A deception-based pursuit method and system for autonomous pursuit missions

Abstract

This invention provides a deception pursuit method and system for autonomous pursuit missions. It defines a high-level intent set and selects decoy and true intentions from it. It establishes behavioral patterns for the escape vessel to update its belief distribution based on the observable motion of the pursuing vessel and select a target exit based on time cost and interception risk cost. The pursuing vessel estimates its current target exit as the induced target exit based on the matching relationship between the escape vessel's course and the exit bearing. It constructs a set of candidate course angles around the pure pursuit direction and the induced blocking direction. It constructs a single-step drift in the logarithmic domain using the posterior ratio of the decoy intention and the true intention and sets drift constraints to quantify the degree to which the current course promotes the escape vessel's belief growth regarding the decoy intention. It calculates the computable lower bound of the drift for each candidate course, filters courses that satisfy the drift constraints to form a feasible set, selects the course with the smallest deviation from the pure pursuit direction as the output course, and if no feasible course is found, it reverts to the pure pursuit direction and generates control commands.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of intelligent navigation and autonomous decision-making control technology for ships, and in particular to a deception pursuit method and system for autonomous pursuit missions. Background Technology

[0002] In restricted navigation waters such as port security, island and reef patrols, and narrow channel control, the autonomous pursuit of escaping vessels by pursuing ships is a crucial research issue in the field of intelligent ship navigation and autonomous decision-making control. The pursuit-escape game is widely present in these application scenarios requiring the interception and control of targets within waterways. Its core lies in the pursuer controlling its own movement through autonomous decision-making to capture the controlled object (escaping vessel) before it reaches a safe area (exit). With the improvement of escaping vessels' autonomous decision-making capabilities and the prevalence of multi-exit environments, escaping vessels can autonomously select and dynamically switch targets among multiple escape exits based on their own state, environmental information, and observed pursuing vessel behavior. This expands the pursuit-escape problem from a purely geometric pursuit under a single exit to an intent inference and strategy game problem under multi-exit conditions. Existing pursuit strategies mainly fall into two categories: one is a pure pursuit strategy, where the pursuing vessel always points towards the escaping vessel's current position to shorten the distance between them; the other is an exit blocking strategy, where the pursuing vessel prioritizes pointing towards its estimated target exit for the escaping vessel to achieve early interception. The above strategies are often used in combination or as basic decision-making logic in multi-exit fugitive pursuit scenarios.

[0003] However, the aforementioned existing pursuit strategies all treat the target (escape vessel) as a simple objective that passively reacts to the environment, failing to consider the escape vessel's active inference capability to deduce the pursuit intention based on the pursuer's observable motion and adjust its exit decision accordingly. While a pure pursuit strategy can maintain a continuous geometrical approach between the pursuer and the escape vessel, when the escape vessel approaches a certain exit, relying solely on geometrical approach often fails to capture it before it reaches the target exit, especially when the speed difference between the two is limited. The pursuer struggles to achieve effective interception using a pure pursuit strategy. Although exit blocking strategies can provide early defense at specific exits, premature or overly forceful blocking can cause the pursuer to deviate excessively from the escape vessel's current position. Once the escape vessel detects the pursuer's blocking intention and switches to another exit, the pursuer will lose its continuous interception capability due to its distance from the escape vessel, potentially leading to a failed pursuit. More importantly, existing methods fail to utilize the observable motion of the pursuing vessel as an active source of information. The pursuing vessel cannot actively influence the escaping vessel's judgment of its intentions by deliberately adjusting its own motion, thus failing to induce the escaping vessel to change its target exit choice. This leaves the pursuing vessel in a passive, reactive position in multi-exit pursuit games; it can only observe and respond to the escaping vessel's exit choice, unable to exert proactive influence on the escaping vessel's exit decision-making process. This fundamentally limits the pursuing vessel's interception effectiveness in multi-exit dynamic games. Furthermore, there is currently a lack of simulation and real-ship closed-loop testing platforms capable of effectively verifying the aforementioned proactively induced pursuit strategies, limiting the efficiency of translating related algorithms from theory to practical application.

[0004] In summary, existing pursuit strategies have two fundamental limitations in multi-exit pursuit scenarios: First, the pursuing vessel cannot actively influence the escape vessel's target exit selection through its own movement, leaving the pursuer in a passive, reactive position throughout the game. Second, existing methods struggle to achieve an effective balance between cognitive guidance and geometric approximation. The pursuing vessel either fails to prevent the escape vessel from approaching the exit due to pure pursuit, or deviates excessively from the escape vessel while attempting to block the exit, losing its interception capability after the escape vessel switches targets. These shortcomings are not caused by human operation or management factors, but rather by the technical architecture of existing strategies themselves. The root cause lies in the fact that existing methods simplify the pursuit problem into a one-sided geometric chase or area defense, failing to establish a coupling mechanism between the observable movement of the pursuing vessel and the inference of the escape vessel's intentions, and failing to construct a quantitative decision-making basis between induced maneuvers and physical approximation. Summary of the Invention

[0005] This invention provides a deception pursuit method and system for autonomous pursuit missions, addressing the technical problems in existing pursuit strategies where the pursuing vessel cannot actively influence the escape vessel's target exit selection through its own movement, making it difficult to achieve an effective balance between cognitive guidance and geometric approximation, resulting in insufficient interception effectiveness, and lacking simulation and water-based closed-loop verification for deception pursuit strategies.

