A digital test scenario design method based on large model agent cooperation
By designing intelligent agents and code-generating intelligent agents to work together, the problems of high threshold, low efficiency and insufficient logic verification in experimental scenario design are solved. It realizes automated closed-loop design and verification from natural language to executable code, thereby improving the quality of experimental scenarios and R&D efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIHANG UNIV
- Filing Date
- 2026-03-13
- Publication Date
- 2026-06-12
AI Technical Summary
In existing technologies, the design of experimental scenarios is difficult and inefficient. Natural language and engineering parameters are separated, and there is a lack of logical consistency verification. The design of evaluation indicators is subjective, which makes it difficult to quantify and evaluate simulation results. Furthermore, large models are prone to errors and cannot be implemented.
A collaborative approach based on large model agents is adopted to construct a design agent and a code generation agent. The design agent performs requirement analysis, logic generation and parameter correction, while the code generation agent performs code conversion and error capture, realizing automated closed-loop design and verification from natural language to executable simulation code.
It has enabled the automated design and verification of experimental scenarios, improved design quality and feasibility, shortened the design cycle, ensured that the generated scenario logic is coherent and the engineering is executable, and significantly improved R&D efficiency.
Smart Images
Figure CN122197598A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of aerospace and complex equipment simulation technology, and particularly relates to a digital test scenario design method based on large model intelligent agent collaboration. Specifically, it relates to a method for automating the design, evaluation index generation, and code-level verification of digital experiment, testing and verification (DETV) scenarios by utilizing large language model (LLM) and multi-agent technology. Background Technology
[0002] In the development of complex equipment systems, DETV (Digital Experiment, Testing and Verification) is a crucial step in ensuring that the equipment performance meets requirements. The test scenario, as the core input to DETV, defines the test environment, physical configuration, mission actions, and evaluation criteria.
[0003] In existing technologies, experimental scenario design mainly faces the following challenges:
[0004] (1) High design threshold and low efficiency: Engineers who are proficient in simulation tools usually need to manually write scripts or configure a large number of parameters, making it difficult to respond quickly to frequently changing test requirements.
[0005] (2) The disconnect between natural language and engineering parameters: Domain experts often require vague natural language, while simulation engines need precise numerical values. Existing conversion processes rely on manual translation, which is prone to errors.
[0006] (3) Lack of logical self-consistency verification: traditionally generated scenarios often only discover logical errors after they are put into operation, lacking an automatic verification mechanism in the design phase.
[0007] (4) Subjective design of evaluation indicators: Often, designers only generate scenarios but do not define scientific success criteria, making it difficult to quantify and evaluate simulation results.
[0008] With the development of Large Language Models (LLMs), generative AI-assisted design has become possible. However, simply relying on LLMs to directly generate code or text can easily lead to illusions, and the generated scenarios often do not conform to physical constraints or cannot be run directly. Therefore, there is an urgent need for an intelligent method that can combine the logical reasoning capabilities of large models with the code execution verification capabilities to achieve the desired "design-verification" closed loop. Summary of the Invention
[0009] The purpose of this invention is to address the problems existing in the prior art and provide a digital test scenario design method based on large-scale model intelligent agent collaboration. This method achieves an automated closed loop from natural language requirements to executable simulation code through the collaborative work of a "design agent" and a "code generation agent."
[0010] To achieve the above objectives, the present invention adopts the following technical solution:
[0011] A digital test scenario design method based on large-scale intelligent agent collaboration includes:
[0012] Step (1): Construct a collaborative system consisting of a Design Agent and a Code Agent, and establish their interaction interface and responsibility boundaries; wherein, the Design Agent has the ability to analyze requirements, retrieve domain knowledge, generate scenario logic and correct parameters, and the Code Agent has the ability to convert code, perform virtual compilation and execution, capture errors and provide log feedback.
[0013] Step (2): Receive the user's test requirement input, which includes structured engineering constraint parameters and unstructured natural language description; using the design agent, based on retrieval augmented generation (RAG) technology, call the domain knowledge base to perform logical reasoning and parameter completion on the input, and generate an initial test scenario natural language text containing a temporal plot and entity behavior tree;
[0014] Step (3): The design agent infers and generates a matching set of evaluation indicators based on the type and target of the generated initial experimental scenario natural language text, and displays it to the user; in response to the user's feedback instructions, it adjusts the initial experimental scenario natural language text or set of evaluation indicators to generate the final confirmed structured scenario configuration file.
