An autonomous decision-making method for aircraft based on a hybrid architecture of finite state machine and behavior tree

Abstract

The application relates to the technical field of aircraft autonomous decision-making, in particular to an aircraft autonomous decision-making method based on a hybrid architecture of a finite state machine and a behavior tree, which comprises the following steps: acquiring multi-source perception data of an aircraft, and writing the multi-source perception data into a shared data center; monitoring a macro state set in the shared data center by using a finite state machine, determining a current macro state according to the macro state set, and indexing and loading a corresponding behavior tree model from a preset behavior tree library by using the macro state set; taking a current control period as a time step, starting from a root node to traverse and execute the behavior tree model, outputting a leaf node action instruction corresponding to an autonomous decision-making action based on the behavior tree model, and controlling the aircraft to execute the autonomous decision-making action based on the leaf node action instruction. Therefore, the problems of state explosion of a finite state machine and insufficient macro control of a pure behavior tree in the related art are solved.

Description

Technical Field

[0001] This application relates to the field of aircraft autonomous decision-making technology, and in particular to an aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree. Background Technology

[0002] As the autonomy and intelligence of unmanned aerial vehicles (UAVs) continue to improve, higher demands are placed on their autonomous decision-making capabilities in complex environments. Currently, most UAVs employ a single architecture for autonomous decision-making. Among related technologies, finite state machines are prone to problems such as state explosion, complex logic, and poor scalability in complex mission scenarios. They struggle to simultaneously manage macro-level missions and execute micro-level actions, resulting in insufficient decision-making efficiency and robustness, and failing to meet the stable autonomous decision-making requirements in highly dynamic flight scenarios. Summary of the Invention

[0003] This application provides an aircraft autonomous decision-making method, device, electronic device, and storage medium based on a hybrid architecture of finite state machine and behavior tree, in order to solve the problems of state explosion in finite state machines and insufficient macroscopic control of pure behavior trees in related technologies.

[0004] The first aspect of this application provides an autonomous decision-making method for aircraft based on a hybrid architecture of finite state machine and behavior tree, comprising the following steps: acquiring multi-source perception data of the aircraft and writing the multi-source perception data into a shared data center; using a finite state machine to monitor a set of macro states in the shared data center, determining the current macro state based on the set of macro states, indexing and loading the corresponding behavior tree model from a pre-set behavior tree library using the set of macro states; traversing and executing the behavior tree model starting from the root node with the current control cycle as the time step, outputting leaf node action instructions corresponding to the autonomous decision-making action based on the behavior tree model, and controlling the aircraft to execute the autonomous decision-making action based on the leaf node action instructions.

[0005] Based on the aforementioned technical means, this application embodiment utilizes a hybrid architecture of finite state machines and behavior trees for collaborative processing. The finite state machine enables macroscopic state monitoring and scheduling, avoiding the problems of state explosion and complex logic associated with single state machines in related technologies. Simultaneously, it relies on dynamic matching and loading of behavior tree models based on macroscopic states to compensate for the shortcomings of pure behavior trees in macroscopic task flow and state maintenance. Combined with a shared data center for unified management of perception data, the decision-making logic is clearly layered and highly modular, effectively improving the real-time performance, stability, and scalability of aircraft autonomous decision-making, and ensuring efficient and reliable decision execution in complex flight scenarios.

[0006] Optionally, the multi-source sensing data includes real-time flight status data of the aircraft, airborne sensor detection data, and data link instructions. Writing the multi-source sensing data into a shared data center includes: constructing a data domain structure for the shared data center, the data domain structure including state vector sets for real-time flight status data, airborne sensor detection data, and data link instructions respectively; mapping the multi-source sensing data into a timestamp-aligned sequence in a target format; determining the state vector sets for real-time flight status data, airborne sensor detection data, and data link instructions based on the timestamp-aligned sequence; and writing the state vector sets into the shared data center.

[0007] Based on the above technical means, the embodiments of this application can achieve unified storage and efficient management of real-time flight status data, airborne sensor detection data and data link instructions by constructing a standardized data domain structure and performing format mapping and timestamp alignment processing on multi-source sensing data. This ensures that the timing of each sensing data is consistent and the format is standardized, providing accurate and reliable data support for subsequent macro-state monitoring and behavior tree decision execution, and improving the accuracy and stability of autonomous decision-making.

[0008] Optionally, the finite state machine includes a topology, a state transition function, flow logic, and a macroscopic set of states. The macroscopic set of states includes takeoff, landing, hovering, waypoint flight, game-theoretic, interception, and refueling states. The topology of the finite state machine includes a finite set of states, a set of events that trigger state transitions, and a state transition function. The state transition function is... The flow logic is as follows ,in, For a finite set of states, For a set of events, For the state at the next moment, The current state. The event conditions that trigger state transitions are defined as follows: A finite state machine includes a topology, a state transition function, transition logic, and a macroscopic set of states. The macroscopic set of states includes takeoff, landing, circling, waypoint flight, game-theoretic, interception, and refueling states. The finite state machine's topology includes a finite set of states, a set of events that trigger state transitions, and a state transition function. The state transition function is... The flow logic is as follows ,in, For a finite set of states, For a set of events, For the state at the next moment, The current state. The event conditions that trigger the state transition.

[0009] Based on the above technical means, the embodiments of this application define the topology, state transition function and strict flow logic of the finite state machine in a standardized manner, clarify multiple macroscopic flight states and state triggering conditions, and can realize the accurate identification, stable flow and reliable switching of the macroscopic mission state of the aircraft. This provides a clear and definite scheduling basis for the index loading of the behavior tree model, and improves the logic, standardization and robustness of the entire hybrid decision architecture.

[0010] Optionally, the behavior tree library includes a behavior tree model constructed for each macro state. Dynamic switching and context activation of the behavior tree model are achieved through the current macro state. The behavior tree model represents the decision logic structure of the current macro state, which includes a mapping function structure, a graph theory structure, sequence control nodes, and selection control nodes. The mapping function structure includes the mapping relationship between the behavior tree library and the macro state set. For any given moment, the activated behavior tree model is uniquely indexed by the current macro state. Dynamic switching of the behavior tree model is achieved based on the mapping relationship and the unique index. The directed acyclic graph includes a root node, a set of control flow nodes, and a set of leaf nodes. Sequence control nodes are used to control takeoff and refueling states. When all child nodes return a success flag in sequence, the parent node returns a success flag. Selection control nodes are used for maneuver decisions in the game state. They traverse child nodes according to priority, and when a child node returns a success flag or a running flag, the parent node stops traversing subsequent child nodes.

[0011] Based on the above technical means, the embodiments of this application achieve accurate matching and rapid activation of decision logic under different task scenarios through one-to-one mapping and dynamic switching between macro states and behavior tree models; at the same time, the use of sequence, selection and other control nodes to distinguish and adapt the execution logic of different macro states makes the decision structure modular, hierarchical and efficient, which not only ensures the determinism and real-time nature of state switching, but also improves the reliability and maintainability of decision logic under complex maneuver and task scenarios.