[0006] In a first aspect, embodiments of the present invention provide a deception pursuit method for autonomous pursuit missions, comprising: S1. Define a set of high-level intentions for the pursuing ship, which includes a first intention to directly pursue the escaping ship and a second intention to block each exit; select one intention from the set of high-level intentions as the decoy intention, and select another intention different from the decoy intention as the real intention; S2. Define the behavior pattern of the escape ship: The escape ship updates its belief distribution on various intentions in the high-level intention set according to the observable motion state of the pursuing ship, and selects the target exit according to the time cost of each exit and the interception risk cost determined based on the belief distribution. S3. The pursuing vessel determines the current target exit of the escape vessel based on the matching relationship between the heading status of the escape vessel and the bearings of each exit; the current target exit is used as the induced target exit, the pure pursuit direction of the pursuing vessel pointing towards the escape vessel and the induced blocking direction of the pursuing vessel pointing towards the induced target exit are determined, and discrete sampling is performed between the pure pursuit direction and the induced blocking direction to construct a candidate heading angle set; S4. Based on the posterior ratio between the decoy intention and the true intention, obtain the single-step expected increment of the posterior ratio in the logarithmic domain, construct a single-step drift amount based on the single-step expected increment, and set drift constraints for filtering the candidate heading angle set based on the single-step drift amount. S5. Calculate the calculable lower bound of the drift for each candidate heading angle in the candidate heading angle set, retain the candidate heading angles that satisfy the drift constraint to form a feasible heading angle set, select the candidate heading angle with the smallest deviation from the pure pursuit direction as the output heading of the pursuing ship when the feasible heading angle set is not empty, use the pure pursuit direction as the output heading of the pursuing ship, and generate the pursuing ship control command according to the output heading of the pursuing ship.

[0007] Secondly, embodiments of the present invention provide a deception pursuit system for autonomous pursuit missions, comprising: The intent definition module is used to define a set of high-level intents for the pursuing vessel. The set of high-level intents includes a first intent to directly pursue the escaping vessel and a second intent to block each exit. An intent is selected from the set of high-level intents as a decoy intent, and another intent different from the decoy intent is selected as the real intent. The escape ship behavior estimation module is used to set the behavior rules of the escape ship: the escape ship updates the belief distribution of the escape ship to various intentions in the high-level intention set according to the observable motion state of the pursuing ship, and selects the target exit according to the time cost of each exit and the interception risk cost determined based on the belief distribution; The target estimation and heading construction module is used by the pursuing ship to determine the current target exit of the escape ship based on the matching relationship between the heading status of the escape ship and the bearings of each exit; using the current target exit as the induced target exit, determining the pure pursuit direction of the pursuing ship pointing towards the escape ship, and the induced blocking direction of the pursuing ship pointing towards the induced target exit, and performing discrete sampling between the pure pursuit direction and the induced blocking direction to construct a set of candidate heading angles; The drift constraint construction module is used to obtain the single-step expected increment of the posterior ratio in the logarithmic domain based on the posterior ratio between the decoy intention and the true intention, construct a single-step drift amount based on the single-step expected increment, and set drift constraints for filtering the candidate heading angle set based on the single-step drift amount. The heading selection and output module is used to calculate the calculable lower bound of the drift of each candidate heading angle in the candidate heading angle set, retain the candidate heading angles that satisfy the drift constraint to form a feasible heading angle set, select the candidate heading angle with the smallest deviation from the pure pursuit direction as the output heading of the pursuing ship when the feasible heading angle set is not empty, use the pure pursuit direction as the output heading of the pursuing ship when the feasible heading angle set is empty, and generate the pursuing ship control command according to the output heading of the pursuing ship.

[0008] This invention provides a deception pursuit method and system for autonomous pursuit missions. The pursuing vessel, through deliberate design of its observable motion behavior, actively manipulates the escape vessel's belief in the pursuit intent, thereby inducing the escape vessel to change its target exit selection. A quantitative balance mechanism is established between belief induction and geometric approximation. Specifically, firstly, a high-level intent set of the pursuing vessel is defined (including direct pursuit intent and intents to block each exit separately). One intent is selected as the decoy intent (i.e., the intent the pursuing vessel wants the escape vessel to mistakenly believe it is executing), and another different intent is selected as the real intent (the intent the pursuing vessel actually executes). The behavioral pattern of the escape vessel is then set: the escape vessel updates its belief distribution on various intents based on the observable motion state of the pursuing vessel, and then selects the target exit based on the time cost of each exit and the interception risk cost under the belief distribution. Based on this, the pursuing vessel estimates its current target exit according to the matching relationship between the escape vessel's course and the bearings of each exit. This estimated target is used as the induced target exit, and a set of candidate course angles is constructed around the pure pursuit direction and the induced blocking direction. Simultaneously, a single-step drift is constructed in the logarithmic domain using the posterior ratio between the decoy's intention and the true intention, and drift constraints are set to quantify the degree to which the current course promotes the escape vessel's belief in the decoy's intention. Finally, the computable lower bound of the drift for each candidate course is calculated, and courses that satisfy the drift constraints are selected to form a feasible set. The course with the smallest deviation from the pure pursuit direction is preferentially selected as the output course. If no feasible course is found, the vessel reverts to the pure pursuit direction and generates control commands. Compared with existing technologies, this method has the following advantages: (1) It breaks through the limitation of the traditional pursuit strategy of only responding passively, and enables the pursuing ship to actively influence the target exit selection by means of the escape ship's own belief inference mechanism, fundamentally changing the passive position of the pursuer in the multi-exit pursuit game. (2) By using drift constraints, a quantitative decision-making basis is established between cognitive induction and physical approximation, avoiding the dual dilemma that the pure pursuit strategy cannot prevent the escape ship from approaching the exit and the exit blocking strategy deviates too much from the escape ship. While inducing the escape ship to form a specific intention belief, it maintains a continuous geometric approximation of the escape ship, which significantly improves the interception success rate under the dynamic game of multiple exits. (3) A complete causal loop is formed between each step: “pursuing ship action → escape ship belief update → escape ship target switch → pursuing ship re-evaluation → pursuing ship re-action”, which integrates deception and autonomous decision-making, and provides systematic technical support for intelligent interception missions in waters with multiple exits and restricted navigation. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart of a deception pursuit method for autonomous pursuit missions provided in an embodiment of the present invention; Figure 2 A flowchart of the escape vessel belief-driven exit decision-making process provided in an embodiment of the present invention; Figure 3 This is a flowchart of a deception course decision-making process under the drift constraint of a pursuing ship, provided in an embodiment of the present invention. Figure 4 This is a schematic diagram illustrating the principle of candidate heading angle and drift selection for pursuing ships provided in an embodiment of the present invention; Figure 5 This is a simulation diagram of a multi-exit pursuit process provided in an embodiment of the present invention; Figure 6 This is a comparison diagram of the trajectories of three pursuit strategies in a multi-exit pursuit game provided in an embodiment of the present invention; Figure 7 A diagram of the aquatic test platform and communication control architecture provided in the embodiments of the present invention; Figure 8 This is a water area test diagram for a pursuit and escape game provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the deception pursuit system for autonomous pursuit missions provided in an embodiment of the present invention. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] Figure 1 This is a flowchart of a deception pursuit method for autonomous pursuit missions according to an embodiment of the present invention, with reference to... Figure 1 The method includes: S1. Define a set of high-level intentions for the pursuing ship, which includes a first intention to directly pursue the escaping ship and a second intention to block each exit. Select one intention from the set of high-level intentions as the decoy intention, and select another intention different from the decoy intention as the real intention.