[0015] Step (4): The design agent sends the structured scenario configuration file to the code generation agent; the code generation agent acts as a virtual verification terminal, attempts to convert the file into executable simulation code and compile and run it; if the running feedback result contains error information, the design agent automatically corrects the generation parameters according to the feedback, regenerates the structured scenario configuration file and triggers verification again until the simulation code has no compilation errors, no logical conflicts and can stably output complete simulation result data, thus completing the verification.
[0016] Furthermore, step (1) includes:
[0017] Step (1.1), the construction of the design agent includes: selecting a general large language model as the base, injecting the role system prompt of "experiment design expert", and attaching a vectorized retrieval interface to connect to the external equipment performance library, environmental effect library and task rule library; at the same time, configuring a structured text parser, and pre-setting a multi-step reasoning template in the system prompt, forcing the model to perform hard constraint condition decomposition, domain knowledge retrieval, physical boundary condition check and logical text synthesis generation in sequence before generating the scenario;
[0018] Step (1.2) The construction of the code generation agent includes: selecting a large language model with code generation capabilities as the base, injecting the role system instructions of "simulation development engineer", and integrating the sandbox running environment and compiler interface to enable it to have the capabilities of code conversion, virtual execution, error capture and log analysis.
[0019] Furthermore, step (2) includes:
[0020] Step (2.1): The design agent identifies the entity object in the input description, queries the equipment performance library, and obtains the kinematic constraints and load performance parameters of the entity object to constrain the behavior trajectory of the generated entity;
[0021] Step (2.2): The designed intelligent agent extracts meteorological and geographical information from the input description, queries the environmental effect library, calculates the sensor attenuation coefficient and communication interference parameters under specific conditions, and writes the calculation results into the desired environmental variables;
[0022] Step (2.3): The designed intelligent agent analyzes the task intent in the input description, queries the task rule base, and matches the corresponding standard collaborative template and work process to fill in the intended task logic details.
[0023] Furthermore, step (3) includes:
[0024] Step (3.1): The designed intelligent agent performs semantic analysis on the generated initial experimental scenario natural language text to determine the core task type of the experiment;
[0025] Step (3.2): Based on the determined task type, retrieve commonly used evaluation dimensions for similar cases from the historical indicator database;
[0026] Step (3.3): Combine the input structured engineering constraint parameters and calculate the success threshold of the recommendation index through reasoning;
[0027] Step (3.4): Output a list of evaluation indicators that includes the indicator name, calculation logic definition and success determination threshold.
[0028] Furthermore, step (4) includes:
[0029] Step (4.1): The design agent serializes the generated structured scenario configuration file into an intermediate format file without specific code logic, and transmits it to the code generation agent;
[0030] Step (4.2): The code-generating agent parses the intermediate format file and calls the integrated sandbox runtime environment to attempt to build a simulation project;
[0031] Step (4.3): If the construction fails, the code generation agent uses log analysis capabilities to generate structured feedback information containing error type and error location, and sends it to the design agent;
[0032] Step (4.4): The designed intelligent agent parses the structured feedback information, searches the knowledge base to match the error cause, automatically modifies the conflict parameters in the intermediate format file, and returns to step (4.1) to re-serialize and transmit.
[0033] Furthermore, the equipment performance database stores the aerodynamic parameters, battery life, maximum available turn overload, and field of view (FOV) of the UAV.
[0034] Furthermore, the entity includes a multi-rotor drone and a lidar obstacle avoidance module.
[0035] Furthermore, the physical boundary conditions include the total length of the planned route and the radius of curvature of the turning nodes.
[0036] Furthermore, the evaluation index list includes result performance indicators and process constraint indicators; among which, result performance indicators include delivery position deviation and mission success rate, and process constraint indicators include maximum attitude deviation angle, maximum wind resistance overload, and remaining power during flight.
[0037] Furthermore, the intermediate state format file contains only a structured description of entity states, event sequences, and environmental parameters.
[0038] The beneficial effects of this invention are as follows:
[0039] (1) By using a dual-agent decoupling and collaboration mechanism, “creative logic design” and “rigorous engineering verification” are completely decoupled: the design agent focuses on the rationality and compliance design of the experimental task logic, while the code generation agent focuses on the executability and stability verification of the simulation engineering. The two achieve data communication through standardized interfaces, which not only gives full play to the logical reasoning ability of the large model, but also avoids the problem of “design and verification disconnect” of a single model, and greatly improves the quality and feasibility of scenario generation.
