Human-computer interaction decision-making assistance system and method driven by knowledge and data hybrid

Abstract

The present application relates to a human-computer interaction decision-making support system and method driven by a hybrid of knowledge and data. The system includes four modules: a human-computer interaction module, an intelligent decision-making algorithm training module, a virtual action support environment module, and a knowledge module, as well as an external data interface design. In this system, human decision makers and intelligent algorithms can work together and complement each other. Human decision makers use their own experience and intuition to guide and adjust the learning of intelligent algorithms, while intelligent algorithms assist humans in making more accurate and scientific decisions by processing and analyzing large amounts of data. In this way, the human-computer interaction decision-making support system driven by a hybrid of knowledge and data can effectively cope with complex and changing decision-making environments and improve the quality and efficiency of decisions, especially in the field of action decisions with high risks and high uncertainties.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a human-computer interaction decision-making assistance system and method driven by a hybrid of knowledge and data. Background Art

[0002] In today's decision science field, data-driven intelligent algorithms are playing an increasingly important role. These algorithms, primarily based on big data, utilize machine learning, deep learning, and other methods to extract information, discover patterns, and predict the future, thereby assisting decision makers in making more scientific and accurate decisions. Data-driven intelligent algorithms have achieved remarkable success in many fields, such as business analytics, medical diagnosis, and market forecasting. Their strength lies in their ability to process and analyze large amounts of data, discover hidden patterns and trends within the data, and make predictions based on historical data.

[0003] However, despite the many advantages of data-driven intelligent algorithms, they also face obvious limitations in complex decision-making scenarios. First, such algorithms usually require a large amount of labeled data to train the model, while in many practical scenarios, especially in action command decision-making, it is very difficult to obtain large amounts of accurate labeled data. Second, data-driven methods often have difficulty handling highly complex and dynamically changing environments because they mainly rely on historical data, while real-world situations are constantly changing, making it difficult for models to be updated and adapted to new situations in real time. In addition, these algorithms often perform poorly in complex scenarios with unclear problem boundaries, unclear rules, and large problem scales, and are unable to provide effective decision support.

[0004] These challenges are even more pronounced in command and decision-making. Real-world operational scenarios are extremely complex, involving not only the coordination of multiple types of actors and the ever-changing dynamics of the two players, but also the influence of numerous uncertainties such as the geographical environment and weather conditions. These factors create a vast problem space rife with uncertainty and complexity. In such an environment, algorithms that rely solely on data-driven algorithms struggle to function effectively, as they cannot accurately capture and understand all these complex variables and struggle to adapt to rapidly changing circumstances. Summary of the Invention

[0005] To at least partially solve the above technical problems, this application proposes a knowledge-data hybrid-driven human-computer interactive decision-making support system, including:

[0006] Human-computer interaction module, used to obtain action task plans through human-computer interaction;

[0007] A virtual action support environment module is used to configure action task parameters using the action task plan to initialize a simulation action model that simulates the execution of the action task corresponding to the action task plan;

[0008] An intelligent decision-making algorithm training module is configured to acquire action rules and action knowledge; utilize the action knowledge and the action task plan to divide the action task corresponding to the action task plan into multiple subtasks; and utilize the action rules and the action knowledge to construct a behavior tree for each subtask; select an action node executable under the latest action state in the behavior tree of the subtask to be executed, and generate an action instruction corresponding to the action node, wherein the latest action state includes the latest action state during the simulated execution of the action task;

[0009] The virtual action support environment module is further used to input the action instructions generated by the intelligent decision algorithm training module into the simulation action model to simulate the execution of the subtask to be executed, and to feed back the latest action status after the execution of the subtask to be executed to the intelligent decision algorithm training module until the action task is completed;

[0010] The human-computer interaction module is also used to perform traceability analysis on the action data generated during the simulation execution of the action task by the virtual action support environment module, and to parse the action strategy generated by the intelligent decision-making algorithm training module. The action strategy includes the division result of the action task, the calculation behavior tree of the subtask and the action instruction, and outputs the traceability analysis result and the parsing result so that the user can adjust the action task plan and / or action rules and action knowledge according to the traceability analysis result and the parsing result.