[0013] The calculated set of high-level intentions for each exit refers to the set of strategic action objectives that the pursuing vessel may take. In a multi-exit pursuit scenario, this set includes two types: the first intention is to directly pursue the escaping vessel to shorten the distance between them; the second intention is to block each exit separately to reach a certain exit first and prevent the escaping vessel from passing. The decoy intention refers to the intention that the pursuing vessel hopes the escaping vessel will mistakenly believe it is carrying out—that is, the intention corresponding to the false belief the pursuing vessel wants the escaping vessel to form. The true intention is the high-level intention that the pursuing vessel actually executes. In a multi-exit navigable waterway, if there are K exits, the set of high-level intentions contains K+1 intentions: one direct pursuit intention plus K blocking intentions corresponding to each exit. High-level intentions .in, The intention was to directly pursue them; The intention was to blockade.

[0014] In existing technologies, the pursuing ship simply follows a fixed strategy, such as always pointing towards the escaping ship or towards a certain exit, without utilizing its own movement as an information source that can be observed and interpreted by the escaping ship. This step, by clearly defining all possible high-level intentions of the pursuing ship and distinguishing between decoy intentions and true intentions, provides a conceptual framework for the pursuing ship to manipulate the escaping ship's beliefs through course selection. It transforms the originally passive pursuit problem into an active belief manipulation problem. The pursuing ship no longer simply responds to the escaping ship's behavior, but by pre-selecting decoy intentions and true intentions, it establishes a clear "guided target direction" and "actual execution direction" for subsequent deceptive course decisions. This ensures that each course selection by the pursuing ship has a dual consideration: to make the escaping ship believe the decoy intention while actually executing the true intention.

[0015] S2. Define the behavior pattern of the escape ship: The escape ship updates its belief distribution on various intentions in the high-level intention set based on the observable motion state of the pursuing ship, and selects the target exit based on the time cost of each exit and the interception risk cost determined based on the belief distribution.

[0016] Here, the belief distribution refers to the escape ship's subjective probability estimate of the various high-level intentions of the pursuing ship. The escape ship cannot directly observe the true intentions of the pursuing ship; it can only maintain a posterior probability value for each type of intention based on the observed position, course, speed, and historical movement behavior of the pursuing ship. The sum of the probabilities of all intentions is 1. The time cost refers to the estimated time required for the escape ship to reach a certain exit from its current position. The interception risk cost refers to the probability estimate of whether the escape ship might be intercepted by the pursuing ship after choosing a certain exit, under the current belief distribution. This value is determined by the conditional risk function under different intentions of the pursuing ship. The total selection cost is the weighted sum of the time cost and the interception risk cost, based on which the escape ship selects its target exit.

[0017] In existing technologies, pursuit strategies often treat the escape vessel as a simple target that passively reacts to the environment, neglecting the escape vessel's proactive inference ability to deduce the pursuit intention based on the pursuer's motion and adjust its own decisions accordingly. This step, by coupling the pursuer's observable motion with the escape vessel's belief update and exit selection, reveals a complete causal chain of "pursuer's actions → escape vessel observation → belief update → exit decision." In this embodiment, S2 provides the theoretical basis for the pursuer to predict the escape vessel's behavior. The pursuer knows that the escape vessel will act according to the rule of "updating beliefs → assessing risks → choosing the exit with the minimum cost." Therefore, the pursuer can actively influence the input (belief distribution) of this rule by changing its own observable motion state, thereby affecting its output (target exit). This makes it possible for the pursuer to shift from "passively responding to the escape vessel's selection" to "actively influencing the escape vessel's decision-making process."

[0018] S3. The pursuing vessel determines the current target exit of the escape vessel based on the matching relationship between the heading status of the escape vessel and the bearings of each exit; the current target exit is used as the induced target exit, the pure pursuit direction of the pursuing vessel pointing towards the escape vessel and the induced blocking direction of the pursuing vessel pointing towards the induced target exit are determined, and discrete sampling is performed between the pure pursuit direction and the induced blocking direction to construct a set of candidate heading angles.

[0019] Here, "current target exit" refers to the exit that the escaping ship is currently heading towards, estimated by the pursuing ship based on the escaping ship's current course. "Pure pursuit direction" refers to the pursuing ship's course towards the escaping ship's current position, representing the geometrically optimal direction for directly approaching the escaping ship. "Induced blocking direction" refers to the pursuing ship's course towards the induced target exit (i.e., the estimated current target exit of the escaping ship), representing the optimal course when intending to block that exit. The candidate course angle set refers to a series of candidate course angle values ​​obtained by discrete sampling between the pure pursuit direction and the induced blocking direction.

[0020] Existing pure pursuit strategies make decisions based on only one direction (towards the escaping ship), while exit blocking strategies make decisions based on only another direction (towards the exit). Neither can simultaneously achieve the dual objectives of "approaching the escaping ship" and "inducing the escaping ship to change its exit." This step constructs a set of candidate heading angles between the pure pursuit direction and the induced blocking direction, allowing the pursuing ship's decision space to simultaneously cover two maneuver tendencies: a heading closer to the pure pursuit direction helps maintain geometrical approximation of the escaping ship, while a heading closer to the induced blocking direction helps convince the escaping ship that the pursuing ship is blocking its target exit. The set of candidate heading angles provides a pool of optional actions for subsequent drift constraint screening. The pursuing ship does not need to make an either-or choice between "pure pursuit" and "exit blocking," but can instead find the optimal compromise within a continuous spectrum between the two. This provides the technical prerequisite for balancing cognitive guidance and physical approximation.