[0040] (2) By mounting a domain-specific knowledge base using RAG technology, the illusion of generating large models is suppressed from the root: by constraining the physical boundaries of entity behavior through the equipment performance library, supplementing the environmental impact parameters of the scene through the environmental effect library, and standardizing the industry standards of task process through the task rule library, the generated scenario content is ensured to fully comply with the laws of physics and industry standards, thus completely solving the problem of "wild imagination and inability to be implemented" in the generation of pure large model content.
[0041] (3) Through the original “generation-compilation-error-correction” fully automated closed-loop iterative process, the pre-verification and self-healing optimization of the scenario design are realized: the logic verification is completed through code-level simulation during the design stage, and the parameter correction and iteration are automatically completed based on structured error feedback. No manual intervention is required for debugging, ensuring that the final output scenario is not only logically sound, but can also run stably in the simulation engine. The design cycle of complex scenarios is shortened from several days to minutes, significantly improving the test efficiency of equipment development. Attached Figure Description
[0042] Figure 1 This is a block diagram illustrating the principle of a digital test scenario design method based on large-scale intelligent agent collaboration according to the present invention.
[0043] Figure 2 This is a diagram illustrating the collaborative interaction and data flow architecture between the designed intelligent agent and the code-generating intelligent agent in this invention.
[0044] Figure 3 This is a schematic diagram of the scenario logic deduction module based on Retrieval Enhancement Generation (RAG) technology in this invention. Detailed Implementation
[0045] The present invention will now be described in further detail with reference to the accompanying drawings.
[0046] This invention discloses a digital test scenario design method based on large-scale intelligent agent collaboration. This embodiment uses a "fixed-point material delivery mission in a complex UAV environment" as an example to demonstrate the specific implementation details of this invention. Figure 1 As shown, the specific steps are as follows:
[0047] Step (1): Design a multi-agent collaborative architecture and a deep requirement analysis module. This module establishes the boundary between logical design and engineering implementation by constructing differentiated agent roles, and uses semantic analysis technology to complete the structured extraction of experimental requirements. The specific implementation is as follows:
[0048] ① Construct a dual-agent collaborative system: Define two core agents—the Design Agent and the Code Agent. The Design Agent is injected with the system prompt "Experiment Design Expert," configured with a chain-of-thought reasoning mode, and responsible for domain logic reasoning and natural language interaction. The Code Agent is injected with the system prompt "Simulation Development Engineer," integrates the simulation engine API documentation index, and is responsible for converting logic into executable code. The two communicate through a standardized JSON (JavaScript Object Notation) intermediate data interface, achieving decoupling between the "business logic layer" and the "underlying code implementation layer."
[0049] like Figure 2 As shown, the construction of the intelligent agent includes: selecting a general large language model as the base, injecting the role system instructions of "experiment design expert", and attaching a vectorized retrieval interface to connect to external equipment performance library, environmental effect library and task rule library; at the same time, configuring a structured text parser, pre-setting multi-step reasoning templates in the system instructions, and forcing the model to perform hard constraint condition decomposition, domain knowledge retrieval, physical boundary condition check and logical text synthesis generation in sequence before generating the scenario;
[0050] The construction of the code generation agent includes: selecting a large language model with code generation capabilities as the base, injecting the role system instructions of "simulation development engineer", and integrating a sandbox runtime environment and compiler interface to enable it to have the capabilities of code conversion, virtual execution, error capture and log analysis.
[0051] The design agent and the code generation agent have conflict resolution reasoning and parameter self-healing capabilities: the design agent generates a structured scenario configuration file based on the domain knowledge base and historical indicator database and sends it to the code generation agent; after verification by the code generation agent, the structured error information is fed back to the design agent, which then completes parameter correction and iterative optimization.
[0052] ② Construct a domain-specific knowledge base: In order to support the decision-making of intelligent agents, construct a vectorized external knowledge base, including: equipment performance base (storing the aerodynamic parameters of UAVs, battery life, maximum available turn overload, field of view (FOV) of electro-optical pods, etc.), environmental effect base, and mission rule base.
