A human-computer collaborative game simulation decision-making method based on linear temporal logic

Abstract

The application discloses a kind of human-computer collaborative game simulation decision-making methods based on linear temporal logic, comprising: using linear temporal logic expression to describe the task between man and machine, and the expression is converted into automaton;Formalized comprehensive modeling is carried out to unmanned aerial vehicle, and initial global path planning of unmanned aerial vehicle system is carried out using planning automaton;Unmanned aerial vehicle runs according to the initial global path planned, when being disturbed by enemy, dynamic situation is handled using man-machine enemy game algorithm.This method describes the task using linear temporal logic, can describe more complex timing task, effective interaction between man and machine;Linear temporal logic constructs controller by the way of synthesis automaton, does not need human intervention, and is guaranteed by mathematics to construct and correct, so that the system is more reliable;Zero-sum game of man-machine-enemy, consider multi-role arena game model, and construct game tree to obtain dynamic optimal strategy under alternation simulation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of unmanned aerial vehicle task planning, in particular to a human-machine collaborative game simulation decision-making method based on linear temporal logic. BACKGROUND

[0002] With the progress of science and technology and the service of advanced fighter aircraft and unmanned aerial vehicles, the future of air combat and the winning mechanism are undergoing profound changes. Currently, advanced manned fighter aircraft have high speed, high stealth, high mobility and high intelligence, and are equipped with advanced radar sensors and data analysis terminals, which have transformed from single combat execution mechanisms to combat command platforms. Compared with manned aircraft, unmanned aerial vehicles do not cause casualties in combat and are more cost-effective, making them suitable for cluster deployment and close-range entry into local airspace to collect battlefield information. The use environment of unmanned aerial vehicles is developing from safe airspace to hostile airspace, and the control requirements for operators and unmanned aerial vehicle autonomous systems are increasing.

[0003] In manned-unmanned aerial vehicle cooperative combat, unmanned aerial vehicles and manned aircraft need to cooperate to complete "perception, decision-making, planning and action". During the combat process, the unmanned aerial vehicle implements the perception of the battlefield state, processes and analyzes complex battlefield data, generates preliminary decision options according to the top-level combat instructions, and executes the final combat instructions under the judgment of the pilot to carry out combat operations on the ground. In traditional manned-unmanned systems, these actions are all performed by the pilot, and because the pilot uses natural language and the unmanned aerial vehicle uses machine language during human-machine interaction, the instructions submitted to the unmanned aerial vehicle can only be specific point-to-point simple tasks such as "start from area A and arrive at area B, avoid area C during the process".

[0004] The state and corresponding actions of the enemy as the object of the combat system are difficult to completely grasp by our side and have a certain degree of deception, belonging to a game of incomplete information; on the other hand, in the antagonistic game between the two sides, the enemy's situation also changes in real time, which increases the difficulty of the game and brings challenges to the application of our combat system. SUMMARY

[0005] The purpose of the present application is to provide a linear temporal logic-based unmanned aerial vehicle task planning and simulation game method, and the purpose is to design an effective controller to enable the unmanned aerial vehicle to complete complex tasks while reducing the losses caused by the game with the enemy units.

[0006] A human-machine collaborative game simulation decision-making method based on linear temporal logic, comprising:

[0007] using a linear temporal logic expression to describe the tasks between the human and the machine, and converting the expression into an automaton;

[0008] Formal comprehensive modeling of the UAV, and using a planning automaton to perform initial global path planning of the UAV system;

[0009] The UAV runs according to the planned initial global path, and when disturbed by the enemy, uses a man-machine enemy game algorithm to process the dynamic situation.

[0010] Further, the task between the man and the machine is described using a linear temporal logic expression, and the expression is converted into an automaton, comprising:

[0011] According to the task requirements, the atomic proposition set is customized, and the task is converted into a linear temporal logic expression; wherein each atomic proposition in the atomic proposition set is used to represent an event in the task description; the linear temporal logic expression is composed of nested logical operators and temporal operators, and the semantics of the task description is linearly converted into a linear temporal logic expression, and the combination expression of the logical operators and the temporal operators is used to represent the task;

[0012] The linear temporal logic expression is converted into an automaton.

