping process is combined with the agent's beliefs to achieve context-aware behavior adaptation.

[0014] S3: Construct a Cousin model layer that is improved by role weighting and parameter dynamics. Take the dynamic parameter set as input, classify the agent's neighbors, calculate the repulsion, synchronization and attraction motion directions, weight and synthesize the desired direction, and combine the agent's fatigue and inertia to complete the motion state update.

[0015] S4: Establish a perception feedback loop to feed back the motion state and environmental perception results updated at each step of the Cousin model layer to the BDI cognitive layer, dynamically update the agent's beliefs, and form a closed loop of "external environmental stimulus → internal state change → decision behavior change → influence on environment and internal state".

[0016] Furthermore, in step S1, the BDI layer adopts a classic formal model, assuming a set of ground-based unmanned swarm intelligence agents. , Represents driverless cars, robots, and intelligent agents. At any moment The belief is a complex structure:

[0017]

[0018] in, Beliefs about the external land environment, including digital terrain elevation, slope, road network, weather, and the user's own location and speed. Observed neighbor set Information, and the intentions of teammates revealed through communication;

[0019] For internal state beliefs:

[0020]

[0021] in, :experience; Survival ability; Loss rate.

[0022] Furthermore, in step S1, the BDI decision function is:

[0023]

[0024] in, It is an intentional selection function; For intelligent agents at any time The desired outcome, namely, the set of objectives for land-based missions; For intelligent agents at any time The intention; For intelligent agents at any time The intention.

[0025] Furthermore, in step S2, the converter will abstract the intent. Translated into a dynamic parameter set that the lower-level Cousin model can understand. :

[0026]

[0027] in, For conversion functions; For intelligent agents At any moment The dynamic parameter set of the Cousin model; the role weight vector: ,and The interaction region radius parameter includes the repulsion radius. Synchronization radius attraction radius .

[0028] Furthermore, in step S2, the mapping rule includes: if the intention includes acting as a mountain marching navigator, then increase... and If the intention includes a dispersed search of the urban environment, then the search volume will increase. and If the intention involves a tight defense against an ambush, then reduce... Increase and increase If the survivability S drops to a low level, then through Dynamically increase the repulsion radius, where, This is the pressure resistance sensitivity coefficient.

[0029] Furthermore, in step S3, the classification method for the agent's neighbors is as follows: based on the interaction region radius parameter in the dynamic parameter set, neighbors within the perception range are divided into exclusion zone neighbors. Neighbors in the synchronization zone Attracting neighbors .

[0030] Furthermore, in step S3, the calculation of the direction of motion includes:

[0031] Repulsion direction:

[0032]

[0033] Synchronization direction:

[0034]

[0035] Direction of attraction:

[0036]

[0037] in, Current intelligent agent At simulation time Position vector; : with the current intelligent agent Interactive neighboring intelligent agents At simulation time Position vector; Neighbor Intelligent Agent At simulation time The velocity vector;

[0038] Simultaneously, combine the target direction analyzed from the BDI intent. and the direction of topographic gradient transformed from landform Participate in sports decision-making.

[0039] Furthermore, in step S3, the formula for the weighted synthesis desired direction is:

[0040]

[0041] in, Fixed weights for each basic behavior; This refers to the target direction after role weighting. It is the direction of motion desired by the intelligent agent.

[0042] Furthermore, in step S3, the agent's movement speed is combined with its fatigue level. calculate: , The fatigue sensitivity coefficient is used; the agent's motion state update is achieved by combining inertia, and the formula is:

[0043]

[0044] in, For intelligent agents At any moment The actual velocity vector; It is the direction difference vector; For simulating step size, This refers to discretizing a continuous motion process; For intelligent agents At any moment The location.

[0045] Furthermore, in step S4, the update formula for the internal state belief is:

[0046]

[0047] in, , , These are update functions for experience, survivability, and attrition, respectively. , , For a moment Internal state parameters; , , For a moment The internal state parameters.