[0053] ③ Joint Semantic Parsing of Requirements: The system jointly parses the natural language description document (e.g., "generating a scenario for precise material delivery to disaster-stricken and trapped areas in rainy weather, avoiding urban high-rise buildings and no-fly zones") and engineering parameter files input by the user. The system combines a domain-specific vocabulary from the aerospace field to accurately identify key elements in the experiment: including equipment entities (e.g., a certain type of multi-rotor UAV, lidar obstacle avoidance module), environmental constraints (e.g., visibility <2km, urban canyon terrain), mission intent (e.g., low-altitude shuttle, precise hovering at the end of the flight path), and hard constraint boundaries (flight speed, altitude, delivery accuracy, remaining battery power).
[0054] ④ Define the boundaries of the digital test: Based on the analyzed application scenario, determine the topology and simulation accuracy of the digital test scenario. The topology range is a 5km×5km urban area, the simulation step size is 100ms, and the simulation accuracy is centimeter-level trajectory control. The core assessment objectives of the test are clearly defined as delivery accuracy and obstacle avoidance reliability in complex environments.
[0055] Step (2): Design a scenario logic deduction module based on Retrieval Enhanced Generation (RAG). This module utilizes an external domain knowledge base to enhance the capabilities of the large model, ensuring that the generated initial scenario conforms to physical laws and safe flight regulations. This includes: receiving user input for experimental requirements, which includes structured engineering constraint parameters and unstructured natural language descriptions; using the designed intelligent agent, based on Retrieval Enhanced Generation (RAG) technology, calling a preset domain knowledge base to perform logical reasoning and parameter completion on the input, generating a natural language text of the initial experimental scenario containing temporal sequence, environment, and entities. The specific implementation is as follows: Figure 3 As shown:
[0056] ① Entity Constraint Extraction and Knowledge Retrieval: The design of an intelligent agent recognizes entity objects in the input description, queries the equipment performance database to obtain the kinematic constraints and load performance parameters of the entity object, so as to constrain the behavior trajectory of the generated entity; the parsed user requirements are transformed into high-dimensional vectors, and the cosine similarity between them and vectors in the knowledge base is calculated to retrieve Top-K related domain knowledge fragments and similar task scenario templates.
[0057] ② Environmental parameter mapping: Design an agent to extract meteorological and geographical information from the input description, query the environmental effects library, calculate the sensor attenuation coefficient and communication interference parameters under specific conditions, and write the calculation results into the desired environmental variables.
[0058] ③ Intent Analysis and Process Matching: Design an intelligent agent to analyze the task intent in the input description, query the task rule base, and match the corresponding standard collaborative templates and work processes to fill in the intended task logic details.
[0059] ④ Multi-step physical logic reasoning and initial scenario synthesis: A designed intelligent agent performs thought chain reasoning. Hard constraints are decomposed and physical boundary conditions are checked (verifying the total planned route length, the radius of curvature of turning nodes, etc.). The reasoning results are integrated to generate a natural language script describing the entire process of the UAV from takeoff and climb, mid-course cruise, terminal approach, to precise airdrop of supplies. This script includes a temporal narrative, entity behavior tree, and environmental configuration, ensuring logical consistency and physical realism.
[0060] Step (3): Design an automatic construction and human-computer interaction optimization module for evaluation indicators. This module dynamically matches evaluation dimensions according to the scenario type and introduces a human-in-the-loop mechanism for logical fine-tuning. This includes the design agent inferring and generating a matching set of evaluation indicators based on the type and target of the generated initial experimental scenario natural language text, and displaying it to the user; responding to user feedback instructions, adjusting the initial experimental scenario natural language text or evaluation indicator set to generate a final confirmed structured scenario configuration file. The specific implementation is as follows:
[0061] ①Task type identification: Design an intelligent agent to perform semantic analysis on the generated initial experimental scenario natural language text to determine that the core task type of the experiment is "precise delivery and obstacle avoidance in complex environments".
[0062] ②Evaluation dimension matching: Based on the defined task type, retrieve commonly used evaluation dimensions of similar cases from the historical indicator database.
[0063] ③ Successful determination threshold calculation: Combining the input structured engineering constraint parameters, the successful determination threshold of the recommendation index is calculated through reasoning.