[0011] This application also provides a knowledge-data hybrid-driven human-computer interaction decision-making assistance method, including:

[0012] Step 1: Obtain the action task plan through human-computer interaction;

[0013] Step 2: configuring action task parameters using the action task plan to initialize a simulation action model that simulates the execution of the action task corresponding to the action task plan;

[0014] Step 3: Acquire action rules and action knowledge;

[0015] Step 4: using the action knowledge and the action task plan to divide the action task corresponding to the action task plan into a plurality of subtasks;

[0016] Step 5: Construct a behavior tree for each subtask using the action rules and the action knowledge;

[0017] Step 6: Select an action node that can be executed in the latest action state on the behavior tree of the subtask to be executed. The latest action state includes the latest action state during the simulation execution of the action task.

[0018] Step 7: Generate an action instruction corresponding to the action node;

[0019] Step 8: Input the action instruction into the simulation action model to simulate the execution of the subtask to be executed, and return to step 6 until the action task is completed, and then execute step 9;

[0020] Step 9: Perform a traceability analysis on the action data generated during the simulation execution of the action task, and parse the action strategy, which includes the division result of the action task, the calculation behavior tree of the subtask and the action instruction. Output the traceability analysis result and the parsing result so that the user can adjust the action task plan and / or action rules, action knowledge according to the traceability analysis result and the parsing result. If the action task plan and / or action rules, action knowledge are adjusted, return to step 3.

[0021] In the systems and methods provided in the embodiments of this application, human decision makers and intelligent algorithms can work together and complement each other. Human decision makers use their experience and intuition to guide and adjust the learning of intelligent algorithms, while intelligent algorithms assist humans in making more accurate and scientific decisions by processing and analyzing large amounts of data. In this way, the human-computer interactive decision-making support system driven by a hybrid of knowledge and data can effectively cope with complex and changing decision-making environments and improve the quality and efficiency of decisions, especially in the field of high-risk and high-uncertainty action decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of the working principle of a knowledge data hybrid-driven human-computer interactive decision-making support system provided in one embodiment of the present application;

[0023] Figure 2 A flow chart of a method for a human-computer interaction decision-making assistance system driven by a hybrid of knowledge and data provided in one embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0025] Human decision makers can leverage their extensive experience, intuition, and creative thinking to solve ambiguous and uncertain problems, capabilities that algorithms relying solely on data struggle to match. Humans are able to rapidly generalize and abstract complex situations, using experience and intuition to rapidly narrow the solution space and identify critical and effective decision points. Furthermore, humans possess a unique ability to understand the diversity, variability, and irregularities of complex environments, enabling them to flexibly adapt to new situations and engage in creative problem-solving. Therefore, combining human knowledge and experience with data-driven algorithms not only leverages the algorithms' ability to process and analyze large amounts of data, but also leverages human intuition and experience to narrow the solution space, improving the quality and efficiency of decision-making. This knowledge- and data-driven, human-machine interactive decision-making system, by integrating human intuitive judgment with machine data processing capabilities, can more effectively address complex and dynamically changing decision-making environments, particularly in high-risk and high-uncertainty action decisions.

[0026] Therefore, in response to the limitations of current data-driven intelligent decision-making algorithms in complex confrontation scenarios with unclear problem boundaries, unclear rules, and large-scale problems, this application constructs a human-computer interaction decision-making support system architecture through a hybrid knowledge and data-driven model. The core of this system is to leverage the complementary advantages of human experience and intuition and the data processing capabilities of machines to achieve more accurate, flexible, and efficient decision support. By effectively integrating the two capabilities of human adaptability and machine calculation, this system not only improves the accuracy and efficiency of decision-making, but also increases the transparency and understandability of the decision-making process, so that it can be parsed and used by various action control information systems to provide auxiliary support for related action decision-making activities.