[0012] Optionally, the leaf node action instructions corresponding to the autonomous decision-making actions are output based on the behavior tree model, including: if the aircraft is in a game-like adversarial state, the leaf node action instructions include executing parallel control logic, which includes: simultaneously running parallel sub-nodes including radar control, infrared control, electronic warfare control, target update and attack decision, and using the selection node to execute chaff release, tailing, return to base or tracking actions according to the battlefield situation; if the aircraft is in a takeoff state, landing state or refueling state, the leaf node action instructions include executing sequence control logic, which includes executing a programmed action sequence including preparation actions, taxiing actions, landing gear retraction and extension actions, docking actions and flight actions according to the time constraints.

[0013] Based on the above technical means, the embodiments of this application can adaptively adopt parallel control or sequential control logic for different macroscopic states, realize multi-sensor and multi-task parallel decision-making and rapid selection of maneuver actions in game-like confrontation state, and strictly execute programmed actions according to the time sequence in takeoff, landing, refueling and other states, which not only improves the real-time performance and flexibility of decision-making in complex situations, but also ensures the standardization and reliability of key mission processes.

[0014] Optionally, controlling the aircraft to perform autonomous decision-making actions based on leaf node action commands includes: parsing the leaf node action commands to determine the aircraft's aerodynamic control surface deflection signal, engine thrust control signal, and load actuation signal, and controlling the aircraft to perform autonomous decision-making actions based on the aerodynamic control surface deflection signal, engine thrust control signal, and load actuation signal.

[0015] Based on the above-mentioned technical means, the embodiments of this application clarify the specific control signals of the aircraft aerodynamic control surfaces, engine thrust and load by parsing the leaf node action commands, so as to realize the accurate conversion of autonomous decision-making actions from command output to actual execution, ensure that the decision commands can be implemented efficiently and accurately, ensure the coordinated response of various aircraft components, further improve the stability and accuracy of autonomous decision-making action execution, and adapt to the flight control requirements under different macroscopic conditions.

[0016] Optionally, parsing the leaf node action commands includes: obtaining a multi-physics domain command mapping model; inputting the leaf node action commands into the multi-physics domain command mapping model, mapping them into flight control commands and mission payload commands through the multi-physics domain command mapping model; calculating aerodynamic control surface deflection signals and engine thrust control signals based on the flight control commands, and calculating payload actuation signals based on the mission payload commands.

[0017] Based on the above technical means, the embodiments of this application use a multi-physical domain command mapping model to accurately map leaf node action commands into two types of commands: flight control and mission payload. Then, the corresponding control signals are calculated separately, realizing the standardized and accurate conversion of decision commands to equipment execution signals, avoiding command parsing deviations, ensuring the coordinated and accurate response of aerodynamic control surfaces, engines, and mission payloads, and further improving the reliability and accuracy of autonomous decision-making action execution.

[0018] A second aspect of this application provides an autonomous decision-making device for aircraft based on a hybrid architecture of finite state machine and behavior tree, comprising: an acquisition module for acquiring multi-source perception data of the aircraft and writing the multi-source perception data into a shared data center; a loading module for using a finite state machine to monitor a set of macro-states in the shared data center, determining the current macro-state based on the set of macro-states, and indexing and loading the corresponding behavior tree model from a pre-set behavior tree library using the set of macro-states; and a control module for traversing and executing the behavior tree model from the root node with the current control cycle as the time step, outputting leaf node action instructions corresponding to the autonomous decision-making action based on the behavior tree model, and controlling the aircraft to execute the autonomous decision-making action based on the leaf node action instructions.

[0019] Optionally, the multi-source sensing data includes real-time flight status data of the aircraft, airborne sensor detection data, and data link instructions. The acquisition module is further used to: construct a data domain structure for the shared data center, the data domain structure including the state vector sets of real-time flight status data, airborne sensor detection data, and data link instructions respectively; map the multi-source sensing data into a timestamp-aligned sequence in a target format; determine the state vector sets of real-time flight status data, airborne sensor detection data, and data link instructions based on the timestamp-aligned sequence; and write the state vector sets into the shared data center.

[0020] Optionally, the finite state machine includes a topology, a state transition function, flow logic, and a macroscopic set of states. The macroscopic set of states includes takeoff, landing, hovering, waypoint flight, game-theoretic, interception, and refueling states. The topology of the finite state machine includes a finite set of states, a set of events that trigger state transitions, and a state transition function. The state transition function is... The flow logic is as follows ,in, For a finite set of states, For a set of events, For the state at the next moment, The current state. The event conditions that trigger the state transition.

[0021] Optionally, the behavior tree library includes a behavior tree model constructed for each macro state. Dynamic switching and context activation of the behavior tree model are achieved through the current macro state. The behavior tree model represents the decision logic structure of the current macro state, which includes a mapping function structure, a graph theory structure, sequence control nodes, and selection control nodes. The mapping function structure includes the mapping relationship between the behavior tree library and the macro state set. For any given moment, the activated behavior tree model is uniquely indexed by the current macro state. Dynamic switching of the behavior tree model is achieved based on the mapping relationship and the unique index. The directed acyclic graph includes a root node, a set of control flow nodes, and a set of leaf nodes. Sequence control nodes are used to control takeoff and refueling states. When all child nodes return a success flag in sequence, the parent node returns a success flag. Selection control nodes are used for maneuver decisions in the game state. They traverse child nodes according to priority, and when a child node returns a success flag or a running flag, the parent node stops traversing subsequent child nodes.

[0022] Optionally, the control module is further configured to: if the aircraft is in a game-like confrontation state, determine that the leaf node action instructions include executing parallel control logic, which includes: simultaneously running parallel sub-nodes including radar control, infrared control, electronic warfare control, target update and attack decision, and using the selected node to execute chaff release, tail placement, return to base or tracking actions according to the battlefield situation; if the aircraft is in a takeoff state, landing state or refueling state, determine that the leaf node action instructions include executing sequence control logic, which includes executing a programmed sequence of actions including preparation actions, taxiing actions, landing gear retraction and extension actions, docking actions and flight actions according to the timing constraints.

[0023] Optionally, the control module is further configured to: parse the leaf node action commands to determine the aircraft's aerodynamic control surface deflection signal, engine thrust control signal, and load actuation signal, and control the aircraft to perform autonomous decision-making actions based on the aerodynamic control surface deflection signal, engine thrust control signal, and load actuation signal.

[0024] Optionally, the control module is further configured to: acquire a multi-physics domain command mapping model; input leaf node action commands into the multi-physics domain command mapping model, and map them into flight control commands and mission payload commands through the multi-physics domain command mapping model; calculate aerodynamic control surface deflection signals and engine thrust control signals based on flight control commands, and calculate payload actuation signals based on mission payload commands.

[0025] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree as described in the above embodiments.

[0026] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree as described in the above embodiments.