[0013] Further, the logical operators include: "Not" (¬) representing negation, "And" (∧) representing and, "Or" (∨) representing or, and "Imply" (→) representing implication;

[0014] The temporal operators include: "Next" (○) representing "next", "Until" (U) representing "until", "Eventually" (◇) representing "eventually", and "Always" (□) representing "always".

[0015] Further, the automaton is represented by a five-tuple:

[0016] A=(Q,Q0,Π,δ,F)

[0017] Wherein Q is a state set, Q0 is an initial state set, Π=2 AP is a set of Boolean expressions defined on the atomic proposition set, AP is an atomic proposition set, i.e. an event set in the task, is a transition relation, and F is an acceptable state set.

[0018] Further, the formal comprehensive modeling of the UAV, and using a planning automaton to perform initial global path planning of the UAV system, comprises:

[0019] Comprehensive modeling of the UAV and its working space, and constructing a weighted state transition system model;

[0020] ​The weighted state transition system model and the automaton are cross-multiplied to obtain a planning automaton;

[0021] Path searching is performed on the planning automaton to obtain an initial global path of the UAV system.

[0022] Further, the UAV and the workspace are comprehensively modeled using the weighted state transition system model, which can be generally expressed as a tuple:

[0023] UAV=(S, Act, s0, →, AP, L, W)

[0024] S is a state set of the UAV, including physical position information; Act is an action set of the UAV, including movement and fire attack; s0 represents an initial state of the UAV; is a transition relation between states, which defines which new states can be reached by which actions from a certain state; AP is an atomic proposition set, and L: S→2 AP is a labeling function, which records a series of atomic propositions corresponding to each system state; W is a mapping from → to a natural number set, representing the weight of each state transition.

[0025] Further, the planning automaton is expressed as:

[0026]

[0027] wherein, respectively represent the cross-multiplied state set, the initial state, and the action set.

[0028] Further, the path searching on the planning automaton to obtain an initial global path of the UAV system includes:

[0029] Path searching is performed on the planning automaton to obtain an initial global path of the UAV system. * or Dijkstra algorithm is used to perform path searching, and the final searched path is the state evolution sequence of the planning automaton. Then, the UAV system evolution sequence is separated from the planning automaton, and a set of action sequences is parsed from the UAV system evolution sequence, thereby obtaining an initial global path.

[0030] Further, the UAV runs according to the planned initial global path, and when disturbed by the enemy, a dynamic situation is handled using a man-machine enemy game algorithm, which includes:

[0031] The UAV runs according to the initial global path planning result to complete the given task in a static environment, and needs to be re-planned through a game algorithm when an enemy unit interferes with the task completion process; when the enemy unit enters a certain preset range, a man-machine-enemy game algorithm is enabled to make decisions on actions:

[0032] According to the state set and action set of the enemy UAV and the UAV, the dynamic game problem is modeled, and a game arena six-tuple is constructed;

[0033] Based on the game arena six-tuple, a winning game strategy considering the worst case is obtained by using the Minimax game tree search algorithm; that is, the Minimax game tree search algorithm is used to search for decisions on the game arena six-tuple obtained in the previous step, starting from the current state of the six-tuple, and searching for the possible state evolution path in the future.

[0034] Further, the zero-sum game of the two parties in turn is expressed as a game arena six-tuple:

[0035]

[0036] Where Q = Q red ∪ Q blue is the state set, including the state Q red of the UAV and the state Q blue of the opponent UAV, Act = Act red ∪ Act blue is the action set, including the action Act red that the UAV can take and the action Act blue that the opponent UAV can take; T: Q × Act→ Q is the transition function, and q0 is the initial state set.

[0037] Further, in the process of the game, the red UAV will always choose the scheme with the minimum cost, and the enemy will always choose the scheme that makes the opponent's cost greater, so the UAV is called min and the opponent is called max; Since the two parties in the game move alternately, the nodes and their parent nodes of the game tree belong to one of the two parties, and their categories also belong to max and min respectively; The node on the game tree corresponds to a depth, and the depth of the parent node is zero;

[0038] Firstly, the maximum search depth D is determined, and then a game tree with the maximum depth D is constructed as a parent node starting from the current state; the nodes on the game tree represent the state of the UAV, and the path of the game tree is equivalent to the action sequence taken by the UAV; at the leaf nodes of the game tree with the maximum depth D, the leaf node value is evaluated using a predefined value evaluation function, and the node causing the task to fail is assigned an infinite value; then the non-leaf nodes are assigned from bottom to top, according to the cost of taking action, wherein the max node takes the maximum value of the child nodes, and the min node takes the minimum value of the child nodes; finally, the game tree after the assignment is used for decision-making in actual action, and when it is my turn, the path of the child node with the value equal to the value of the min node is selected.