[0021] S4. Based on the posterior ratio between the decoy intention and the true intention, obtain the single-step expected increment of the posterior ratio in the logarithmic domain, construct a single-step drift amount based on the single-step expected increment, and set drift constraints for filtering the candidate heading angle set based on the single-step drift amount.

[0022] Here, the posterior ratio refers to the ratio of the escaping vessel's probability of believing in the decoy's intention to its probability of believing in the true intention, used to quantify the escaping vessel's relative degree of belief in the two intentions. The single-step drift is the expected increment of the posterior ratio in the logarithmic domain, used to characterize the trend of the pursuing vessel's current course movement in terms of expected impact on the belief in the decoy's intention. The drift constraint is a screening rule based on the condition that the single-step drift exceeds a preset drift threshold; only candidate courses that meet this condition are considered to have an effective inducing effect.

[0023] In existing technologies, the pursuing vessel cannot determine whether its course maneuvers have truly induced the escape vessel to change its intentions, lacking a real-time means of evaluating the deception effect. This step introduces a posterior ratio between the decoy's intention and the true intention, defining its expected increment in the logarithmic domain as the single-step drift. This transforms the abstract concept of induction effectiveness into a calculable and comparable numerical indicator. A positive single-step drift indicates that the current course maneuver will increase the escape vessel's belief in the decoy's intention in the expected value (i.e., the induction is effective); a negative drift indicates that the induction has failed. The drift constraint ensures that each course choice by the pursuing vessel has a guaranteed induction effect. Only courses that can increase the escape vessel's belief in the decoy's intention are allowed to pass the screening, fundamentally ensuring that the pursuing vessel's maneuvers always serve the deception goal of making the escape vessel believe in the decoy's intention.

[0024] S5. Calculate the calculable lower bound of the drift for each candidate heading angle in the candidate heading angle set, retain the candidate heading angles that satisfy the drift constraint to form a feasible heading angle set, select the candidate heading angle with the smallest deviation from the pure pursuit direction as the output heading of the pursuing ship when the feasible heading angle set is not empty, use the pure pursuit direction as the output heading of the pursuing ship, and generate the pursuing ship control command according to the output heading of the pursuing ship.

[0025] The calculated drift lower bound refers to the lower bound value of the single-step drift for each candidate course, calculated based on the proximity of that course to the true intention direction and the decoy's intention direction. The closer the candidate course is to the decoy's intention direction and the farther it is from the true intention direction, the larger the drift lower bound. The feasible course angle set refers to the subset of course angles in the candidate course angle set that satisfy the drift constraint (i.e., the calculated drift lower bound is greater than the drift threshold). The pursuer's output course is the course that is finally selected to generate control commands. When the feasible set is not empty, the course with the smallest deviation from the pure pursuit direction is selected; when the feasible set is empty, it reverts to the pure pursuit direction.

[0026] Existing technologies lack an effective balancing mechanism when faced with the conflicting objectives of "inducement" and "approach." They either excessively deviate from the escape vessel in pursuit of inducement (a risk of blocking strategies) or abandon inducement in pursuit of approach (a limitation of pure pursuit strategies). This step resolves this contradiction through a two-stage decision-making logic of "constraint first, then selection": The first stage uses drift constraints to filter out routes that cannot effectively induce belief growth, ensuring that the retained routes are all deceptively effective; the second stage selects the route with the smallest deviation from the pure pursuit direction among the routes that meet the inducement requirements, maintaining geometrical approximation to the escape vessel as much as possible while ensuring the inducement effect. By adopting the decision rule of "selecting the one with the smallest deviation when the feasible set is not empty, and retreating to pure pursuit when it is empty", the pursuing ship achieves a dynamic balance between cognitive guidance and physical approximation. It has both clear guidance effect guarantee (drift constraint) and clear approximation capability guarantee (using the pure pursuit direction as the selection criterion and retreat direction), so that the pursuing ship has both initiative (inducing the escape ship to change its choice) and robustness (it can still maintain the pursuit posture even if the guidance fails) in the multi-exit pursuit game.

[0027] Based on the above embodiments, as a preferred implementation, in step S2, the belief distribution includes the posterior probability maintained by the escape ship for each intention in the set of high-level intentions, and the sum of the posterior probabilities of all intentions is 1; the escape ship updates the belief distribution according to the observable motion state of the currently acquired pursuing ship in a Bayesian recursive manner.

[0028] Specifically, in a multi-exit pursuit scenario, the escaping vessel cannot directly observe the true intentions of the pursuing vessel. It can only infer the higher-level intentions the pursuing vessel is pursuing based on observable information such as its position, course, speed, and historical movement behavior. Therefore, this embodiment predefines a set of higher-level intentions for the pursuing vessel. Φ This includes the primary intention of directly pursuing the escaping ship and the secondary intention of separately blocking each exit.

[0029] like Figure 2 As shown, the escape ship's exit selection process first defines the high-level intentions of the pursuing ship; then, it inputs the pursuing ship's status to form observation information. Based on the above observation information, the escape ship forms a basis for judging the pursuing ship's intentions and updates its belief distribution on various intentions of the pursuing ship based on Bayesian recursion.

[0030] Specifically, the distribution of beliefs about the pursuing ship's intentions regarding the escape ship's maintenance:

[0031] in, b i ( () indicates the time the escaped ship was at. i It is believed that the pursuing ships are carrying out the intentions of higher command. The posterior probability; i For a moment i The true intention of the pursuing ships; The cutoff time for the escaped ship i The history of observation; Φ A collection of intentions from higher levels; The probability normalization condition means that the sum of the belief probabilities of the escape ship regarding all intentions in the set of high-level intentions is 1.