[0048] This invention provides a method for constructing a multi-agent simulation behavior model for ground-based unmanned systems based on BDI, which has the following beneficial effects:

[0049] 1. This method for constructing a multi-agent simulation behavior model for ground-based unmanned systems based on BDI significantly improves the naturalness of motion while ensuring task efficiency through a bidirectional coupling architecture. By establishing a transparent bridge from human commands to swarm behavior through a quantitative mapping chain of intent to parameters, it enhances human trust and controllability over unmanned swarms, effectively solving the problem of the disconnect between cognition and behavior in traditional models. This allows the agent swarm to exhibit smooth and natural self-organized motion characteristics when performing tasks, and to efficiently complete predetermined task objectives. It significantly improves the realism of the simulation scene and the natural coordination of agent behavior, making the simulation results closer to the actual operational performance of unmanned swarms.

[0050] 2. The method for constructing a multi-agent simulation behavior model for ground-based unmanned systems based on BDI establishes a transparent and controllable bridge from task instructions to swarm behavior by constructing an explicit mapping mechanism between intent and behavioral parameters. This allows operators to intuitively predict and control the macroscopic movement of the swarm. At the same time, relying on a closed-loop design of perception feedback, the unmanned system can quickly complete cognitive replanning and adaptive behavioral adjustment when facing complex and dynamic environmental changes, effectively improving the robustness of task execution and adaptability to changing scenarios. Attached Figure Description

[0051] Figure 1 is a flowchart illustrating the steps of the method for constructing a multi-agent simulation behavior model of a ground-based unmanned system based on BDI according to the present invention.

[0052] Figure 2 is a schematic diagram of the ground unmanned system behavior modeling framework of the ground unmanned system multi-agent simulation behavior model construction method based on BDI of the present invention;

[0053] Figure 3 is a schematic diagram of the internal architecture of the intelligent agent in the BDI-based multi-agent simulation behavior model construction method for ground unmanned systems of the present invention.

[0054] Figure 4 is a schematic diagram of the intent-parameter mapping rules for the regional reconnaissance mission of the BDI-based multi-agent simulation behavior model construction method for ground unmanned systems according to the present invention. Detailed Implementation

[0055] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0056] As shown in Figures 1-4, this invention provides a technical solution: a method for constructing a multi-agent simulation behavior model for ground unmanned systems based on BDI. This method constructs a two-layer coupled architecture of "cognition-behavior" within the agent and establishes a perception feedback loop to form a decision-action closed loop. To improve the realism and intelligence of multi-agent simulation of ground unmanned systems, this technical solution designs a ground unmanned system behavior modeling framework, as shown in Figure 2. First, relatively simple and autonomous individual agents are designed, and a "cognition-behavior" closed loop is established within the agent. Based on this, multiple agents can interact locally according to rules, enabling them to exhibit intelligent, adaptive, and robust cluster collaboration capabilities at the system level.

[0057] Here, swarm intelligence is not "programmed" but rather generated spontaneously and emergently by equipping each individual with a cognitively driven, parameterized "biological instinct" interaction rule and allowing them to interact freely in simulation. This establishes a "cognitive-guided self-organizing" mechanism to ensure more realistic simulation of complex, intelligent, and adaptive unmanned swarm behavior.

[0058] The "cognition-behavior" two-layer architecture established inside the intelligent agent is shown in Figure 3. The upper BDI engine is responsible for rational decision-making and intent generation at the task level, while the lower Cousin dynamics engine is responsible for real-time motion control based on local interaction. The two are bidirectionally coupled through an "intent-behavior converter".

[0059] Upper level: Based on beliefs, select goals and form planning intentions.

[0060] The bottom layer is responsible for fast and smooth maneuver control based on local information, realizing the top-level intent (translating the top-level intent into parameters or rule weights of a bio-inspired model, calculating information about neighbors / obstacles in the cluster in real time, and performing actions such as collision avoidance, regrouping, and formation maintenance).