[0039] Compared with the prior art, the application has the following technical characteristics:

[0040] 1. Compared with the traditional task planning method of constructing differential equations, the method uses linear temporal logic to describe the task, which can describe more complex temporal tasks, and linear temporal logic is also a good formal language for effective interaction between man and machine.

[0041] 2. Linear temporal logic constructs a controller by synthesizing an automaton, which does not require human intervention during the process and is mathematically guaranteed to be correct, making the system more reliable.

[0042] 3. For the zero-sum game between man and machine-enemy, an Arena game model considering multiple roles is constructed, and a game tree is constructed to obtain the dynamic optimal strategy under alternating simulation using the Minimax algorithm. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 Man-machine cooperative task flow chart

[0044] Figure 2 Linear temporal logic semantic diagram

[0045] Figure 3 Task Converted into an automaton, where init is the initial state and the double circle represents an acceptable state

[0046] Figure 4 Initial task planning flow chart of the UAV DETAILED DESCRIPTION

[0047] Referring to the drawings, the application provides a man-machine cooperative game simulation decision-making method based on linear temporal logic, which includes the following steps:

[0048] Step 1, use linear temporal logic expression to describe the task between man and machine, and convert the expression into an automaton

[0049] Step 1.1, customizing the set of atomic propositions according to the task requirements, and converting the task into a linear temporal logic expression; wherein each atomic proposition in the set of atomic propositions is used to represent an event in the task description; the linear temporal logic expression is composed of nested logical operators and temporal operators, and the semantics of the task description is converted into linear temporal logic, and the combination of the logical operators and the temporal operators is expressed as an expression to represent the task.

[0050] In human-computer collaborative tasks, human-computer pilots have rich combat experience and play the role of decision-makers in collaboration, and unmanned aerial vehicles have stronger mobility and serve as task executors in collaboration. Therefore, effective, accurate and efficient task communication between the two is required for human-computer collaboration. Natural language itself has the characteristics of ambiguity and ambiguity, and the language characteristics of each pilot are very different, and the performance of the onboard computer is limited, making it difficult to process natural language tasks, so this scheme considers using linear temporal logic for task interaction between humans and computers.

[0051] Temporal logic belongs to the field of formal method theory, and is a theoretical system that integrates description, verification and synthesis. It can be said that the birth of temporal logic perfectly meets the demand of human-computer collaboration at a higher level and convenience. Similar to the natural language-based human-computer interaction mode, temporal logic also aims to simplify the learning and operation cost in the process of human-computer interaction. Although it is close to the expression mode of natural language, compared with natural language, temporal logic has more obvious advantages, with rich expression ability and no ambiguity.

[0052] The syntax of the linear temporal logic language can be recursively defined by the following expression:

[0053]

[0054] This formula is the syntax definition formula of the linear temporal logic language, which reflects a legal linear temporal logic expression is constructed by a series of symbol nesting. Considering the simplest case, this scheme only has an atomic proposition a, and a series of symbol nesting of a is as follows: a) is a linear temporal logic sentence that meets the syntax. Nesting is from the bottom up in this scheme, but from the top down, a better definition can be obtained, that is, to define in a recursive form, the symbol ::= represents recursion, and the end of recursion is True or a such basic elements. True represents the atomic proposition true, that is, a state that is always satisfied, that is, always true without any conditions, which is actually less used; a is a certain atomic proposition, which is the basic element of complex and ordered tasks; represents a task.