[0032] When the escape ship was at the moment t After obtaining observation information from the pursuing ships, based on the observation history:

[0033] Update its belief distribution. Belief updates are performed using a Bayesian recursive approach:

[0034] That is, based on the belief of the previous moment b i ( Current observation information and in different intentions The likelihood probability of generating this observation information Calculate the posterior probability of the pursuing ship's various intentions at the current moment. b t+1 ( This completes the recursive update of the belief distribution. Among them, s t For a moment t The system status, For a moment t Control actions of the pursuing ship.

[0035] To facilitate real-time online calculations, this implementation interprets the observable actions of the pursuing ship as the result of rational control under a certain higher-level intention, defining the control cost as follows: in, The cost function for the control actions of the pursuing ship represents the state of the pursuing ship. s t Take control action Then, predict the endpoint of the trajectory. With the intentions of higher authorities Corresponding target direction μ ( s t The deviation between ) is half the square of the value; To control the actions of the pursuing ship From state under action s t The predicted endpoint of the starting trajectory; μ ( s t (This is the intention of the higher-ups) The target direction in the current state; α For accuracy parameters, α =1 / σ 2 , σ It indicates the clarity of the observed action.

[0036] Construct a likelihood function based on the proximity of the pursuing ship's current course to the directions of each intended target:

[0037] The likelihood function indicates that the current predicted trajectory endpoint of the pursuing ship... With a certain intention and target direction μ The closer the likelihood, the higher the probability of generating the current observation under that intention, and correspondingly, the greater the increase in the belief probability of that intention after the update. σ Parameters indicating the clarity of the observed action: σ The smaller the value, the clearer the observation and the faster the belief is updated; σ The larger the value, the greater the observation noise and the slower the belief update.

[0038] Substituting the likelihood function into the Bayesian recursive formula, the belief update becomes an explicitly computable form:

[0039] The denominator is the sum of all high-level intentions. The normalization constant obtained by summation. This explicit form is directly based on the currently observable control actions of the pursuing ship. Calculate the updated belief probability for each intention.

[0040] Through the aforementioned recursive process, the observable motion of the pursuing ship not only alters the geometrical relationship between the two sides but can also be interpreted by the escaping ship as having different intentions, such as directly pursuing, blocking the current exit, or blocking other exits. This causes the escaping ship's belief distribution to dynamically evolve with the pursuing ship's movement. Thus, a quantifiable correlation mechanism is established between the observable motion of the pursuing ship and the changes in the escaping ship's intentional beliefs: each course maneuver made by the pursuing ship affects the likelihood probability through the control cost function, and then influences the escaping ship's posterior beliefs about various intentions through Bayesian recursion, ultimately affecting the escaping ship's subsequent exit selection and replanning decisions. This belief update mechanism provides a quantitative basis for subsequent pursuing ships to actively manipulate the escaping ship's judgment of the pursuit intention through deliberate course design, thereby inducing the escaping ship to change its target exit selection.

[0041] Based on the above embodiments, as a preferred implementation, step S2, selecting the target exit according to the time cost of each exit and the interception risk cost determined based on the belief distribution, specifically includes: The escape vessel calculates the total choice cost for each exit, which is the sum of the calculated time cost of the exit and the calculated interception risk cost of the exit under the current belief distribution.

[0042] The interception risk cost is determined based on the conditional risk function under different intentions of the pursuing vessel. The conditional risk function includes: the direct pursuit risk function when the pursuing vessel executes the first intention; the blocking risk function for the calculated exit when the pursuing vessel executes the second intention for the calculated exit; and the indirect risk function for the calculated exit when the pursuing vessel executes the second intention for other exits besides the calculated exit. The indirect risk function is set with a reduction coefficient based on the spatial distance between the exits, and the reduction coefficient is negatively correlated with the spatial distance between the exits.

[0043] The escape vessel selects the exit with the lowest total cost as its target exit.

[0044] After obtaining the distribution of the pursuing ship's intentions and beliefs, the escaping ship calculates the total risk of its current target exit under different interception risk conditions. Let the current target exit of the escaping ship be the weighted sum of the probabilities of each intention and belief and the corresponding interception risk:

[0045] in, For the estimated time to reach the exit; For the assessment of interception risk under the current belief; Risk weights.

[0046] The escape ship ultimately chooses the exit that minimizes the total cost:

[0047] Conditional interception risks include at least three categories: direct pursuit risk when the pursuing vessel directly pursues the escaping vessel, current exit blocking risk when the pursuing vessel blocks the current exit, and indirect risk to the current exit when the pursuing vessel blocks other exits.

[0048] The pursuers directly chased the escape ship, when the pursuers intended to... At that time, the risk depends on whether the escape vessel will be overtaken before reaching the exit. This is due to the speed of the pursuers. greater than the escape ship speed The estimated time for the pursuers to catch up with the escape ship can be defined as:

[0049] At this point, the conditional risk function is defined as:

[0050] in, For the logistic function:

[0051] parameter This is the adjustment coefficient.

[0052] The pursuer blocks the current exit (exit blocking strategy). When the pursuer intends to block the exit... At that time, the risk is determined by the relative arrival times of both parties at the exit. The pursuer's arrival time at the exit is:

[0053] The corresponding conditional risk is defined as:

[0054] in To adjust the parameters.

[0055] The pursuers blocked other exits; when the pursuers intended to block other exits... At that time, the export may still be under some indirect threat, but the risk is lower than directly blocking the current export. Conditional risk is defined as:

[0056] in, This is a reduction factor. This reduction factor can be defined based on the spatial distance between outlets:

[0057] in This is the attenuation parameter.

[0058] Based on the above embodiments, as a preferred implementation, the escape vessel takes the target exit as the current target exit. When the interception risk cost of the current target exit continues to exceed a preset threshold for a preset duration, exit replanning is triggered, the total selection cost of each exit is recalculated, and the exit with the minimum total selection cost is updated as the new target exit.