[0061] The main logic is to parse and map the high-level "intention" output by the BDI model into dynamic role parameters (such as "leader", "follower", "explorer") and dynamic region radius parameters in the Cousin model, so that the emergent behavior of the agent cluster serves the cognitive goals of the high level. At the same time, the environmental state (such as neighbor location and obstacle information) generated by the lower-level dynamics is fed back to the BDI layer in real time to update its "belief", forming a decision-making closed loop.

[0062] Taking the "area reconnaissance mission" as an example, the mapping from the intention to the dynamic parameters of the Cousin model is shown in Figure 4;

[0063] The specific plan includes the following steps:

[0064] S1. Construct the BDI cognitive layer inside the intelligent agent:

[0065] The BDI layer adopts a classic formal model, assuming a set of ground-based unmanned swarm intelligence agents. ( (Can represent driverless cars, robots, etc.) Intelligent agents At any moment The belief is a complex structure:

[0066]

[0067] in: Beliefs about the external land environment, including digital terrain elevation, slope, road network, weather, and the user's own location and speed. Observed neighbor set Information, and the intentions of teammates revealed through communication;

[0068] For internal state beliefs:

[0069]

[0070] in, Experience: A high experience value can improve the ability to optimize path selection in complex terrain (such as gravel ground and bushes) and the speed of sensor data parsing; Survivability is related to the platform's physical protection and electronic countermeasures capabilities. When subjected to simulated fire threats or strong electromagnetic interference, this value decreases dynamically, leading to a reduction in the sensing range, a decrease in communication success rate, and a decrease in decision confidence. Wear and tear is related to platform endurance, component wear and tear, and energy consumption. It accumulates with mission time and motion load, resulting in a decrease in maximum speed and a slower maneuver response.

[0071] desire : A set of land-based mission objectives from the cluster command system or generated autonomously, such as {reaching the navigation point, maintaining a wedge formation, conducting a thorough search of the target area, avoiding dynamic threats, and occupying high ground};

[0072] intention : A current action plan chosen and committed to be executed based on current desires, the terrestrial environment, and one's own state of mind, such as "detour along a valley towards point B";

[0073] BDI decision function:

[0074]

[0075] in It is an intention selection function that is based on current beliefs and desires. It may retain, adjust or generate new intentions. For example, when the belief perceives an obstacle ahead, the priority of "avoiding the obstacle" in the desire increases, and the intention may switch from "going straight" to "going around".

[0076] S2. Set the intent-behavior converter:

[0077] The converter will abstract the intent. Translated into a dynamic parameter set that the lower-level Cousin model can understand. :

[0078]

[0079] in, This mainly includes: Role weight vector: ,and This vector directly determines the agent's behavioral tendencies within the cluster;

[0080] Interaction region radius: Repulsion radius Ensure safe distance; synchronization radius It affects the range of motion coordination; the radius of attraction. This affects the scope of cluster aggregation and goal orientation; it influences beliefs. As a converter The input is the core of achieving "context-aware behavior adaptation," which means that even if two agents have the same intention... If their environmental conditions and their own states (beliefs) are different, they may adopt different behavioral parameters to ensure that their intentions can be achieved more safely and effectively.

[0081] Mapping rules:

[0082] like If it includes "serving as a mountain marching navigator", then Significantly increased Enlarge it appropriately to attract followers;

[0083] like Including "dispersed search of urban environment", then Significantly increased Enlarge it appropriately to attract followers;

[0084] like If it includes "ambushed, defend tightly", then , Increase, then Significant increase;

[0085] If experience Individuals with higher synergy will have a higher weight in their synchronization direction when "forming" a group. This results in a more stable and efficient formation;

[0086] If the pressure resistance If it drops to a low level due to an attack, then Will be through formula The dynamic increase manifests as risk-avoidance behavior after being startled, among which... This is the compressive strength sensitivity coefficient;

[0087] S3. Construct the Cousin model layer:

[0088] The traditional Cousin model defines three concentric perception regions (repulsion, attraction, and synchronization) for each agent. This method builds upon this by adding role weighting and dynamic parameterization for the agents. :

[0089] Neighbor classification: based on The radius parameter in the sensor classifies neighbors within the sensing range into three categories: exclusion zone neighbors. Neighbors in the synchronization zone Attracting neighbors ;

[0090] The direction is calculated as follows:

[0091] Repulsion direction:

[0092]

[0093] Right now Given the combined position of all nearby neighbors, it is a unit direction vector pointing outward (i.e. away from them). It indicates the direction in which the agent should move to avoid colliding with all too close neighbors. Its calculation depends only on position and is independent of the velocity of the neighbors. It is a pure anti-collision force.