[0027] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0028] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0029] Figure 1 The flowchart shows an aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree provided according to an embodiment of this application.

[0030] Figure 2 This is a schematic diagram of the macroscopic state transition topology of the finite state machine provided in the embodiments of this application;

[0031] Figure 3 This is a schematic diagram of the parallel behavior tree logic structure for multi-dimensional decision-making involving maneuvering, weapons, and sensors under game states, according to an embodiment of this application.

[0032] Figure 4 This is a schematic diagram of the behavior tree logic structure under the aerial refueling state provided in the embodiments of this application;

[0033] Figure 5 This is a schematic diagram of the behavior tree logic structure under autonomous takeoff state provided in the embodiments of this application;

[0034] Figure 6 This is a schematic diagram of the behavior tree logic structure under autonomous landing state according to an embodiment of this application;

[0035] Figure 7 This is a flowchart of an aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree according to an embodiment of this application;

[0036] Figure 8 This is a block diagram illustrating an aircraft autonomous decision-making device based on a hybrid architecture of finite state machine and behavior tree, according to an embodiment of this application.

[0037] Figure 9 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0038] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0039] Currently, most autonomous decision-making systems for unmanned aerial vehicles (UAVs) adopt a single control architecture of finite state machines or behavior trees. These technologies have the following shortcomings in multi-task modal autonomous decision-making under complex flight environments:

[0040] (1) In related technologies, finite state machines are prone to "state explosion" when dealing with micro-complex maneuvering tasks. The models are large and difficult to debug, and their scalability and robustness are extremely poor.

[0041] (2) Pure behavior trees lack the implicit memory of long-term macro task states. When dealing with time-constrained tasks, a large number of complex blackboard variables need to be introduced, which destroys the simplicity of the structure and easily confuses the decision logic.

[0042] (3) The relevant technical methods are difficult to achieve parallel coordination of multi-dimensional decision-making elements such as flight maneuvering, load control and sensor management, which can easily lead to action conflicts and reduce mission execution efficiency.

[0043] In summary, the relevant technologies suffer from problems such as unreasonable architecture design, poor decision adaptability, and insufficient mission execution efficiency, making it difficult to meet the needs of high-performance unmanned aerial vehicles for fully autonomous situational awareness, mission planning, and decision execution in complex environments with high dynamics, strong competition, and incomplete information.

[0044] The following description, with reference to the accompanying drawings, outlines an aircraft autonomous decision-making method, apparatus, electronic device, and storage medium based on a hybrid architecture of finite state machines and behavior trees, representing embodiments of this application. Addressing the issues raised in the background section regarding the limitations of single control architectures in constructing multi-task modal, highly dynamic aircraft autonomous decision-making systems, such as "state explosion," difficulty in maintaining long-cycle mission logic, severe coupling of maneuvering-firepower-sensor multi-dimensional decision-making, and the difficulty in simultaneously managing the entire mission process and achieving refined decision-making, this application provides an aircraft autonomous decision-making method based on a hybrid architecture of finite state machines and behavior trees. This method employs a hierarchical governance and concern separation strategy, constructing a two-layer decision-making architecture consisting of a top-level state machine and a bottom-level behavior tree. The finite state machine manages seven macro-level mission states, including takeoff and landing, and enables automatic transitions. Within each state node, micro-level tactical logic is constructed using a behavior tree, with game states embedded in the game behavior tree—containing three types of decisions: maneuvering, weaponry, and sensors. A shared blackboard mechanism and a hybrid logic topology strategy are used to decouple decision logic and achieve closed-loop control, combining the advantages of both to overcome the shortcomings of a single control model. Simultaneously, a game-state hybrid reasoning model addresses the multi-dimensional resource coordination problem, ensuring that the aircraft maintains both macro-level mission stability and micro-level decision-making flexibility in complex scenarios. This enhances the aircraft's autonomous adaptability and decision-making efficiency in multi-mission scenarios, achieving efficient and autonomous control throughout the entire process in complex battlefield environments, demonstrating significant engineering practical value.

[0045] Specifically, Figure 1 This is a flowchart illustrating an aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree, provided in an embodiment of this application.

[0046] like Figure 1 As shown, the aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree includes the following steps:

[0047] In step S101, multi-source perception data of the aircraft is acquired and written into the shared data center.

[0048] It is understood that by uniformly writing multi-source sensing data into a shared data center, this application embodiment achieves centralized management and standardized storage of various sensing information such as real-time flight status of the aircraft, detection results of airborne sensors, and data link instructions. This breaks down the barriers of scattered storage of various types of data and ensures that subsequent steps such as macro-state monitoring of finite state machines, loading of behavior tree model indexes, and generation of decision instructions can obtain basic data support that is consistent in time sequence, standardized in format, and accurate and reliable. This lays the data foundation for the efficient and accurate operation of the entire hybrid architecture's autonomous decision-making.

[0049] Specifically, the multi-source perception data acquired in this application embodiment includes real-time flight status data of the aircraft, airborne sensor detection data, and data link instructions. The real-time flight status data covers three-dimensional position, geocentric velocity vector, Euler angle attitude, remaining fuel, and the number of currently mounted weapons. The airborne sensor detection data includes target relative distance, relative azimuth angle, elevation angle, friend or foe identification attributes, and threat warning information. The data link instructions include tactical role identifiers, preset waypoint sequences, engagement authorization flags, and formation coordination instructions. The process of writing the above-mentioned multi-source perception data into the shared data center is not a simple storage process. Instead, a data domain structure of the shared data center is first constructed. This data domain structure corresponds to three independent sets of state vectors, corresponding to the state vector sets of real-time flight status data, airborne sensor detection data, and data link instructions, respectively. Then, the multi-source perception data is mapped into a timestamp-aligned sequence in the target format. The time sequence alignment eliminates the temporal deviation of data from different sources. Based on this alignment sequence, the state vector sets corresponding to each type of data are determined. Finally, the standardized state vector sets are uniformly written into the shared data center to form a complete decision data set, providing a unified data context interface for all subsequent decision-making steps.

[0050] In this embodiment, the multi-source sensing data includes real-time flight status data of the aircraft, airborne sensor detection data, and data link instructions. Writing the multi-source sensing data into a shared data center includes: constructing a data domain structure for the shared data center, the data domain structure including state vector sets for each of the real-time flight status data, airborne sensor detection data, and data link instructions; mapping the multi-source sensing data into a timestamp-aligned sequence in a target format; determining the state vector sets for each of the real-time flight status data, airborne sensor detection data, and data link instructions based on the timestamp-aligned sequence; and writing the state vector sets into the shared data center.

[0051] It is understood that, by constructing a standardized data domain structure and performing format mapping and timestamp alignment on multi-source sensing data, the embodiments of this application can achieve unified storage and efficient management of real-time flight status data, airborne sensor detection data, and data link instructions, ensuring that the timing of each sensing data is consistent and the format is standardized, providing accurate and reliable data support for subsequent macro-state monitoring and behavior tree decision execution, and improving the accuracy and stability of autonomous decision-making.