[0059] Specifically, such as Figure 2 As shown, after calculating the total risk of the current target exit, the escape vessel determines whether the total risk exceeds a preset threshold. The escape vessel selects a target exit at the initial moment and maintains this selection in subsequent stages; the escape vessel will only proceed if the risk of the current target exit continuously exceeds the threshold. achieve τ hold 1 Only after a few seconds can the exit selection be restarted:

[0060] If the total risk of the current target export does not exceed the preset threshold, the escape vessel will continue to maintain the current target export and will not initiate export replanning.

[0061] After calculating the total cost of each exit, the escape vessel selects the exit with the lowest total cost as its current target exit. If exit replanning is triggered, the exit with the lowest total cost is updated as the new current target exit; if exit replanning is not triggered, the original target exit is maintained. Through the above process, the escape vessel's belief update, conditional risk assessment, total risk judgment, exit replanning, and target exit output form a closed-loop decision-making process.

[0062] Based on the above embodiments, as a preferred implementation, in step S3, the pursuing vessel determines the current target exit of the escape vessel according to the matching relationship between the course status of the escape vessel and the bearings of each exit, specifically including: Obtain the current heading angle of the escape vessel and the azimuth angle of the escape vessel to each exit.

[0063] Calculate the heading angle deviation between the current heading angle and each of the azimuth angles.

[0064] Each of the heading angle deviations is converted into a matching score between the escape vessel's heading and the corresponding exit bearing, and the heading angle deviation and the matching score are negatively correlated.

[0065] The exit with the highest matching score is determined as the current target exit of the escape vessel.

[0066] Specifically, such as Figure 3 As shown, before executing a deceptive pursuit course decision, the pursuing ship first needs to estimate the target exit that the escape ship is currently heading towards. The pursuing ship cannot directly know the escape ship's internal decision-making state, but it can infer its target exit by observing the escape ship's heading motion. When the escape ship is heading towards a certain exit, the closer its current heading is to the direction of that exit, the greater the probability that exit is the escape ship's target. Therefore, the pursuing ship introduces a target estimation method based on heading observations, using the degree of matching between the escape ship's current heading angle and the azimuth angles of each exit to estimate the escape ship's target exit.

[0067] Let the heading angle of the escape ship be... The escape ship reached the exit The azimuth is The angle error is defined as: .

[0068] Mapping errors to observation scores .

[0069] Pick This serves as an estimate of the escape vessel's target exit for the pursuer. To suppress jitter, this embodiment uses exponential moving average smoothing on the observed course and sets a duration threshold. Only if the new estimate remains continuously above The pursuer's estimated target is only updated periodically.

[0070] The pursuer constructs candidate heading angles only between two reference directions, corresponding to direct pursuit and induced blockade, respectively. The heading of the pursuer directly pointing to the current position of the escape vessel is defined as:

[0071] Let the current target exit be... Its representative point location is The course of the pursuer pointing towards the exit of the induced target is defined as... .

[0072] exist and Between these points, M+1 candidate heading angles are obtained by uniform discretization, denoted as:

[0073] in, .

[0074] Based on the above embodiments, as a preferred implementation, after determining the current target exit of the escape vessel, the method further includes: Multiple consecutive time-series determined results are smoothed using an exponential moving average method to obtain a smoothed exit determination result.

[0075] The pursuing vessel is also equipped with a duration threshold. Only when the same smoothed exit determination result is held continuously for a period of time exceeding the duration threshold will the pursuing vessel output the smoothed exit determination result as the updated current target exit of the escape vessel.

[0076] Specifically, in real-world aquatic environments, due to factors such as observation noise, wind and wave disturbances, and natural fluctuations in the escape vessel's course, the target exit determined by the pursuing vessel based on a single course observation may frequently change, meaning the estimation result switches back and forth between multiple exits within a short period. If the pursuing vessel directly uses each estimation result to determine the induced blocking direction, it will cause the pursuing vessel's induced blocking direction to oscillate frequently, thus affecting the stability and consistency of the deceptive pursuit course decision. To suppress the aforementioned jitter, this embodiment smooths multiple time-series consecutive determination results and sets a duration threshold as an update condition.

[0077] Therefore, after obtaining a confirmed exit result for a target at each moment, the pursuing vessel smooths multiple consecutive confirmed results using an exponential moving average method to obtain a smoothed exit determination result. The exponential moving average method assigns different weights to historical confirmed results, with greater weights for results closer to the current moment and smaller weights for results further away from the current moment. This effectively suppresses the impact of short-term jitter on the estimation results while preserving current observation information.

[0078] Meanwhile, the pursuing ship is also equipped with a duration threshold. T hold Only if the same smoothed exit determination result is continuously maintained for a period exceeding the duration threshold. T hold Only then will the pursuing ship use the smoothed exit determination result as the updated current target exit output of the escaping ship.

[0079] In other words, after the pursuing ship obtains a smoothed exit determination result at each moment, it compares it with the smoothed exit determination result from the previous moment. If they match, the continuous holding time of that result is accumulated; if they do not match (i.e., a switch has occurred), the continuous holding time is reset, and the timing restarts. Only when the continuous holding time of the same smoothed exit determination result exceeds a preset duration threshold will the process continue. T hold Only then will the pursuing vessel formally determine the smoothed exit determination result as the updated current target exit of the escape vessel, and use it to subsequently determine the direction of induced blockade.

[0080] If the continuous holding time of the smoothed exit determination result does not exceed the duration threshold. T hold If the pursuing ship does not update the current target exit, it will continue to maintain the determined result from the previous moment.

[0081] By combining the above-mentioned exponential moving average smoothing with duration threshold processing mechanism, the frequent jumps in estimation results caused by observation noise, environmental disturbances or instantaneous jitter of the escape vessel's course can be effectively suppressed, ensuring the stability, continuity and reliability of the pursuit vessel's estimation of the escape vessel's current target exit, avoiding frequent oscillations in the pursuit vessel's induced blocking direction due to the instability of the estimation results, thereby ensuring the consistency and effectiveness of subsequent deception pursuit course decisions.

[0082] Based on the above embodiments, as a preferred implementation, in step S3, discrete sampling is performed between the pure pursuit direction and the induced blocking direction to construct a candidate heading angle set, specifically including: The pure pursuit direction is taken as the first boundary direction of the candidate heading angle set, and the induced blocking direction is taken as the second boundary direction of the candidate heading angle set.