[0094] Synchronization direction:

[0095]

[0096] Right now It is the vector average of the velocities of local neighbors.

[0097] Direction of attraction:

[0098]

[0099] Right now To face the center of the neighborhood group;

[0100] Target direction (from BDI): The specific direction of movement derived from the intent; the direction of the terrain gradient: This transforms the terrain slope and travel cost into an additional terrain gradient direction. This guides the agent to prioritize choosing flat paths.

[0101] The repulsion direction, alignment direction, and attraction direction mentioned above together constitute the three basic vectors driving group behavior in the Cousin model. The meanings of each direction are shown in Table 1 below:

[0102] Table 1

[0103] The core purpose of this system is to avoid collisions, maintain a minimum safe distance ("personal space"), coordinate movement, ensure that individuals move in the same direction as their neighbors, and achieve coordinated maneuvers such as group turning. It also maintains aggregation, prevents individuals from leaving the group, maintains group integrity, and facilitates convergence towards the target. The calculation is based on the relative position (direction) of neighbors. The formula is: summing the unit direction vectors pointing to neighbors and then inverting the result. The velocity vector (direction) of neighbors is calculated using the formula: summing the velocity vectors of neighbors and then normalizing (calculating the average direction). The relative position (center of mass) of the neighbor group is calculated using the formula: summing the position vectors pointing to neighbors (calculating the direction of the center of mass) and then normalizing. The inner region of the perception area (repulsion zone) has a radius of... Typically the smallest. Middle layer region (alignment area), radius... It lies between the repulsion region and the attraction region. The outer region (attraction region) has a radius of... Typically the largest. Biological metaphors for an immediate defensive response to intrusion into personal space. Imitation and coordination of group movement trends to achieve smooth collective movement. Tendency toward group spatial location and toward resources / goals. Dynamics and enhancements in the model through modulation. Directly control the size of "personal space" to achieve variations in formation density (such as tight assault formations or dispersed search formations). 1. By adjusting... Controlling the "coordination range" affects the consistency and stability of formation direction. 2. By adjusting the alignment weights. This controls the degree to which individuals follow the group's movement trends. 1. By adjusting... 1. Control the range of "group vision" or "target vision". 2. Core Enhancement: Direct the target direction generated by BDI intent. As the core source of attraction, it integrates role weighting to achieve goal-oriented group movement. Its function can be analogous to "push": a short-range, powerful, and precise separating force that ensures individual safety. "Steering force" or "coordination force": a medium-range, adjusting force used for matching, ensuring the consistency of group movement. "Pull force": a long-range, gentle, and holistic cohesive force that ensures group unity and goal orientation. surface

[0104] Weighted synthesis expected direction:

[0105]

[0106] in, Fixed weights for each basic behavior, This refers to a goal direction that has been weighted by roles, such as what a leader might assign. Higher weight;

[0107] The speed and fatigue of the agent Affecting the agent's mobility:

[0108]

[0109] This is the fatigue sensitivity coefficient; as fatigue increases, the agent's movement speed decreases.

[0110] Motion updates are performed by incorporating the agent's "inertia":

[0111]

[0112] In the formula: For intelligent agents At any moment The actual velocity vector; It is the direction difference vector; For simulating step size, This refers to discretizing a continuous motion process; For intelligent agents At any moment The location.