[0027] The knowledge data hybrid driven human-computer interaction decision-making assistance system and method provided in the embodiments of the present application can be applied to, but not limited to, drone action scenarios.

[0028] Specifically, an embodiment of the present application provides a knowledge-data hybrid-driven human-computer interaction decision-making support system, including a human-computer interaction module, an intelligent decision-making algorithm training module, and a virtual action support environment module, and optionally, also includes a knowledge module and an external data interface.

[0029] The functions and implementation methods of each module are described in detail below.

[0030] Human-computer interaction module, this module is used to acquire, analyze and display data through human-computer interaction. Specifically, the human-computer interaction module is used to acquire action task plans through human-computer interaction; it is also used to trace the action data generated during the simulation of the action task by the virtual action support environment module, and to parse the action strategy generated by the intelligent decision-making algorithm training module. The action strategy includes the division result of the action task, the calculation behavior tree of the subtask and the action instruction, and outputs the traceability analysis result and the parsing result so that the user can adjust the action task plan and / or action rules and action knowledge according to the traceability analysis result and the parsing result. By way of example and not limitation, the functions of the human-computer interaction module include: action task plan editing, action process display, action plan evaluation, action case recommendation and data traceability, etc.

[0031] The virtual action support environment module is used to configure the action task parameters using the action task plan to initialize a simulation action model that simulates the execution of the action task corresponding to the action task plan. The action task parameters include map data of the environment where the action task is located, simulation model parameters for the simulated execution of the action task, instruction rule encoding, time management configuration parameters, and task management configuration parameters. Furthermore, the map data may include detailed terrain and weather condition map data. The virtual action support environment module is also used to input the action instructions generated by the intelligent decision algorithm training module into the simulation action model to simulate the execution of the subtask to be executed, and to feed back the latest action status after the execution of the subtask to be executed to the intelligent decision algorithm training module until the action task is completed.

[0032] The intelligent decision-making algorithm training module is used to obtain action rules and action knowledge; use the action knowledge and the action task plan to divide the action task corresponding to the action task plan into multiple subtasks, and use the action rules and the action knowledge to construct a behavior tree for each subtask; select an executable action node in the latest action state in the behavior tree of the subtask to be executed, and generate an action instruction corresponding to the action node, where the latest action state includes the latest action state during the simulated execution of the action task.

[0033] The knowledge module is used to store action rules, action knowledge, and action task plans.

[0034] External data interfaces, which primarily include the human-computer interaction module interface, the intelligent decision-making algorithm training module interface, the virtual action support environment module interface, and the knowledge module interface, are used to complete system integration and ensure that various modules and algorithms can work together seamlessly.

[0035] The embodiment of the present application designs an external data interface to standardize and format all data passing through the interface to ensure that data can be transmitted and interacted between various modules in the system. The external data interface of the embodiment of the present application is shown in Table 1:

[0036] Table 1 Main system interfaces

[0037]

[0038] In a more specific embodiment, the architecture and function implementation process of the knowledge data hybrid driven human-computer interaction decision support system provided by the embodiment of the present application are as follows: Figure 1 shown.

[0039] The human-computer interaction module includes a solution editing submodule, a solution evaluation submodule, a data traceability submodule, an action process display submodule, and a case recommendation submodule. Furthermore, a human-computer interaction interface can be used to provide functional access to each of these submodules, allowing users to select the desired submodule function through the interface.

[0040] After the user selects the plan editing function, the human-computer interaction module calls the plan editing interface, obtains the action task plan input by the user through human-computer interaction, and outputs the action task plan input by the user to the virtual action support environment module and the intelligent decision algorithm training module respectively. In the embodiment of the present application, the action task plan includes but is not limited to the identification of the action participants, the scale of the action participants, the type of action means of the action participants, the type of action tools of the action participants, the goals of the action participants, the action restrictions of the action participants, etc. After the user selects the plan evaluation function, the human-computer interaction module calls the plan evaluation interface, evaluates the action task plan, and visually displays the evaluation results.