[0083] Within the angular range between the first boundary direction and the second boundary direction, uniform discrete sampling is performed according to a preset number of samples to obtain multiple sampling directions.

[0084] The first boundary direction, the second boundary direction, and each of the sampling directions are collectively used as candidate heading angles, and the candidate heading angles form a candidate heading angle set.

[0085] Based on the above embodiments, as a preferred implementation, step S4 specifically includes: The posterior ratio is defined as the ratio of the probability that the escape vessel believes in the decoy's intention to the probability that the escape vessel believes in the true intention.

[0086] Obtain the single-step expected increment of the posterior ratio in the logarithmic domain, and use the single-step expected increment as the single-step drift.

[0087] When the single-step drift is positive, the current course of the pursuing vessel promotes the escape vessel's belief in the decoy's intention; when the single-step drift is negative, the current course of the pursuing vessel inhibits the escape vessel's belief in the decoy's intention.

[0088] The drift constraint is that the single-step drift amount is greater than a preset drift threshold.

[0089] Specifically, after constructing the candidate heading angle set, the pursuing ship introduces a posterior ratio between the decoy's intention and the true intention to characterize the evolution of the escape ship's belief over time. Let the true intention be... * The bait is intended to d Define the posterior ratio:

[0090] in, R t For a moment t The posterior ratio between decoy intent and true intent b t ( d (time) t The escaping ship believed the pursuing ship was carrying out a decoy maneuver. d The posterior probability, b t ( * (time) t The escaping ship believed the pursuing ship was carrying out its true intentions. * The posterior probability. This posterior ratio R t This reflects the relative degree of belief of the escape vessel regarding the decoy's intentions versus its true intentions: R t The larger the value, the more the escaping ship believes that the pursuing ship is carrying out a decoy mission, and the better the deception effect. R t The smaller the value, the more likely the escaping ship is to believe that the pursuing ship is carrying out its true intentions.

[0091] Based on the Bayesian update, the posterior ratio can be written in logarithmic recursive form:

[0092] Among them, increment Z t+1 for:

[0093] Right now Z t+1 In the state s t and control actions Under the conditions, the decoy's intention d Observational likelihood probability and true intention * The logarithm of the ratio of the observed likelihood probabilities.

[0094] Define single-step drift (conditional expected increment):

[0095] in, This represents the single-step drift amount, indicating the drift in the current state. s t Belief Distribution b t and the control actions of the pursuit ship Below, the log-likelihood ratio is greater than the increment. Z t+1 The expected value of the condition. This single-step drift amount. This characterizes the degree to which the pursuer's current course maneuver contributes to the escape vessel's belief in the decoy's intention. Drift depicts the tendency of the pursuer's maneuver, in the current state, to increase or decrease the belief in the decoy's intention in a desired sense. If the posterior probability of the decoy intention is greater than 0, then the expected probability increases, meaning the current course of the pursuing vessel promotes the escape vessel's increased belief in the decoy intention; if If the value is less than 0, the posterior probability of the decoy's intention decreases in expectation, meaning the current course of the pursuing vessel inhibits the escape vessel's belief in the decoy's intention from growing. The drift constraint is the single-step drift amount. It is greater than the preset drift threshold (usually 0).

[0096] The pursuing vessel further calculates the computable lower bound of drift for each candidate course. Let the observed features extracted by the escaping vessel from the pursuer's motion be the unit direction vector of the pursuer's next step, i.e.:

[0097]

[0098] For any high-level intention Define the target direction:

[0099] Assume the intentional conditional observation model of the escape vessel is as follows:

[0100] Among them, noise level σ Errors can be estimated from simulation / actual ship observations.

[0101] Consider the intention of the decoy d With true intention * Define the noise-free "actual direction":

[0102] Under the aforementioned Gaussian model, the log-likelihood ratio increment Z t+1 The conditional expectation can be written as (from the standard Gaussian log-likelihood algebra):

[0103] Therefore, as long as the actual direction is closer to the decoy's direction than the true direction, the drift is positive, thus obtaining a calculable lower bound for the drift, which can be obtained using the identity. and The above equation simplifies to:

[0104] This expression indicates that if the pursuer's actual course is closer to the direction corresponding to the decoy's intention and relatively farther from the true intention direction, its guiding effect is stronger, and the corresponding lower bound of drift is larger. Conversely, if the pursuer's movement is closer to the true intention direction, the lower bound of drift will decrease, and may even be negative.

[0105] By comparing the lower bound of the drift of each candidate course with a preset drift threshold, the pursuing ship retains candidate courses that meet the drift threshold, forming a set of feasible courses. If the set of feasible courses is not empty, the pursuing ship selects the candidate course with the smallest deviation from the pure pursuit direction as the optimal course; if the set of feasible courses is empty, the pursuing ship retreats to the pure pursuit direction. This rule ensures that the pursuing ship minimizes deviation from the pure pursuit direction while maintaining the growth of the decoy's intention belief, thus balancing belief induction and physical approximation.

[0106] like Figure 4 As shown, the candidate heading angle and drift selection principle can be further expressed as follows: The pursuing ship constructs a candidate heading region with the pure pursuit direction and the induced target exit direction as the boundary directions; it judges whether the candidate heading meets the induced requirements based on the drift lower bound corresponding to the candidate heading; the candidate heading that meets the drift threshold is taken as the feasible heading; finally, the feasible heading with the smallest angle with the pure pursuit direction is selected as the output heading of the pursuing ship. If all candidate headings do not meet the drift threshold, the pure pursuit direction is output.

[0107] Based on the above embodiments, as a preferred implementation, the pursuing ship outputs a heading that is either one of the candidate heading angles in the set of candidate heading angles or the pure pursuing direction; The pursuing ship control commands include at least one of the following: desired course command, desired yaw rate command, and desired speed command; the pursuing ship generates a desired yaw rate based on the course control deviation between the current course and the output course of the pursuing ship, and adjusts the desired longitudinal speed based on the turning amplitude corresponding to the course control deviation.