[0055] The physical meaning of atomic proposition is that the user defines the event of interest in the actual system running process. For the task, the atomic proposition represents an event in the task. For example, "our fighter is in the enemy radar detection area" is defined as an atomic proposition a, and a is true at a certain time, which means that the event actually occurs at that moment. Therefore, the subsequent text scheme will describe the arrival of the radar detection area as reaching a, which actually means reaching a place to make event a occur. Of course, the atomic proposition can also be defined as an event such as "damage cost less than 100". How to define the atomic proposition depends entirely on the user's task description requirements and the difficulty of distinguishing between event occurrence and non-occurrence. The task is represented, and the task can be nested. The task contains two symbols, namely logical operators ( "Not" ( ) represents not, "And" ( ∧ ) represents and, "Or" ( ∨ ) represents or, and "Imply" ( → ) represents implication) and time sequence operators ( "Next" ( ○ ) represents "next", "Until" ( U ) represents "until", "Eventually" ( ◇ ) represents "eventually", and "Always" ( □ ) represents "always"). There is a certain conversion relationship between the symbols, such as "Eventually" can be represented by "Until" : , "Always" can be represented by "Eventually" combined with "Not" : , and "Imply" can be realized by "Not" and "Or" : . The semantic diagram is shown in the attached Figure 2 .

[0056] With such a representation tool as linear temporal logic, it is possible to represent tasks in a fairly large range of task space, such as the task description is "eventually reach task point a and need to avoid obstacle d, and need to visit task point b before visiting task point a", the task can be expressed as

[0057] Step 1.2, convert the linear temporal logic expression into an automaton

[0058] Linear temporal logic can be converted into an equivalent büchi automaton. Büchi automaton is a commonly used tool for process modeling. A büchi automaton can usually be represented by a five-tuple:

[0059] A = ( Q, Q0, Π, δ, F )

[0060] where Q is the state set, Q0 is the initial state set, Π = 2 APAP is a set of Boolean expressions defined on a set of atomic propositions, where AP is the set of atomic propositions, i.e., the set of events in a task. It is a transition relationship, and F is the set of acceptable states.

[0061] An automaton is a tool for parsing the semantics of a given linear sequential logic expression. Essentially, it's a directed graph with a set of active nodes (called states). Each node can be considered to encode the progress of a stage in the corresponding linear sequential logic expression (for example, a task requires three stages to complete, and a node indicates that the first stage is complete but the second stage is not). The automaton continuously reads in events occurring at the current moment (all atomic propositions that are true at this moment), and then it undergoes state transitions (active nodes change). At each moment, it checks whether its state satisfies a certain condition to determine whether the corresponding linear sequential logic expression is satisfied, i.e., whether the task is completed. For the process of converting linear sequential logic expressions into automata, there are already mature automaton construction algorithms, and stable software tools (such as ltl2ba) implement these algorithms, enabling fully automatic conversion from linear sequential logic expressions to automata, which is sufficient for the application scenario of this solution. The resulting automaton is attached. Figure 3 .

[0062] Step 2: Perform formal synthesis modeling of the UAV and use a planning automaton to perform initial global path planning for the UAV system.

[0063] Step 2.1: Perform comprehensive modeling of the drone and its workspace to construct a weighted state transition system model.

[0064] An environmental model is the primary prerequisite for UAV mission planning. The UAV's workspace should be three-dimensional; however, to reduce mission planning complexity, this approach considers building the UAV's workspace into a two-dimensional planar model. A weighted state transition system model is used to comprehensively model the UAV and its workspace, typically represented as a tuple:

[0065] UAV=(S, Act, s0, →, AP, L, W)

[0066] Where S is the drone's state set, including information such as physical location; Act is the drone's action set, including movement and fire strikes. s0 represents the drone's initial state, typically the core command area. For the transition relationships between states, define which actions a given state can transition to which new states; AP is the set of atomic propositions, consistent with step 1, serving as a bridge between the system model and the task description; L: S→2 APis a set of atomic propositions (may be empty, single or multiple atomic propositions) corresponding to each system state, meaning the events happened in this state; W is a mapping from Q to the weight of each state transition (physical meaning is the distance between two states).

[0067] Step 2.2, cross-multiply the weighted state transition system model and the automaton obtained in step 1 to obtain the planning automaton

[0068] The required task is described using linear temporal logic, and its task state automaton A = (Q, Q0, Π, δ, F) is obtained. Cross-multiplying the task automaton and the state transition system model of the UAV can obtain the planning automaton:

[0069]

[0070] wherein respectively represent the state set after cross-multiplication, the initial state, and the action set.