[0041] After the user selects the case traceability function, the human-computer interaction module invokes the case traceability interface to trace the action plan and visually display the traceability results. By way of example and not limitation, evaluating the action plan includes analyzing the action strategy corresponding to the action plan, and the evaluation results include the analysis results.

[0042] After the user selects the action process display function, the human-computer interaction module calls the action process display interface to visualize the action process. Specifically, the received action data is visualized in real time so that the user can intuitively understand the execution status of the action task.

[0043] After the user selects the case recommendation function, the human-computer interaction module calls the case recommendation interface, filters and displays recommended cases according to predetermined conditions and strategies. The present embodiment does not limit the conditions and strategies for case recommendation. In actual applications, the conditions and strategies for case recommendation can be determined as needed.

[0044] The knowledge module includes a basic database, an action task plan library and a calculation tool library. Among them, the basic database stores the basic data required for the operation of the system. This application does not specifically limit the content of the basic data. In actual applications, the basic data and the basic database can be defined as needed. The action task plan library stores historical action task plans. The calculation tool library stores tools such as action rules and action knowledge. On the one hand, the knowledge module receives and saves the updated action task plans, updated action knowledge, and updated action rules output by the human-computer interaction module, and on the other hand, outputs action rules and action knowledge to the intelligent decision-making algorithm training module. Among them, the action task plan is used to guide the action strategy of the entities on both sides of the action, the action rules define the constraints on the entity behavior, and the action knowledge accumulates past action experience and action layout.

[0045] The virtual action support environment module includes a map data submodule, a time management submodule, a command rule encoding submodule, a task management submodule, and a simulation action model. The virtual action support environment module outputs action states to the intelligent decision-making algorithm training module and receives action commands from the intelligent decision-making algorithm training module. The action states include the state of the action after the corresponding action command is executed, including map data, the size of the action participants, the types of action methods used by the action participants, and the types of action tools used by the action participants.

[0046] The intelligent decision-making algorithm training module includes a behavioral component, an adversarial data management model, and a learning algorithm development framework. It obtains action status and map data from the virtual action support environment module, action task plans from the human-computer interaction module, and action rules and action knowledge from the knowledge module. It then outputs action instructions to the virtual action support environment module and strategy plans and action process data to the human-computer interaction module.

[0047] The virtual action support environment module inputs the action instructions generated by the intelligent decision algorithm training module into the simulated action model to simulate the execution of the pending subtask. It then feeds the latest action status after executing the pending subtask back to the intelligent decision algorithm training module. The virtual action support environment module and the intelligent decision algorithm training module repeat this process until the action task is completed.

[0048] The human-computer interaction module performs traceability analysis on the action data generated by the virtual action support environment module during the simulation of the action task, and parses the action strategy generated by the intelligent decision-making algorithm training module. The action strategy includes the division results of the action task, the calculation behavior tree and action instructions of the subtask, and outputs the traceability analysis results and parsing results so that the user can adjust the action task plan and / or action rules and action knowledge based on the traceability analysis results and parsing results.

[0049] In the disclosed embodiment, the action task parameters include map data of the environment where the action task is located, simulation model parameters for simulating the execution of the action task, instruction rule coding, time management configuration parameters, and task management configuration parameters.

[0050] In an embodiment of the present disclosure, the action task corresponding to the action task plan is divided into multiple subtasks using the action knowledge and the action task plan, including: inputting the action task into a pre-trained task division model to obtain multiple subtasks output by the task division model, and the task division model is trained using the action knowledge.

[0051] In an embodiment of the present disclosure, selecting an executable action node in a latest action state on a behavior tree of a subtask to be executed includes: inputting the behavior tree of the subtask to be executed and the latest action state into a pre-trained action node selection model, and obtaining the executable action node in the latest action state output by the action node selection model.