[0064] ④ Output of indicator list: The output includes an evaluation indicator list containing indicator names, calculation logic definitions and success judgment thresholds (specifically including result performance indicators such as delivery position deviation and mission success rate, and process constraint indicators such as maximum attitude deviation angle, maximum wind resistance overload, and remaining power during flight).
[0065] ⑤ Visual Interaction and Logic Adjustment: An interactive interface is established to display the initial design concept and indicator set to the user. In response to user feedback and modification commands, the design agent automatically updates the design scheme and parameters using relevant templates from a domain-specific knowledge base, ultimately generating a final, confirmed structured design concept configuration file.
[0066] Step (4): Design a closed-loop verification module based on sandbox feedback. This module verifies the executability of the scenario through code-level simulation and uses error logs to drive the agent to correct parameters. This includes the design agent sending the structured scenario configuration file to the code generation agent; the code generation agent, acting as a virtual verification terminal, attempts to convert the file into executable simulation code and compile and run it; if the running feedback contains error information, the design agent automatically corrects the generated parameters based on the feedback, regenerates the structured scenario configuration file, and triggers verification again until the simulation code has no compilation errors, no logical conflicts, and can stably output complete simulation result data, thus completing the verification. The specific implementation is as follows:
[0067] ① Intermediate State Serialization and Transmission: The design agent serializes the structured scenario configuration file confirmed in step (3) into an intermediate state format file that does not contain specific code logic, and transmits it to the code generation agent. This file only contains a structured description of entity states, event sequences, and environmental parameters.
[0068] ② Simulation Code Parsing and Sandbox Construction: The code generation agent parses the intermediate format file and calls the integrated isolated sandbox runtime environment to attempt to build a simulation project. It maps the classes and objects of the simulation engine and uses pre-set code templates and code generation models to transform them into compilable simulation project code (including complete simulation step size control, entity instantiation, and data recording logic).
[0069] ③ Error Feature Capture and Feedback Generation: Attempt to pre-compile and test-run the generated simulation code. If the simulation test-run fails, the code generation agent uses regular expressions to extract error features, generates structured feedback information containing error type, error location, and physical cause, and sends it to the design agent.
[0070] ④ Conflict Resolution and Iterative Verification: The intelligent agent receives and parses structured feedback information, triggering conflict resolution reasoning. By searching the knowledge base to match the cause of the error, it automatically modifies the conflict parameters in the intermediate format file and returns to step ① to re-serialize and transmit. This process is repeated iteratively until the simulation code can pass without errors and output complete flight track and flight control data, thus outputting a high-quality DETV scenario that has been verified by both logic and engineering.
[0071] In summary, this invention discloses a digital test scenario design method based on large-scale intelligent agent collaboration. This method includes designing a multi-agent collaborative architecture and a deep requirements analysis module; designing a scenario logic deduction module based on retrieval enhancement generation; designing an automatic evaluation index construction and human-computer interaction optimization module; and designing a sandbox feedback-based closed-loop verification and self-healing module. This invention effectively solves the problem of logic and code disconnect in traditional scenario design. Through intelligent agent collaboration and closed-loop verification, it significantly reduces manual debugging time, ensuring that the generated test scenarios meet both experimental requirements and are engineering-executable.
[0072] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A digital test scenario design method based on large-scale intelligent agent collaboration, characterized in that, The method includes: Step (1): Construct a collaborative system consisting of a design agent and a code generation agent, and establish their interaction interface and responsibility boundaries; wherein, the design agent has the ability to analyze requirements, retrieve domain knowledge, generate scenario logic and correct parameters, and the code generation agent has the ability to convert code, perform virtual compilation and execution, capture errors and provide log feedback. Step (2): Receive the user's test requirement input, which includes structured engineering constraint parameters and unstructured natural language description; using the design agent, based on retrieval enhancement generation technology, call the preset domain knowledge base to perform logical reasoning and parameter completion on the input, and generate an initial test scenario natural language text containing time sequence, environment, entity and entity behavior tree; Step (3): The design agent, based on the task type and test objective of the generated initial test scenario natural language text, and a preset historical index database, infers and generates a matching evaluation index set, which is then displayed to the user; in response to the user's feedback instructions, the initial test scenario natural language text or evaluation index set is adjusted to generate the final confirmed structured scenario configuration file. Step (4): The design agent sends the structured scenario configuration file to the code generation agent; the code generation agent, acting as a virtual verification terminal, converts the file into executable simulation code and compiles and runs it; if the running feedback results contain error information or logical conflicts, the design agent automatically corrects the generation parameters according to the feedback, regenerates the structured scenario configuration file and triggers verification again until the simulation code has no compilation errors, no logical conflicts and can stably output complete simulation result data, thus completing the verification.