[0052] Specifically, the data domain structure of the shared data center strictly corresponds to the entire set of decision data. ,in This is a set of local state vectors, corresponding to various parameters of the real-time flight state data, specifically: , This is a set of environmental state vectors, corresponding to airborne sensor detection data, specifically... , This is a set of task instruction vectors, corresponding to data link instructions, specifically... The format mapping and timestamp alignment of multi-source sensing data are achieved through a multi-source data fusion function. , to achieve, among which This represents the discrete time step of the current control cycle. This is a fusion function used to combine airborne sensor data. Data link data and inertial navigation data Timestamp alignment and coordinate system transformation are performed to ensure that the time sequence and format of various types of data are consistent. Based on the alignment sequence, three types of state vector sets are extracted and improved, and then written into the shared data center to achieve standardized centralized management of multi-source heterogeneous data.

[0053] In step S102, a finite state machine is used to monitor the macro-state set in the shared data center, the current macro-state is determined based on the macro-state set, and the corresponding behavior tree model is indexed and loaded from the preset behavior tree library using the macro-state set.

[0054] It is understood that the embodiments of this application use a finite state machine to monitor and determine the macro state in real time, which can accurately determine the current mission state of the aircraft. Based on this state, the matching decision logic model is quickly indexed and loaded from the pre-set behavior tree library, realizing the dynamic binding between the macro state and the micro behavior tree. This ensures clear and stable state scheduling, and enables the decision logic to switch adaptively with the scenario, thereby improving the modularity and scenario adaptability of the system.

[0055] Specifically, the finite state machine first scans the global situational data in the shared data center at a fixed frequency, focusing on monitoring key parameters in the local state variables, environmental state variables, and task instruction variables. Combined with preset task logic and constraints, it then determines whether the trigger event for a state transition is met. The macroscopic state set specifically includes seven discrete states: takeoff, landing, hovering, waypoint flight, game-theoretic confrontation, interception, and refueling. The corresponding state sets are as follows: Finite state machines use state transition functions According to the current state With triggering events Determine the state at the next moment. This allows for the accurate determination of the current macroscopic state. A one-to-one mapping relationship exists between the pre-set behavior tree library and the set of macroscopic states. After the finite state machine determines the current macroscopic state, it uniquely indexes and loads the corresponding behavior tree model from the behavior tree library through this mapping relationship. It also initializes the behavior tree model based on the current state, ensuring that the loaded behavior tree model fully matches the current macroscopic task mode, thus providing a corresponding logical benchmark for subsequent microscopic decision-making reasoning.

[0056] In this embodiment, the finite state machine includes a topology, a state transition function, flow logic, and a macroscopic state set. The macroscopic state set includes takeoff, landing, circling, waypoint flight, game-theoretic, interception, and refueling states. The finite state machine's topology includes a finite state set, a set of events triggering state transitions, and a state transition function. The state transition function is... The flow logic is as follows ,in, For a finite set of states, For a set of events, For the state at the next moment, The current state. The event conditions that trigger state transitions are defined as follows: A finite state machine includes a topology, a state transition function, transition logic, and a macroscopic set of states. The macroscopic set of states includes takeoff, landing, circling, waypoint flight, game-theoretic, interception, and refueling states. The finite state machine's topology includes a finite set of states, a set of events that trigger state transitions, and a state transition function. The state transition function is... The flow logic is as follows ,in, For a finite set of states, For a set of events, For the state at the next moment, The current state. The event conditions that trigger the state transition.

[0057] It is understood that the embodiments of this application, by defining the topology, state transition function and strict flow logic of the finite state machine in a standardized manner, clarify multiple macroscopic flight states and state triggering conditions, can achieve accurate identification, stable flow and reliable switching of the aircraft's macroscopic mission states, provide clear and definite scheduling basis for the index loading of the behavior tree model, and improve the logic, standardization and robustness of the entire hybrid decision architecture.

[0058] Specifically, the topology of a finite state machine is defined as a triplet. ,in It is a finite set of states, specifically including takeoff. ,landing , hover Waypoint flight Game theory and confrontation Interception ,come on Seven mutually exclusive discrete states, each corresponding to a different macroscopic mission phase of the aircraft, are connected by a directed topological network, such as... Figure 2As shown in the figure, the arrows indicate the direction of transition between states; This is a set of events that trigger state transitions. Each event consists of logical operations on variables in the shared blackboard. A typical event includes the refueling trigger event. Game-triggered events wait; This is the state transition function, and its core logic is as follows: That is, the state at the next moment. From the current state and the event conditions that are met The system collectively decides to periodically scan the event set. After capturing a valid trigger signal, the atomic transition of the execution state is performed to ensure the determinism and stability of the macro state transition. At the same time, it provides a clear state basis for the index loading of the behavior tree model, ensuring the coordination and consistency between macro scheduling and micro decision-making.

[0059] In this embodiment, the behavior tree library includes a behavior tree model constructed for each macro state. Dynamic switching and context activation of the behavior tree model are achieved through the current macro state. The behavior tree model is the decision logic structure of the current macro state, which includes a mapping function structure, a graph theory structure, sequence control nodes, and selection control nodes. The mapping function structure includes the mapping relationship between the behavior tree library and the macro state set. For any behavior tree model activated at any time, it is uniquely indexed by the current macro state. Dynamic switching of the behavior tree model is achieved based on the mapping relationship and the unique index. The directed acyclic graph includes a root node, a set of control flow nodes, and a set of leaf nodes. The sequence control nodes are used to control the takeoff and refueling states. When all child nodes return a success flag in sequence, the parent node returns a success flag. The selection control nodes are used for maneuver decisions in the game state. They traverse child nodes according to priority. When a child node returns a success flag or a running flag, the parent node stops traversing subsequent child nodes.

[0060] It is understood that the embodiments of this application achieve accurate matching and rapid activation of decision logic in different task scenarios through one-to-one mapping and dynamic switching between macro states and behavior tree models; at the same time, the use of sequence, selection and other control nodes to distinguish and adapt the execution logic of different macro states makes the decision structure modular, hierarchical and efficient, ensuring both the determinism and real-time nature of state switching, and improving the reliability and maintainability of decision logic in complex maneuver and task scenarios.