[0071] The physical meaning of cross-multiplication is to associate the automaton and the system model into a large system, and to combine the automaton state and the system state into a new system state (for example, (coordinate (3, 5), complete the first phase of the task) is a cross-multiplication state). The planning automaton is a combination of the original automaton and the original system model, and they communicate with each other through the atomic proposition set AP. The system model continuously transmits the currently true atomic propositions to the original automaton to drive the original automaton to transition, so that they can operate in coordination. The advantage of cross-multiplication is to use the automaton that monitors the evolution of the system to analyze the progress of the current task, which facilitates the next step of planning.

[0072] Step 2.3, perform path search on the planning automaton to obtain the initial global path of the UAV system

[0073] After obtaining the planning automaton, the state evolution sequence σ of the UAV system (i.e., the sequence of states generated by the UAV system in chronological order) will only be accepted by the automaton if it satisfies the linear temporal logic specification, i.e., the task specification. The planning automaton is a large system that contains both the system model and the task completion progress model (i.e., the automaton). Therefore, any legal evolution behavior of the planning automaton is the coupling of the two sequences, i.e., the original system state evolution sequence and the automaton state evolution sequence. Therefore, path search can be performed on the directed graph corresponding to the planning automaton from the initial point to the cross-multiplication states (which may be multiple) containing the automaton acceptable states. Specifically, A *and Dijkstra algorithm, the final search path is the state evolution sequence of the planning automaton, and then the evolution sequence of the UAV system is separated from it, and a set of action sequences (e.g., walk east, walk north, fire…) are parsed from it, so as to obtain an initial global path. The planning process is shown in FIG. 8. It is called an initial global path because a static system model is used for planning in this step, and it cannot cope with dynamic situations such as encountering enemy aircraft on the path. The processing of dynamic situations needs to use the game algorithm later. Figure 4

[0074] Step 3, the UAV runs according to the initial global path planning result, and when it is disturbed by the enemy, the man-machine enemy game algorithm is used for dynamic situation processing

[0075] The UAV runs according to the initial global path planning result to ensure that the given task is completed in a static environment. When enemy units interfere with the task completion process, the game algorithm is used for re-planning. In order to reduce the computational complexity, when the enemy units enter a certain preset range, the man-machine-enemy game algorithm is used to make decisions, as follows:

[0076] Step 3.1, according to the state set and action set of the enemy UAV and our UAV, the dynamic game problem is modeled, and a game arena six-tuple is constructed

[0077] First, the game problem is modeled, and the man-machine-enemy game process is constructed into a dynamic zero-sum game model.

[0078] According to whether the action of the first mover in the game affects the second mover, the game problem can be divided into static game and dynamic game. In static game, the order of action of the two parties does not affect their decision-making behavior, which can be described by a game matrix. If the decision of the second party in the game process is based on the action of the first party, the game process is dynamic, which is generally described by a game tree.

[0079] For the man-machine-enemy game process, this scheme can be considered as a dynamic game process, that is, both sides take turns to make actions. And this game process can be regarded as a zero-sum game between the two parties. Our side pursues to complete all tasks as soon as possible, and the enemy side will try to make the task impossible to complete or increase the cost of task completion. The two-party turn-based zero-sum game can be expressed as a game arena six-tuple:

[0080]

[0081] where Q = Q red ∪ Q blue is the state set, which includes the state Q red ​Also includes the state Q of the opponent UAV blue , Act = Act red ∪ Act blue is the action set, also includes both the actions Act red that the UAV can take and the actions Act blue that the opponent UAV can take, T: Q x Act→ Q is the transition function, q0 is the initial state set, and the remaining symbols are defined as before; both sides take turns to take actions. The UAV needs to consider the enemy's decision while going in the direction of a larger winning side. Since the enemy's strategy is uncertain, it is assumed that the opponent will always choose the action that maximizes the interference to the UAV. The Minimax algorithm is used, which is a zero-sum algorithm, that is, one side chooses the selection that maximizes its advantage from the available options, while the other side chooses the method that minimizes the opponent's advantage.