[0052] The present application also provides a knowledge data hybrid driven human-computer interaction decision-making method, the specific implementation of which is as follows: Figure 2 As shown, the specific process is:

[0053] Step 1: Obtain the action task plan through human-computer interaction;

[0054] Step 2: configuring action task parameters using the action task plan to initialize a simulation action model that simulates the execution of the action task corresponding to the action task plan;

[0055] Step 3: Acquire action rules and action knowledge;

[0056] Step 4: using the action knowledge and the action task plan to divide the action task corresponding to the action task plan into a plurality of subtasks;

[0057] Step 5: Construct a behavior tree for each subtask using the action rules and the action knowledge;

[0058] Step 6: Select an executable action node in the latest action state on the behavior tree of the subtask to be executed. The latest action state includes the latest action state during the simulation execution of the action task.

[0059] Step 7: Generate an action instruction corresponding to the action node;

[0060] Step 8: Input the action instruction into the simulation action model to simulate the execution of the subtask to be executed, and return to step 6 until the action task is completed, and then execute step 9;

[0061] Step 9: Perform a traceability analysis on the action data generated during the simulation execution of the action task, and parse the action strategy generated by the intelligent decision-making algorithm training module. The action strategy includes the division result of the action task, the calculation behavior tree of the subtask and the action instruction. Output the traceability analysis result and the parsing result so that the user can adjust the action task plan and / or action rules, action knowledge according to the traceability analysis result and the parsing result. If the action task plan and / or action rules, action knowledge are adjusted, return to step 3.

[0062] In a more specific implementation, the knowledge data hybrid-driven human-computer interaction decision-making assistance method provided in the embodiment of the present application is implemented as follows:

[0063] The user uses the human-computer interaction module to edit the action plan, inputs the action plan into the virtual action support environment module through the external data interface between the human-computer interaction module and the virtual action support environment module, and inputs the action plan into the intelligent decision algorithm training module through the external data interface between the human-computer interaction module and the intelligent decision algorithm training module.

[0064] The virtual action support environment module instantiates a simulated action scenario. Specifically, it sets various action task parameters based on the received action plan, including map data, the scale and type of action participants, the expected action effect of the own side, the main actions of the other side, etc.

[0065] The intelligent decision algorithm training module receives the latest action status and map data from the virtual action support environment module, receives action rules, action knowledge, action instructions, etc. from the knowledge module, and receives action plans from the human-computer interaction module;

[0066] The intelligent decision-making algorithm training module divides the action tasks corresponding to the action plan into multiple subtasks based on the action plan and action knowledge. .

[0067] The intelligent decision-making algorithm training module constructs a Build a behavior tree. This application embodiment does not limit the specific construction process of the behavior tree. In actual application, the behavior tree can be constructed using existing behavior tree construction technology as needed. The behavior tree can be a two-layer behavior tree or a multi-layer behavior tree. In the embodiments of the present application, the subtasks may have a serial timing relationship, a parallel timing relationship, or a combination of serial and parallel timing relationships (i.e., some subtasks have a serial timing relationship, while others have a parallel timing relationship). In addition, all subtasks may need to be executed, or some subtasks may be conditionally triggered and not executed if the triggering condition is not met during the action.

[0068] The intelligent decision-making algorithm training module uses the latest action state as the basis for determining the conditional nodes of the behavior tree, outputs the action nodes that can be executed under the current latest action state, and generates action instructions from the action nodes. The action nodes that can be executed under the current latest action state can be action nodes for the current subtask or action nodes for other subtasks.

[0069] The intelligent decision-making algorithm training module transmits action commands to the virtual action environment, driving the execution of the simulated action model. After the simulated action model executes an action command, it sends data to the intelligent decision-making algorithm training module, including the latest action status and map data. The intelligent decision-making algorithm manages the data generated during the action using the adversarial data management model.

[0070] The intelligent decision-making algorithm training module transmits the intelligent algorithm strategy or plan, as well as the action process data to the human-computer interaction module.