[0061] Specifically, behavior treebank With macroscopic state set There is a one-to-one mapping relationship. For any control moment Activated behavior tree model The state output by the current state machine Unique index, i.e. The system uses this mapping relationship and state pointers to achieve rapid dynamic switching and context initialization of behavior tree models corresponding to different macro-states, ensuring that the execution logic of the behavior tree model matches the current macro-task after switching; each behavior tree model Both are directed acyclic graphs. Its graph theory structure includes a root node. Control flow node set and the set of leaf nodes that perform actions or conditional judgments. The system performs a depth-first traversal on T, and the traversal result is... The results of logical operations at each level are recursively synthesized: Control flow nodes control the execution logic of child nodes. Parent nodes refer to sequence, selection, and other control flow nodes, while child nodes can include leaf nodes (specific behavior nodes) or nested control flow nodes. Based on the execution status of its child nodes (complete, successful, failed, etc.), a control flow node determines whether and in what order subsequent nodes will execute. The set of control flow nodes primarily includes sequence control nodes. and select control node In the attached diagram, the sequence control node (corresponding to the symbol →) represents the sequence logic. The parent node only returns "Success" when all child nodes have returned "Success" in sequence. This is mainly used in the takeoff state. Refueling status The logic control for landing status is defined as follows: This ensures strict adherence to timing constraints; the selection control node (corresponding symbol ?) represents the selection logic, traversing child nodes according to a preset priority. Once a child node returns "Success" or "Running," the parent node immediately stops traversing subsequent child nodes. This is primarily used for flexible decision-making in game-playing situations, and its logic is defined as follows: This enables mutually exclusive selection of maneuver strategies; the return status of all nodes belongs to... The Running state is used to maintain continuous actions across cycles, preventing the behavior tree from frequently resetting before the action is completed. (Behavior Tree Model Library) It contains pre-defined decision logic files for seven macroscopic states, namely... , This file represents the decision logic for the pre-set takeoff state. This file represents the decision logic for the preset landing state. This is a file representing the pre-defined decision logic for the hovering state. A file representing the preset decision logic for waypoint flight states. A file representing the pre-defined decision logic for a game-theoretic adversarial situation. This represents the decision logic file that pre-sets the interception state. This file represents the pre-defined decision logic for the refueling status.

[0062] In step S103, the behavior tree model is traversed starting from the root node with the current control cycle as the time step. Based on the behavior tree model, the leaf node action instructions corresponding to the autonomous decision-making actions are output, and the aircraft is controlled to execute autonomous decision-making actions based on the leaf node action instructions.

[0063] It is understood that the embodiments of this application use the control cycle as the time step, traverse the execution behavior tree model from the root node, and can stably and orderly output the leaf node action instructions that match the current scene, and directly drive the aircraft to perform autonomous decision-making actions according to the instructions, ensuring the real-time, continuity and reliability of decision execution, so that the hybrid decision architecture forms a complete closed loop from state scheduling to action implementation.

[0064] Specifically, the system uses a fixed control cycle as the time step to process the behavior tree model activated in step S102. A depth-first traversal is performed, starting from the root node of the behavior tree and sequentially traversing each level of child nodes according to the logical rules of the control flow nodes until the set of leaf nodes is reached. During the traversal, based on the behavior tree topology corresponding to the current macro state, mixed sequence / selection logic or pure sequence logic is used to execute corresponding logical reasoning. Combined with real-time situational data in the shared data center, the return status of each node is determined. Finally, the leaf nodes output leaf node action commands corresponding to autonomous decision-making actions that match the current scenario and task requirements. These commands cover maneuvering actions, sensor control actions, and task payload actions. Subsequently, the output leaf node action commands are parsed. The system converts abstract action commands into executable physical control signals at the aircraft's underlying level. These signals include aerodynamic control surface deflection signals, engine thrust control signals, and load actuation signals. Based on these control signals, the system drives the aircraft's aerodynamic control surfaces, engines, and mission payloads to coordinate their actions and complete the execution of autonomous decision-making actions. Simultaneously, the system feeds back the aircraft's physical responses, such as changes in position, speed, and resource consumption, to a shared data center to update the corresponding state vector set and proceed to the next control cycle. This forms a complete closed loop of perception, macroscopic decision-making, microscopic reasoning, execution, and feedback, ensuring the real-time performance, continuity, and reliability of decision-making execution.

[0065] In this embodiment, the leaf node action instructions corresponding to the autonomous decision-making actions are output based on the behavior tree model, including: if the aircraft is in a game-like adversarial state, the leaf node action instructions include executing parallel control logic, which includes: simultaneously running parallel sub-nodes including radar control, infrared control, electronic warfare control, target update and attack decision, and using the selection node to execute chaff release, tailing, return to base or tracking actions according to the battlefield situation; if the aircraft is in a takeoff state, landing state or refueling state, the leaf node action instructions include executing sequence control logic, which includes executing a programmed action sequence including preparation actions, taxiing actions, landing gear retraction and extension actions, docking actions and flight actions according to the timing constraints.

[0066] It is understood that the embodiments of this application can adaptively adopt parallel control or sequential control logic for different macroscopic states, realize multi-sensor and multi-task parallel decision-making and rapid selection of maneuver actions in game-like adversarial states, and strictly execute programmed actions according to the time sequence in states such as takeoff, landing, and refueling, which not only improves the real-time performance and flexibility of decision-making in complex situations, but also ensures the standardization and reliability of key mission processes.

[0067] Specifically, when outputting leaf node action commands based on the behavior tree model, the corresponding control logic will be adaptively adopted according to the current macroscopic state of the aircraft: when the aircraft is in a game-like adversarial state ( When executing hybrid parallel control logic, the root node of the behavior tree is a sequence logic operator, and its child node queue is arranged according to priority as follows:

[0068] .

[0069] like Figure 3 As shown in the diagram, "→" represents sequence logic, and "?" represents selection logic; where The system selects a node for maneuver decision-making, with the rest being parallel sub-nodes. After the maneuver decision-making node returns to a non-failure state, the system simultaneously runs the parallel sub-nodes corresponding to radar control, infrared control, electronic warfare control, target update, and attack decision-making. This enables parallel decision-making across multiple sensors and tasks, while utilizing a priority queue of the selection control node. Traversing maneuvering sub-nodes, prioritizing chaff release, tail placement, return to base, and tracking based on battlefield situation such as threat alerts, fuel levels, and position, stops subsequent traversal once a maneuvering sub-node returns successfully or displays a running status, and outputs the corresponding chaff release, tail placement, return to base, or tracking action command. When the aircraft is in takeoff, landing, or refueling mode, pure sequence control logic is executed, with the root node of the behavior tree being the sequence control node and the child nodes being programmatic action sequences. The execution logic satisfies strict timing constraints. The Boolean chain multiplication constraint, the specific process is as follows:

[0070] Takeoff procedure: Perform the following steps in sequence: preparation, taxiing, test speed brakes, release speed brakes, throttle at rest, throttle fully extended, afterburner activation, takeoff commencement, liftoff, retract landing gear, and release direct throttle control. Figure 5 As shown in the diagram, "→" indicates sequential logic, and each child node is executed in order;

[0071] Landing procedure: The sequence is as follows: return, approach, flight along the flight path, descent, and landing. Figure 6 As shown in the diagram, "→" represents sequential logic, ensuring timing constraints;

[0072] Refueling procedure: Perform the following steps in sequence: approach, maintain formation, refuel, return to base. Figure 4 As shown in the figure, "→" represents the sequence logic, which ensures the stability of the formation before docking;

[0073] Execution is strictly carried out in accordance with the timing constraints. Subsequent action sub-nodes are only executed when the preceding action sub-node returns a success flag, and the corresponding timing-based action instructions are output to ensure the standardization and security of the process during high-risk task phases.