[0113] S4. Establish a sensing feedback loop:

[0114] The results of each step update at the bottom layer of the model (position, velocity, perceived terrain and obstacles, real-time distribution of neighbors) and the new environmental perception are used to update... On the other hand, the effects of the action itself and the perceived urgency of the event serve as inputs to the internal state update function, dynamically adjusting... :

[0115]

[0116] in, , , These are update functions for experience, survivability, and attrition, respectively. , , For a moment Internal state parameters; , , For a moment Internal state parameters;

[0117] This forms a closed loop of "external environmental stimuli → changes in internal state → changes in decision-making behavior → influence on the environment and internal state";

[0118] Based on the above description, this invention constructs a two-layer hybrid agent model with a closed-loop feedback of "cognition-behavior" and designs an "intent-behavior converter". "This will output the high-level task intent from the BDI model." The parameters are analyzed and quantized into a set of dynamic control parameters for the underlying Cousin model. This includes role weight vectors and interaction radii. This mapping relationship is explicit and adjustable, allowing operators to indirectly and precisely control the macroscopic movement pattern of the cluster (such as from "dispersed" to "gathered") by changing task instructions. This achieves interpretable and predictable guidance from cognitive objectives to emergent cluster behavior. Through the above architecture, the agent can complete high-level cognitive tasks (such as area blockade and reconnaissance) in complex land environments, and also exhibit natural and smooth low-level cluster movement. Compared with the isolated model, this approach can shorten the task completion time when completing the same complex reconnaissance task, while reducing the average alignment error of cluster movement and reducing the number of collisions.

[0119] In summary, this invention provides an efficient, natural, and interpretable behavior modeling method. Through a bidirectional coupling architecture, it significantly improves the naturalness of movement (formation maintenance error, trajectory smoothness) while ensuring task efficiency (task completion time, coverage). By establishing a quantitative mapping chain of "intent → parameters," a transparent bridge is built from human instructions to swarm behavior, enhancing human trust and controllability over unmanned swarms. This effectively solves the problem of the disconnect between cognition and behavior in traditional models, allowing intelligent agent swarms to exhibit smooth and natural self-organized movement characteristics when performing tasks, and efficiently complete predetermined task objectives. It significantly improves the realism of simulation scenarios and the natural coordination of intelligent agent behavior, making simulation results closer to the actual operational performance of unmanned swarms.

[0120] By constructing an explicit mapping mechanism between intent and behavioral parameters, a transparent and controllable bridge is established from task instructions to cluster behavior, allowing operators to intuitively predict and regulate the macroscopic movement of the cluster. At the same time, relying on the closed-loop design of perception feedback, the unmanned system can quickly complete cognitive replanning and behavioral adaptive adjustment when facing complex and dynamic environmental changes, effectively improving the robustness of task execution and adaptability to changing scenarios.

[0121] The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described in order to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and to design various embodiments with various modifications suitable for a particular purpose.

Claims

1. A method for constructing a multi-agent simulation behavior model for ground-based unmanned systems based on BDI, characterized by: Construct a two-layer coupled architecture of "cognition-behavior" within the intelligent agent, and establish a perception feedback loop to form a decision-making-action closed loop. Includes the following steps: S1: Construct the BDI cognitive layer inside the agent, and generate task-level planning intentions based on beliefs and desires. The beliefs include external land environment beliefs and internal state beliefs. The intentions are updated and generated through the BDI decision function. S2: Set up an intention-behavior converter to parse and map the abstract intentions output by the BDI cognitive layer into a dynamic parameter set of the Cousin model. The dynamic parameter set includes role weight vectors and interaction area radius parameters. The mapping process is combined with the agent's beliefs to achieve context-aware behavior adaptation. S3: Construct a Cousin model layer improved with role weighting and parameter dynamics. Take the dynamic parameter set as input, classify the agent's neighbors, calculate the repulsion, synchronization, and attraction motion directions, synthesize the desired direction with weights, and complete the motion state update by combining the agent's fatigue and inertia. S4: Establish a perception feedback loop to feed back the motion state and environmental perception results updated by the Cousin model layer at each step to the BDI cognitive layer, dynamically update the agent's beliefs, and form a closed loop of "external environmental stimulus → internal state change → decision behavior change → influence on environment and internal state".