[0071] The human-computer interaction module conducts traceability analysis on the action process data, parses the intelligent algorithm solution for user reference, and passes the knowledge or rules condensed from the analysis to the knowledge module.

[0072] Users provide feedback based on the presented options and revise their action plans. By continuously optimizing behavior trees and reinforcement learning models, we ensure that the decision-making process continues to evolve and adapt to the changing environment.

[0073] In addition, the system is extensively tested to ensure that it can provide reliable and effective decision support in a variety of operational environments.

[0074] The knowledge and data hybrid driven human-computer interaction decision-making assistance system and method provided in the embodiments of the present application are applied to drone action scenarios, and can fully utilize the advantages of the decision-making system driven by both knowledge and data to make drone action decisions.

[0075] In an application scenario based on drone action decision-making, the method provided by the embodiment of the present application is as follows:

[0076] At the UAV action decision-making and control end, the UAV action task plan is obtained through human-computer interaction; at the decision-making and control end, the UAV action task plan is used to configure the UAV action task parameters to initialize the simulation action model for simulating the execution of the action task corresponding to the UAV action task plan; at the decision-making and control end, the UAV action rules and UAV action knowledge are obtained, and the action task corresponding to the action task plan is divided into multiple subtasks using the UAV action knowledge and the UAV action task plan; at the decision-making and control end, the UAV action rules and UAV action knowledge are used to construct a behavior tree for each subtask, and the action node that can be executed under the latest action state is selected on the behavior tree of the subtask to be executed, and the latest action state includes the latest action state during the simulation execution of the action task .... The control end generates action instructions corresponding to the action node; the decision control end inputs the action instructions into the UAV simulation action model to simulate the execution of the subtask to be executed; the above operations are repeated for each subtask until the action task is completed; the action data generated during the simulation execution of the action task is traced and analyzed at the decision control end, and the UAV action strategy is parsed. The UAV action strategy includes the division results of the action task, the calculation behavior tree and action instructions of the subtask, and the traceability analysis results and parsing results are output so that the user can adjust the action UAV task plan and / or UAV action rules, UAV action knowledge according to the traceability analysis results and parsing results. When the UAV action task plan and / or UAV action rules, UAV action knowledge are adjusted, the above processing process is repeated.

[0077] The technical effects of this application are described as follows:

[0078] (1) In this technical solution, the knowledge-data hybrid-driven human-computer interaction decision-making support system uses the idea of ​​intelligence and modularization to establish a module for each function such as simulation deduction, human-computer interaction, intelligent decision-making algorithm, and knowledge storage. The four modules are integrated into a whole using standardized interfaces to form a complete system architecture, providing intelligent decision-making support information for user decision-making.

[0079] (2) In this technical solution, the knowledge-data hybrid-driven human-computer interaction decision-making support system is a closed-loop decision-making system, in which the intelligent decision-making algorithm module can perform offline training of the model. Users input manual decision results and knowledge into the intelligent decision-making algorithm training process through the human-computer interaction module, continuously optimizing the model to achieve the goal of improving the efficiency and accuracy of decision-making support.

[0080] (3) In this technical solution, the system has excellent adaptability and can adjust its support strategy according to different decision-making scenarios and user needs. Whether it is a change in the opponent's strategy, the input of new action states, or changes in the environment, the system can flexibly adjust to ensure the provision of the most appropriate decision support.