[0074] In this embodiment of the application, controlling the aircraft to perform autonomous decision-making actions based on leaf node action commands includes: parsing the leaf node action commands to determine the aircraft's aerodynamic control surface deflection signal, engine thrust control signal, and load actuation signal, and controlling the aircraft to perform autonomous decision-making actions based on the aerodynamic control surface deflection signal, engine thrust control signal, and load actuation signal.

[0075] It is understood that the embodiments of this application, by parsing the leaf node action commands, clarify the specific control signals of the aircraft aerodynamic control surfaces, engine thrust and load, realize the precise conversion of autonomous decision-making actions from command output to actual execution, ensure that decision commands can be implemented efficiently and accurately, guarantee the coordinated response of various aircraft components, further improve the stability and accuracy of autonomous decision-making action execution, and adapt to flight control requirements under different macroscopic conditions.

[0076] Specifically, the core of controlling an aircraft to perform autonomous decision-making actions based on leaf node action commands is to complete command parsing and execution. First, the leaf node action commands output from the behavior tree are parsed hierarchically to distinguish different types of commands, such as maneuvering commands, sensor control commands, and mission payload commands. For maneuvering commands, such as tailing, tracking, and taxiing, the required target attitude and target overload parameters are determined after parsing, and then the aerodynamic control surface deflection angles of the elevators, ailerons, and rudder, as well as engine thrust control signals, such as throttle opening, are calculated. For sensor control commands, such as radar activation and infrared detection, and mission payload commands… Commands, such as landing gear retraction and extension, weapon launch, and chaff release, are analyzed to determine the corresponding load actuation signals. Subsequently, the analyzed aerodynamic control surface deflection signals, engine thrust control signals, and load actuation signals are synchronously transmitted to the corresponding actuators on the aircraft, driving aerodynamic control surface deflection, engine thrust adjustment, and mission load to perform corresponding actions. This achieves coordinated response among components and ensures accurate execution of autonomous decision-making actions. At the same time, based on the control requirements of different macroscopic states, the parameter range and response priority of each control signal are dynamically adjusted to adapt to flight control requirements in different scenarios such as takeoff, game, and refueling, further improving the stability and accuracy of decision execution.

[0077] In this embodiment of the application, parsing the leaf node action command includes: obtaining a multi-physics domain command mapping model; inputting the leaf node action command into the multi-physics domain command mapping model, and mapping it into flight control command and mission payload command through the multi-physics domain command mapping model; calculating the aerodynamic control surface deflection signal and engine thrust control signal according to the flight control command, and calculating the payload actuation signal according to the mission payload command.

[0078] It is understood that the embodiments of this application use a multi-physics domain command mapping model to accurately map leaf node action commands into two types of commands: flight control and mission payload. Then, the corresponding control signals are calculated separately, realizing the standardized and accurate conversion of decision commands to equipment execution signals, avoiding command parsing deviations, ensuring the coordinated and accurate response of aerodynamic control surfaces, engines and mission payloads, and further improving the reliability and accuracy of autonomous decision-making action execution.

[0079] Specifically, the process of parsing leaf node action instructions follows a standardized mapping and solution logic, first obtaining a pre-set multi-physics domain instruction mapping model. This model is used to achieve a precise mapping from abstract leaf node action instructions to underlying control instructions; subsequently, the set of leaf node action instructions output by the behavior tree is... The multi-physics domain command mapping model is input, and through model computation, it is mapped into two types of standardized commands: continuous flight control commands. and discrete task payload commands Their mapping relationship satisfies For flight control commands Based on a preset feedback control law Target overload included in the instruction or target posture Error feedback calculations are performed to accurately determine the deflection angles of the elevator, ailerons, and rudder, as well as the engine throttle opening, generating aerodynamic control surface deflection signals and engine thrust control signals; for mission payload commands... Specifically, it is The instructions are mapped to Boolean or enumerated signals using a lookup table method. For example, when the attacking node returns to the Running state, a firing signal is set. (High-level pulse); When the landing gear control node (retraction / extension) is activated in the sequence logic, the hydraulic actuation signal is set. To correspond to high / low levels, when the radar control node is activated, the mode word configuration is based on the output of the decision subtree. (e.g., search / track pattern codes); finally, the standardized load actuation signal is calculated to ensure that the command parsing is error-free and to guarantee the coordinated and accurate response of each actuator.

[0080] The aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree proposed in this application utilizes the division of labor and cooperation between the hybrid architecture of finite state machine and behavior tree. The finite state machine realizes macroscopic state monitoring and scheduling, avoiding the problems of state explosion and complex logic in single state machines in related technologies. At the same time, it relies on dynamic matching of macroscopic states and loading of behavior tree models to make up for the shortcomings of pure behavior trees in macroscopic task flow and state maintenance. Combined with a shared data center for unified management of perception data, the decision logic is clearly layered and highly modular, effectively improving the real-time performance, stability and scalability of aircraft autonomous decision-making, and ensuring efficient and reliable decision execution in complex flight scenarios.

[0081] The following section will elaborate on the aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree proposed in this application through a specific embodiment, such as... Figure 7 As shown, the specific steps are as follows:

[0082] In step one: Multi-source information perception and battlefield situation update are performed to acquire real-time flight status data of the aircraft, airborne sensor detection data and data link instructions. The above multi-source heterogeneous data are fused and processed and written into a shared data center to form a full set of decision data, providing a unified data context interface for subsequent decision steps.

[0083] In step two: macro-state monitoring and flow determination are performed. The finite state machine is used to monitor the global situation data in the shared data center in real time. Based on the preset task logic and constraints, it is determined whether the trigger event for state transition is met, and the current macro-task state of the aircraft is accurately determined, providing a clear state basis for the index loading of the behavior tree model.

[0084] In step three: Dynamic mapping and activation of the behavior tree model are performed. Based on the current macro state determined in step two, the corresponding behavior tree model is uniquely indexed and loaded from the preset behavior tree library. The context activation and initialization of the behavior tree model are realized through the state pointer to ensure that the loaded behavior tree model is completely matched with the current macro task modality.

[0085] In step four: execute micro-tactical logic hierarchical / parallel reasoning, take the current control cycle as the time step, traverse and execute the activated behavior tree model starting from the root node, and adaptively adopt hybrid parallel control logic or pure sequence control logic according to the current macro state, and output the leaf node action instructions corresponding to the autonomous decision action that matches the current scenario and task requirements.

[0086] In step five: execution command settlement and execution, parsing leaf node action commands, mapping them to flight control commands and mission payload commands, calculating aerodynamic control surface deflection signals, engine thrust control signals and payload actuation signals, driving the aircraft to perform corresponding maneuvers or mission operations, and feeding back the physical response after execution to the shared data center, completing the closed-loop iteration of a single perception, decision-making and execution control cycle.