2. The method for constructing a multi-agent simulation behavior model of a ground-based unmanned system based on BDI according to claim 1, characterized in that: In step S1, the BDI layer adopts a classic formal model, assuming a set of ground-based unmanned swarm intelligence agents. ， Represents driverless cars, robots, and intelligent agents. At any moment The belief is a complex structure: in, Beliefs about the external land environment, including digital terrain elevation, slope, road network, weather, and the user's own location and speed. Observed neighbor set Information, and the intentions of teammates revealed through communication; Beliefs about internal states: in, :experience; Survival ability; Loss rate.

3. The method for constructing a multi-agent simulation behavior model of a ground-based unmanned system based on BDI as described in claim 2, characterized in that: In step S1, the BDI decision function is: in, It is an intentional selection function; For intelligent agents at any time The desired outcome, namely, the set of objectives for land-based missions; For intelligent agents at any time The intention; For intelligent agents at any time The intention.

4. The method for constructing a multi-agent simulation behavior model of a ground-based unmanned system based on BDI according to claim 1, characterized in that: In step S2, the converter will abstract the intent. Translated into a dynamic parameter set that the lower-level Cousin model can understand. ： in, For conversion functions; For intelligent agents At any moment The dynamic parameter set of the Cousin model; the role weight vector: ,and The interaction region radius parameter includes the repulsion radius. Synchronization radius attraction radius 。 5. The method for constructing a multi-agent simulation behavior model of a ground-based unmanned system based on BDI according to claim 4, characterized in that: In step S2, the mapping rule includes: if the intention includes serving as a mountain marching navigator, then increase... and If the intention includes a dispersed search of the urban environment, then the search volume will increase. and If the intention involves a tight defense against an ambush, then reduce... Increase and increase If the survivability S drops to a low level, then through Dynamically increase the repulsion radius, where, This is the sensitivity coefficient for compressive strength.

6. The method for constructing a multi-agent simulation behavior model of a ground-based unmanned system based on BDI according to claim 1, characterized in that: In step S3, the classification method for the agent's neighbors is as follows: based on the interaction region radius parameter in the dynamic parameter set, neighbors within the perception range are divided into exclusion zone neighbors. Neighbors in the synchronization zone Attracting neighbors 。 7. The method for constructing a multi-agent simulation behavior model of a ground-based unmanned system based on BDI as described in claim 6, characterized in that: In step S3, the calculation of the direction of motion includes: repulsion direction: Synchronization direction: Direction of attraction: in, Current intelligent agent At simulation time Position vector; : with the current intelligent agent Interactive neighboring intelligent agents At simulation time Position vector; Neighbor Intelligent Agent At simulation time The velocity vector; combined with the target direction resolved from the BDI intent. and the direction of topographic gradient transformed from landform Participate in sports decision-making.

8. The method for constructing a multi-agent simulation behavior model of a ground-based unmanned system based on BDI as described in claim 7, characterized in that: In step S3, the formula for the weighted synthesis desired direction is: in, Fixed weights for each basic behavior; This refers to the target direction after role weighting. It is the direction of motion desired by the intelligent agent.

9. The method for constructing a multi-agent simulation behavior model of a ground-based unmanned system based on BDI as described in claim 8, characterized in that: In step S3, the agent's movement speed is combined with its fatigue level. calculate: ， The fatigue sensitivity coefficient is used; the agent's motion state update is achieved by combining inertia, and the formula is: in, For intelligent agents At any moment The actual velocity vector; It is the direction difference vector; For simulating step size, This refers to discretizing a continuous motion process; For intelligent agents At any moment The location.

10. The method for constructing a multi-agent simulation behavior model of a ground-based unmanned system based on BDI according to claim 1, characterized in that: In step S4, the formula for updating the internal state belief is: in, 、 、 These are update functions for experience, survivability, and attrition, respectively. 、 、 For a moment Internal state parameters; 、 、 For a moment The internal state parameters.

Images