[0082] Step 3.2, based on the game arena six-tuple, the winning game strategy considering the worst case is obtained by using the Minimax game tree search algorithm; that is, the Minimax game tree search algorithm is used to search for decisions on the game arena six-tuple obtained in the previous step, starting from the current state of the six-tuple, searching for possible future state evolution paths (the path is actually determined by the actions of the enemy and myself), the specific steps are as follows:

[0083] (1) First, determine the maximum search depth D;

[0084] (2) At the leaf nodes of the game tree with a maximum depth of D, use a pre-defined value evaluation function (such as the degree of impact or the probability of being detected, depending on actual needs) to evaluate the value of the leaf nodes;

[0085] (3) Assign values to non-leaf nodes from bottom to top. The max node takes the maximum value of the child nodes, and the min node takes the minimum value of the child nodes;

[0086] (4) Each time it is my turn, choose the path of the child node whose value is equal to the value of this min node.

[0087] Finally, through the Minimax game tree search algorithm, the human-computer collaborative game decision is completed.

[0088] Minimax is a pessimistic algorithm, that is, assuming that the opponent will lead me to the node with the minimum value from the current theoretical value, that is, the opponent has perfect decision-making ability. Therefore, the strategy of me should be to choose the best among the worst case that the opponent can reach, that is, the loss caused by the opponent to me under perfect decision-making. Minimax does not find the theoretically optimal solution, because the theoretically optimal solution often depends on whether the opponent is stupid enough, in Minimax, me has full control, if the opponent makes perfect decision every step, me can achieve the expected minimum loss result, if the opponent does not walk out of perfect decision, me can achieve better results than the worst case. In short, me is to choose the best in the worst case.

[0089] In the process of game, the red unmanned aerial vehicle always chooses the scheme with the minimum cost, and the enemy always chooses the scheme that makes the opponent's cost greater, so the me is called min, and the opponent is called max. Because the two parties of the game are alternately moved, the nodes of the game tree and their parent nodes belong to one of the two parties, and their categories belong to max and min respectively. The nodes on the game tree correspond to a depth, and the parent node has a depth of zero. First, determine the maximum search depth D, then start from the current state as the parent node to build a game tree with a maximum depth of D. The nodes on the game tree represent the state of the unmanned aerial vehicle, and the path of the game tree is equivalent to the action sequence taken by the unmanned aerial vehicle. At the leaf nodes of the pattern tree with a maximum depth of D, the pre-defined value evaluation function is used to evaluate the value of the leaf nodes, and the nodes that cause the task to fail are assigned a value of infinity. Then, from bottom to top, the non-leaf nodes are assigned values according to the cost of the action taken. Among them, the max node takes the maximum value of the child nodes, and the min node takes the minimum value of the child nodes. Finally, when actually acting, according to the game tree after the value assignment, when it is my turn (at this time, it must be at a min node of the pattern tree), the path of the child node with the value equal to the value of the min node is selected.