[0087] In summary, the embodiments of this application have at least the following beneficial effects:

[0088] (1) Resolve hierarchical conflicts and reduce decision complexity: By using a hybrid architecture of finite state machine to manage lifecycle and behavior tree to execute specific behaviors, macro task flow and micro single machine decision are handled in layers. This avoids the problem of introducing a large number of complex blackboard variables when behavior tree maintains long-term states, and solves the "state explosion" phenomenon of state machine in high dynamic scenarios in related technologies, which significantly reduces the overall complexity of the decision model.

[0089] (2) Achieve three-dimensional parallel collaboration and improve comprehensive game effectiveness: In the game state, abandon the serial decision-making logic, construct the decision subtrees of maneuver, firepower and perception in parallel, realize the synchronous optimization of flight attitude, weapon launch conditions and detection coverage, so that the UAV can still maintain target tracking and capture the launch window in violent maneuvering, and greatly improve the comprehensive combat effectiveness in complex air combat environment.

[0090] (3) It has strong scalability and facilitates rapid tactical iteration: thanks to the modular design, the behavior tree logic of each state node is independent of each other. When upgrading a specific tactic, only the corresponding sub-module needs to be replaced. There is no need to modify the top-level state machine or other state code. The high cohesion and low coupling characteristics significantly reduce the development and verification costs, and facilitate rapid iteration of the tactical library to adapt to different battlefield needs.

[0091] (4) Balancing decision-making real-time performance with task security: The determinism of finite state machines serves as a safety barrier, ensuring that the system is forced to jump to a safe state under critical conditions, thus avoiding the loss of control caused by the dead loop of behavior tree logic; at the same time, the fast traversal characteristics of behavior trees are used to respond to sudden threats, meet the high real-time operation requirements of embedded platforms, and improve system reliability and robustness.

[0092] Next, referring to the accompanying drawings, an aircraft autonomous decision-making device based on a hybrid architecture of finite state machine and behavior tree proposed according to an embodiment of this application is described.

[0093] Figure 8 This is a block diagram of an aircraft autonomous decision-making device based on a hybrid architecture of finite state machine and behavior tree according to an embodiment of this application.

[0094] like Figure 8 As shown, the aircraft autonomous decision-making device 80 based on a hybrid architecture of finite state machine and behavior tree includes: an acquisition module 801, a loading module 802, and a control module 803.

[0095] The acquisition module 801 is used to acquire multi-source perception data of the aircraft and write the multi-source perception data into the shared data center; the loading module 802 is used to monitor the macro-state set in the shared data center using a finite state machine, determine the current macro-state based on the macro-state set, and index and load the corresponding behavior tree model from the preset behavior tree library using the macro-state set; the control module 803 is used to traverse and execute the behavior tree model starting from the root node with the current control cycle as the time step, output the leaf node action command corresponding to the autonomous decision action based on the behavior tree model, and control the aircraft to execute the autonomous decision action based on the leaf node action command.

[0096] In this embodiment, the multi-source sensing data includes real-time flight status data of the aircraft, airborne sensor detection data, and data link instructions. The acquisition module 801 is further used to: construct a data domain structure of a shared data center, the data domain structure including state vector sets for real-time flight status data, airborne sensor detection data, and data link instructions respectively; map the multi-source sensing data into a timestamp-aligned sequence in a target format; determine the state vector sets for real-time flight status data, airborne sensor detection data, and data link instructions based on the timestamp-aligned sequence; and write the state vector sets into the shared data center.

[0097] In this embodiment, the finite state machine includes a topology, a state transition function, flow logic, and a macroscopic state set. The macroscopic state set includes takeoff, landing, circling, waypoint flight, game-theoretic, interception, and refueling states. The finite state machine's topology includes a finite state set, a set of events triggering state transitions, and a state transition function. The state transition function is... The flow logic is as follows ,in, For a finite set of states, For a set of events, For the state at the next moment, The current state. The event conditions that trigger the state transition.

[0098] In this embodiment, the behavior tree library includes a behavior tree model constructed for each macro state. Dynamic switching and context activation of the behavior tree model are achieved through the current macro state. The behavior tree model is the decision logic structure of the current macro state, which includes a mapping function structure, a graph theory structure, sequence control nodes, and selection control nodes. The mapping function structure includes the mapping relationship between the behavior tree library and the macro state set. For any behavior tree model activated at any time, it is uniquely indexed by the current macro state. Dynamic switching of the behavior tree model is achieved based on the mapping relationship and the unique index. The directed acyclic graph includes a root node, a set of control flow nodes, and a set of leaf nodes. The sequence control nodes are used to control the takeoff and refueling states. When all child nodes return a success flag in sequence, the parent node returns a success flag. The selection control nodes are used for maneuver decisions in the game state. They traverse child nodes according to priority. When a child node returns a success flag or a running flag, the parent node stops traversing subsequent child nodes.

[0099] In this embodiment, the control module 803 is further configured to: if the aircraft is in a game-like confrontation state, determine that the leaf node action command includes executing parallel control logic, which includes: simultaneously running parallel sub-nodes including radar control, infrared control, electronic warfare control, target update and attack decision, and using the selected node to execute chaff release, tail placement, return to base or tracking actions according to the battlefield situation; if the aircraft is in a takeoff state, landing state or refueling state, determine that the leaf node action command includes executing sequence control logic, which includes executing a programmed action sequence including preparation action, taxiing action, landing gear retraction and extension action, docking action and flight action according to the timing constraints.

[0100] In this embodiment, the control module 803 is further configured to: parse the leaf node action command to determine the aerodynamic control surface deflection signal, engine thrust control signal and load actuation signal of the aircraft, and control the aircraft to perform autonomous decision-making actions based on the aerodynamic control surface deflection signal, engine thrust control signal and load actuation signal.

[0101] In this embodiment, the control module 803 is further configured to: acquire a multi-physics domain command mapping model; input leaf node action commands into the multi-physics domain command mapping model, and map them into flight control commands and mission payload commands through the multi-physics domain command mapping model; calculate aerodynamic control surface deflection signals and engine thrust control signals based on the flight control commands, and calculate payload actuation signals based on the mission payload commands.

[0102] It should be noted that the foregoing explanation of the aircraft autonomous decision-making method embodiment based on the hybrid architecture of finite state machine and behavior tree also applies to the aircraft autonomous decision-making device based on the hybrid architecture of finite state machine and behavior tree in this embodiment, and will not be repeated here.

[0103] The aircraft autonomous decision-making device based on a hybrid architecture of finite state machine and behavior tree proposed in this application achieves a division of labor and cooperation between the two architectures. The finite state machine realizes macroscopic state monitoring and scheduling, avoiding the problems of state explosion and complex logic in single state machines in related technologies. At the same time, it relies on dynamic matching and loading of behavior tree models based on macroscopic states to make up for the shortcomings of pure behavior trees in macroscopic task flow and state maintenance. Combined with a shared data center for unified management of perception data, the decision logic is clearly layered and highly modular, effectively improving the real-time performance, stability and scalability of aircraft autonomous decision-making, and ensuring efficient and reliable decision execution in complex flight scenarios.