2. The digital test scenario design method based on large-scale intelligent agent collaboration according to claim 1, characterized in that, Step (1) includes: Step (1.1), the construction of the design agent includes: selecting a general large language model as the base, injecting the role system instructions of "experiment design expert", and attaching a vectorized retrieval interface to connect to the external equipment performance library, environmental effect library and task rule library; at the same time, configuring a structured text parser, pre-setting a multi-step reasoning template in the system instructions, and forcing the model to perform hard constraint condition decomposition, domain knowledge retrieval, physical boundary condition check and logical text synthesis generation in sequence before generating the scenario; Step (1.2) The construction of the code generation agent includes: selecting a large language model with code generation capabilities as the base, injecting the role system instructions of "simulation development engineer", and integrating the sandbox running environment and compiler interface to enable it to have code conversion, virtual execution, error capture and log analysis capabilities.
3. The digital test scenario design method based on large-scale intelligent agent collaboration according to claim 2, characterized in that, Step (2) includes: Step (2.1): The design agent identifies the entity object in the input description, queries the equipment performance library, and obtains the kinematic constraints and load performance parameters of the entity object to constrain the behavior trajectory of the generated entity; Step (2.2): The designed intelligent agent extracts meteorological and geographical information from the input description, queries the environmental effect library, calculates the sensor attenuation coefficient and communication interference parameters under specific conditions, and writes the calculation results into the desired environmental variables; Step (2.3): The designed intelligent agent analyzes the task intent in the input description, queries the task rule base, and matches the corresponding standard collaborative template and work process to fill in the intended task logic details.
4. The digital test scenario design method based on large-scale intelligent agent collaboration according to claim 1, characterized in that, Step (3) includes: Step (3.1): The designed intelligent agent performs semantic analysis on the generated initial experimental scenario natural language text to determine the core task type of the experiment; Step (3.2): Based on the determined task type, retrieve the evaluation dimensions of similar cases from the historical indicator database; Step (3.3): Combine the input structured engineering constraint parameters and calculate the success threshold of the recommendation index through reasoning; Step (3.4): Output a list of evaluation indicators that includes the indicator name, calculation logic definition and success determination threshold.
5. The digital test scenario design method based on large-scale intelligent agent collaboration according to claim 1, characterized in that, Step (4) includes: Step (4.1): The design agent serializes the generated structured scenario configuration file into an intermediate format file without specific code logic, and transmits it to the code generation agent; Step (4.2): The code-generating agent parses the intermediate format file and calls the integrated sandbox runtime environment to attempt to build a simulation project; Step (4.3): If the build fails or a logical conflict occurs during execution, the code generation agent uses log analysis capabilities to generate structured feedback information containing error type, error location and conflict cause, and sends it to the design agent; Step (4.4): The designed intelligent agent parses the structured feedback information, searches the domain knowledge base to match the error cause, automatically modifies the conflict parameters in the intermediate format file, and returns to step (4.1) to re-serialize and transmit.
6. The digital test scenario design method based on large-scale intelligent agent collaboration according to claim 2, characterized in that, The equipment performance database stores the drone's aerodynamic parameters, battery life, maximum available hovering overload, and field of view (FOV) of the electro-optical pod.
7. The digital test scenario design method based on large-scale agent collaboration according to claim 1, characterized in that, The entities include multi-rotor drones and lidar obstacle avoidance modules.
8. The digital test scenario design method based on large-scale intelligent agent collaboration according to claim 2, characterized in that, The physical boundary conditions include the total length of the planned route and the radius of curvature of the turning nodes.
9. The digital test scenario design method based on large-scale intelligent agent collaboration according to claim 4, characterized in that, The evaluation index list includes result performance indicators and process constraint indicators; among which, result performance indicators include delivery position deviation and mission success rate, and process constraint indicators include maximum attitude deviation angle, maximum wind resistance overload, and remaining power during flight.
10. The digital test scenario design method based on large-scale intelligent agent collaboration according to claim 5, characterized in that, The intermediate state format file contains only a structured description of entity states, event sequences, and environmental parameters.