[0081] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A knowledge-data hybrid-driven human-computer interactive decision-making support system, characterized by: include: Human-computer interaction module, used to obtain action task plans through human-computer interaction; A virtual action support environment module is used to configure action task parameters using the action task plan to initialize a simulation action model that simulates the execution of the action task corresponding to the action task plan; Intelligent decision-making algorithm training module, used to acquire action rules and action knowledge; Dividing the action task corresponding to the action task plan into a plurality of subtasks using the action knowledge and the action task plan, and constructing a behavior tree for each subtask using the action rules and the action knowledge; Selecting an executable action node in the latest action state on the behavior tree of the subtask to be executed, and generating an action instruction corresponding to the action node, wherein the latest action state includes the latest action state during the simulation execution of the action task; The virtual action support environment module is further configured to input the action instructions generated by the intelligent decision-making algorithm training module into the simulation action model to simulate the execution of the subtask to be executed, and to feed back the latest action status after the execution of the subtask to be executed to the intelligent decision-making algorithm training module until the action task is completed; The human-computer interaction module is also used to perform traceability analysis on the action data generated during the simulation execution of the action task by the virtual action support environment module, and to parse the action strategy generated by the intelligent decision-making algorithm training module. The action strategy includes the division result of the action task, the calculation behavior tree of the subtask and the action instruction, and outputs the traceability analysis result and the parsing result so that the user can adjust the action task plan and / or action rules and action knowledge according to the traceability analysis result and the parsing result.

2. The system according to claim 1, wherein: The action task parameters include map data of the environment where the action task is located, simulation model parameters of the action task simulation execution, instruction rule coding, time management configuration parameters, and task management configuration parameters.

3. The system according to claim 1, wherein: The step of utilizing the action knowledge and the action task plan to divide the action task corresponding to the action task plan into a plurality of subtasks includes: The action task is input into a pre-trained task division model to obtain a plurality of subtasks output by the task division model, wherein the task division model is trained using the action knowledge.

4. The system according to claim 1, wherein: The step of selecting an executable action node in the latest action state on the behavior tree of the subtask to be executed includes: The behavior tree of the subtask to be executed and the latest action state are input into a pre-trained action node selection model to obtain an executable action node under the latest action state output by the action node selection model.

5. The system according to claim 1, wherein: The system further comprises a knowledge module for storing action rules, action knowledge and action task plans.

6. The system according to claim 1, wherein: The system further comprises an external data interface, which is used for data communication between various modules.

7. A knowledge-data hybrid-driven human-computer interaction decision-making method, characterized in that: include: Step 1: Obtain the action task plan through human-computer interaction; Step 2: configuring action task parameters using the action task plan to initialize a simulation action model that simulates the execution of the action task corresponding to the action task plan; Step 3: Acquire action rules and action knowledge; Step 4: using the action knowledge and the action task plan to divide the action task corresponding to the action task plan into a plurality of subtasks; Step 5: Construct a behavior tree for each subtask using the action rules and the action knowledge; Step 6: Select an action node that can be executed in the latest action state on the behavior tree of the subtask to be executed. The latest action state includes the latest action state during the simulation execution of the action task. Step 7: Generate an action instruction corresponding to the action node; Step 8: Input the action instruction into the simulation action model to simulate the execution of the subtask to be executed, and return to step 6 until the action task is completed, and then execute step 9; Step 9: Perform a traceability analysis on the action data generated during the simulation execution of the action task, and parse the action strategy, which includes the division result of the action task, the calculation behavior tree of the subtask and the action instruction. Output the traceability analysis result and the parsing result so that the user can adjust the action task plan and / or action rules, action knowledge according to the traceability analysis result and the parsing result. If the action task plan and / or action rules, action knowledge are adjusted, return to step 3.

8. The method according to claim 7, characterized in that The action task parameters include map data of the environment where the action task is located, simulation model parameters of the action task simulation execution, instruction rule coding, time management configuration parameters, and task management configuration parameters.

9. The method according to claim 7, characterized in that The step of utilizing the action knowledge and the action task plan to divide the action task corresponding to the action task plan into a plurality of subtasks includes: The action task is input into a pre-trained task division model to obtain a plurality of subtasks output by the task division model, wherein the task division model is trained using the action knowledge.

10. The method according to claim 7, characterized in that The step of selecting an executable action node in the latest action state on the behavior tree of the subtask to be executed includes: The behavior tree of the subtask to be executed and the latest action state are input into a pre-trained action node selection model to obtain an executable action node under the latest action state output by the action node selection model.

Images