[0104] Figure 9 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0105] The memory 901, the processor 902, and the computer program stored on the memory 901 and capable of running on the processor 902.

[0106] When the processor 902 executes the program, it implements the aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree provided in the above embodiments.

[0107] Furthermore, electronic devices also include:

[0108] Communication interface 903 is used for communication between memory 901 and processor 902.

[0109] The memory 901 is used to store computer programs that can run on the processor 902.

[0110] The memory 901 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0111] If the memory 901, processor 902, and communication interface 903 are implemented independently, then the communication interface 903, memory 901, and processor 902 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0112] Optionally, in a specific implementation, if the memory 901, processor 902, and communication interface 903 are integrated on a single chip, then the memory 901, processor 902, and communication interface 903 can communicate with each other through an internal interface.

[0113] The processor 902 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0114] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree.

[0115] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0116] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0117] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0118] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (FPGAs), field-programmable gate arrays (FPGAs), etc.

[0119] Those skilled in the art will understand that all or part of the steps of the methods implementing the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0120] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. An autonomous decision-making method for aircraft based on a hybrid architecture of finite state machine and behavior tree, characterized in that, Includes the following steps: Acquire multi-source perception data of the aircraft and write the multi-source perception data into a shared data center; A finite state machine is used to monitor the set of macro states in the shared data center, the current macro state is determined from the set of macro states, and the corresponding behavior tree model is indexed and loaded from a pre-set behavior tree library using the current macro state; Using the current control cycle as the time step, the behavior tree model is traversed and executed starting from the root node. Based on the behavior tree model, the leaf node action instructions corresponding to the autonomous decision-making action are output. Based on the leaf node action instructions, the aircraft is controlled to execute the autonomous decision-making action. The macroscopic state set includes takeoff state, landing state, circling state, waypoint flight state, game confrontation state, interception state, and refueling state.

2. The aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree as described in claim 1, characterized in that, The multi-source sensing data includes real-time flight status data of the aircraft, airborne sensor detection data, and data link instructions. Writing the multi-source sensing data into the shared data center includes: A data domain structure for a shared data center is constructed, which includes a set of state vectors for real-time flight status data, airborne sensor detection data, and data link instructions. The multi-source sensing data is mapped into a timestamp-aligned sequence in a target format. Based on the timestamp-aligned sequence, the state vector sets of real-time flight status data, airborne sensor detection data, and data link instructions are determined, and the state vector sets are written into the shared data center.

3. The aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree as described in claim 1, characterized in that, The finite state machine includes a topology, a state transition function, flow logic, and a macroscopic set of states. The topology includes a finite set of states, a set of events that trigger state transitions, and a state transition function. The state transition function is... The flow logic is as follows: ,in, For a finite set of states, For a set of events, For the state at the next moment, The current state. The event conditions that trigger the state transition.

4. The aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree as described in claim 1, characterized in that, The behavior tree library includes a behavior tree model constructed for each macro state. The behavior tree model is dynamically switched and context activated through the current macro state. The behavior tree model is the decision logic structure of the current macro state. The decision logic structure includes a mapping function structure, a graph theory structure, a sequence control node, and a selection control node. The mapping function structure includes the mapping relationship between the behavior tree library and the macro state set. For any behavior tree model activated at any time, it is uniquely indexed by the current macro state. The dynamic switching of the behavior tree model is realized based on the mapping relationship and the unique index. The graph theory structure includes a root node, a set of control flow nodes, and a set of leaf nodes. The root node is the entry point of the entire behavior tree. Each traversal of all nodes in the tree starts from the root node. The control flow nodes are used to control the execution logic of the leaf nodes, including the order and selection type. Based on the success flags returned by all its child nodes, the corresponding success flag is returned. Leaf nodes are used to control the aircraft to perform specific maneuvers. For the behavior tree of refueling, takeoff, and landing, the control flow node is a sequence control node. If it returns True, its child nodes are traversed in sequence until a node returns False, at which point the traversal stops. For the behavior tree of the game state, the control flow node is the selection control node. Its child nodes are traversed in order. When a child node returns the True flag, the traversal of subsequent child nodes stops.

5. The aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree as described in claim 1, characterized in that, The leaf node action instructions corresponding to the autonomous decision-making actions output based on the behavior tree model include: If the aircraft is in a game-like confrontation state, then the leaf node action command is determined to include the execution of parallel control logic. The execution of parallel control logic includes: simultaneously running parallel sub-nodes including radar control, infrared control, electronic warfare control, target update and attack decision, and using the selected node to execute chaff release action, tail placement action, return to base action or tracking action according to the battlefield situation. If the aircraft is in takeoff, landing, or refueling state, then the leaf node action command is determined to include execution sequence control logic. The execution sequence control logic includes a programmed action sequence that is executed sequentially according to timing constraints, including preparation action, taxiing action, landing gear retraction and extension action, docking action, and flight action along the flight path.

6. The aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree as described in claim 1, characterized in that, The control of the aircraft to perform the autonomous decision-making action based on the leaf node action command includes: The leaf node action command is analyzed to determine the aerodynamic control surface deflection signal, engine thrust control signal, and load actuation signal of the aircraft. Based on the aerodynamic control surface deflection signal, the engine thrust control signal, and the load actuation signal, the aircraft is controlled to execute the autonomous decision-making action.

7. The aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree as described in claim 6, characterized in that, The parsing of the leaf node action instructions includes: Obtain the multi-physics domain instruction mapping model; The leaf node action command is input into the multi-physics domain command mapping model, and mapped into flight control command and mission payload command through the multi-physics domain command mapping model; The aerodynamic control surface deflection signal and the engine thrust control signal are calculated based on the flight control command, and the load actuation signal is calculated based on the mission load command.

8. An autonomous decision-making device for aircraft based on a hybrid architecture of finite state machine and behavior tree, characterized in that, include: The acquisition module is used to acquire multi-source perception data of the aircraft and write the multi-source perception data into a shared data center. The loading module is used to monitor the set of macro states in the shared data center using a finite state machine, determine the current macro state from the set of macro states, and use the current macro state to index and load the corresponding behavior tree model from a preset behavior tree library; The control module is used to traverse and execute the behavior tree model starting from the root node with the current control cycle as the time step, output the leaf node action instructions corresponding to the autonomous decision-making action based on the behavior tree model, and control the aircraft to execute the autonomous decision-making action based on the leaf node action instructions. The macroscopic state set includes takeoff state, landing state, circling state, waypoint flight state, game confrontation state, interception state, and refueling state.

9. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed, they implement the aircraft autonomous decision-making method based on a hybrid architecture of finite state machine and behavior tree as described in any one of claims 1-7.

Images