[0090] The above examples are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for human-in-the-loop game simulation decision-making based on linear temporal logic, characterized in that, The application relates to a linear temporal logic-based human-machine cooperative game simulation decision method. A task between a human and a machine is described by using a linear temporal logic expression, and the expression is converted into an automaton; An unmanned aerial vehicle (UAV) is formally and comprehensively modeled, and an initial global path of the UAV system is planned by using a planning automaton; The UAV runs according to the planned initial global path, and when disturbed by an enemy, the UAV uses a human-enemy game algorithm to process a dynamic situation; The UAV is formally and comprehensively modeled, and the initial global path of the UAV system is planned by using the planning automaton, which comprises the following steps. The UAV and a working space where the UAV is located are comprehensively modeled to construct a weighted state transition system model; The weighted state transition system model and the automaton are cross-multiplied to obtain a planning automaton; An initial global path of the UAV system is searched on the planning automaton; The UAV and the working space are comprehensively modeled by using the weighted state transition system model, and are expressed as a tuple: ； wherein is a set of states of the UAV, including physical position information; is a set of actions of the UAV, including movement and fire attack; denotes the initial state of the UAV; is the transition relation between states, defining which new states a certain state can be transferred to through which actions; is a set of atomic propositions, is a labeling function, recording a series of atomic propositions corresponding to each system state; is is a mapping to the set of natural numbers, representing the weight of each state transition; The planning automaton is expressed as: ； wherein, respectively represent the state set after cross multiplication, the initial state, and the action set; is a set of Boolean expressions defined on the set of atomic propositions, is a transition relation, is a set of acceptable states; The UAV runs according to the planned initial global path, and when disturbed by an enemy, the UAV uses a human-enemy game algorithm to process a dynamic situation, which comprises the following steps. The UAV runs according to the initial global path planning result to complete a given task in a static environment, and when an enemy unit interferes with the task completion process, the UAV needs to be re-planned by using a game algorithm; when the enemy unit enters a certain preset range, a human-enemy game algorithm is used to make a decision on the action: A dynamic game problem is modeled according to state sets and action sets of enemy UAVs and our UAVs to construct a game arena six-tuple; Based on the game arena six-tuple, a Minimax game tree search algorithm is used to obtain a winning game strategy considering the worst case; that is, the Minimax game tree search algorithm is used to make a decision search on the game arena six-tuple obtained in the previous step, and the future possible state evolution path is searched from the current state of the six-tuple. 2.The human-machine collaborative game simulation decision-making method based on linear temporal logic according to claim 1, wherein, The task between the human and the machine is described by using the linear temporal logic expression, and the expression is converted into the automaton, which comprises the following steps. An atomic proposition set is defined according to a task requirement, and the task is converted into a linear temporal logic expression; each atomic proposition in the atomic proposition set is used to represent an event in the task description; the linear temporal logic expression is composed of nested logical operators and temporal operators, and the expression is used to linearly and temporally convert the semantics of the task description; the task is expressed by the combination of the logical operators and the temporal operators. The linear temporal logic expression is converted into the automaton. 3.The human-machine collaborative game simulation decision-making method based on linear temporal logic of claim 1, wherein, The automaton is expressed as a five-tuple: ； wherein is a set of states, is a set of initial states, is a set of Boolean expressions defined on a set of atomic propositions, is a set of atomic propositions, i.e. a set of events in the task, is a transition relation.

4. The linear temporal logic-based human-machine cooperative game simulation decision method according to claim 1, wherein The initial global path of the UAV system is searched on the planning automaton, which comprises the following steps. The path search from the initial point to the fork state containing the receivable state of the automaton is performed on the directed graph corresponding to the planning automaton, and the path search is performed by using A * or Dijkstra algorithm, and the final searched path is the state evolution sequence of the planning automaton, and then the evolution sequence of the unmanned aerial vehicle system is separated from the state evolution sequence, a set of action sequences is analyzed, and thus an initial global path is obtained.

5. The human-computer collaborative game simulation decision-making method based on linear temporal logic according to claim 1, characterized in that, A zero-sum game between two parties is expressed as a game arena six-tuple: ； wherein S is a set of states, including both states of our drone Q red and states of the opponent drone Q blue , A is a set of actions, including both actions our drone can take Act red and actions the opponent drone can take Act blue ; T is a transition function, q 0 is an initial state set.

6. The human-computer collaborative game simulation decision-making method based on linear temporal logic according to claim 1, characterized in that, In the process of game, the red unmanned aircraft always chooses the minimum cost solution, while the enemy always chooses the solution that makes the other side more costly, so the former is called min and the latter is called max; since the game is alternately moved, the nodes and their parent nodes of the game tree belong to one of the two parties, and the types of the nodes and their parent nodes of the game tree also belong to max and min respectively; the nodes on the game tree correspond to a depth, and the depth of the parent node is zero; Firstly, the maximum search depth D is determined, and then the game tree with the maximum depth D is constructed from the current state as the parent node; the nodes on the game tree represent the state of the unmanned aircraft, and the path of the game tree is equivalent to the action sequence taken by the unmanned aircraft; at the leaf nodes of the game tree with the maximum depth D, the value of the leaf nodes is evaluated using a predefined value evaluation function, and the nodes leading to task failure are assigned a value of infinity; then the non-leaf nodes are assigned values from bottom to top, according to the cost of the action taken, wherein the max node takes the maximum value of the child nodes, and the min node takes the minimum value of the child nodes; finally, the actual action is decided according to the game tree after the assignment, and every time it is the turn of the former, the path of the child node with the value equal to the value of the min node is selected.

Images