[0108] The pursuing vessel converts the selected course into control commands. These control commands may include desired course, desired yaw rate, or desired speed. Specifically, the pursuing vessel can generate a desired yaw rate based on the angular error between the current course and the output course, and adjust the desired longitudinal speed according to the turning angle; when the turning angle is large, the longitudinal speed is reduced to ensure turning execution, and when the turning angle is small, the longitudinal speed is increased to shorten the distance to the escaping vessel. After receiving the control commands, the underlying control system drives the propulsion actuator, causing the pursuing vessel to move according to the deceptive pursuing course.

[0109] In this embodiment, the deception pursuit method is also verified in a closed loop based on a simulation-water verification platform. After completing the steps of the above embodiments, this embodiment further verifies the deception pursuit method in a closed loop based on a simulation-water verification platform for ship pursuit missions.

[0110] In one specific implementation, the simulation verification unit can be implemented based on the VRX platform or other simulation platforms with water surface motion modeling capabilities. The simulation verification unit includes a scene configuration module, a ship model module, a pursuit and escape decision module, a disturbance setting module, and a data recording module. The scene configuration module is used to set multiple exit water areas, exit locations, obstacle areas, capture radius, and mission termination conditions; the ship model module is used to provide motion models for the pursuing and escaping ships; the pursuit and escape decision module is used to run the deception pursuit method for autonomous pursuit missions in Embodiment 1; the disturbance setting module is used to set undisturbed, wind and wave disturbance, or other environmental disturbance conditions; and the data recording module is used to record the state, beliefs, risks, heading, drift, and control data during the pursuit and escape process.

[0111] like Figure 5As shown, in the simulation verification, a multi-exit waterway pursuit simulation scenario can be constructed, setting up a pursuing vessel, an escape vessel, multiple escape exits, and mission termination conditions. During the simulation, the escape vessel executes the belief-driven exit decision process in the above embodiment, while the pursuing vessel executes the deception pursuit course decision process under drift constraints in the above embodiment, thus forming a complete pursuit simulation closed loop. The simulation process can record process data such as pursuit trajectory, escape vessel belief changes, current exit risk, target exit switching, candidate courses, drift lower bound, pursuit distance, and control commands.

[0112] like Figure 6 As shown, to illustrate the difference between the described deceptive pursuit method and traditional strategies, a pure pursuit strategy, an exit blocking strategy, and the deceptive pursuit method of this invention can be configured under the same or comparable initial conditions. The pure pursuit strategy ensures the pursuing vessel always points towards the current position of the escaping vessel; the exit blocking strategy prioritizes the pursuing vessel towards the current target exit of the escaping vessel; and the deceptive pursuit method of this invention selects a course based on drift constraints between the pure pursuit direction and the induced target exit direction. By comparing the pursuit and escape processes under different strategies, the effectiveness of the method of this invention in inducing the escaping vessel to adjust its exit selection and maintaining the pursuing vessel's approach capability can be verified.

[0113] In simulation verification, both undisturbed and wind / wave-disturbed conditions can be set. By changing wind field, wave, or other environmental disturbance parameters, the actual motions of the pursuing and escaping vessels are affected by disturbances, and the belief update, exit replanning, and pursuit course selection processes under disturbance conditions are recorded. This verifies that the method can still complete pursuit / escape decision calculations and control execution under environmental disturbances.

[0114] like Figure 7 As shown, the water-based test unit includes a shore-based terminal, a communication network, a pursuit vessel's end-of-ship system, and an escape vessel's end-of-ship system. The shore-based terminal is used to configure pursuit and escape missions, set the operational area, set exit locations, display the status of the pursuit and escape vessels, and save test data. The communication network is used to enable data interaction between the shore-based terminal and the pursuit and escape vessel's end-of-ship systems, including mission parameter distribution, vessel status feedback, dual-vessel status sharing, and test data transmission.

[0115] Both the pursuing vessel's and the escaping vessel's end-board systems can include a positioning and orientation module, a communication module, an onboard computing module, a flight control module, and a propulsion execution module. The positioning and orientation module acquires the vessel's position, heading, and speed; the communication module facilitates data transmission between the vessel and the shore, as well as between the two vessels; the onboard computing module runs the pursuit / escape decision-making algorithm; the flight control module receives high-level control commands output by the onboard computing module and converts these commands into low-level control signals that the propulsion execution module can execute; the propulsion execution module drives the vessel's movement based on these low-level control signals.

[0116] like Figure 8 As shown, during the water trial, the shore-based system first sets the operating area, exit position, and pursuit / escape mission parameters, and then sends these parameters to the pursuing and escaping vessels via a communication network. The escaping vessel's end-board system operates a belief-driven exit decision-making process based on its own state, the pursuing vessel's state, and the exit position to obtain the current target exit. The pursuing vessel's end-board system operates a deception pursuit course decision-making process under drift constraints based on the escaping vessel's state and exit information, generating pursuit vessel control commands. After receiving the control commands output by the onboard computing module, the flight control module drives the propulsion execution module to execute the corresponding motion. During the motion, the pursuing and escaping vessels continuously transmit information such as position, heading, speed, target exit, and mission status back to the shore-based system, enabling the shore-based system to display and record the pursuit / escape process in real time.

[0117] Through the coordinated operation of the shore-based system, communication network, shipboard system, and propulsion actuators, the water-based test unit forms a verification architecture of "shore-based mission configuration - shipboard status perception - pursuit and escape decision calculation - flight control execution and control - status data feedback". This architecture can transform the deception pursuit method described in Example 1 from an algorithm calculation process into a closed-loop ship control system in actual waters, used to verify the feasibility of the method under real positioning, communication, and control execution conditions.

[0118] Secondly, embodiments of the present invention provide a deception pursuit system for autonomous pursuit missions, based on the methods described in the above embodiments, such as... Figure 9 As shown, the system includes: The intent definition module 901 is used to define a set of high-level intents for the pursuing ship. The set of high-level intents includes a first intent to directly pursue the escaping ship and a second intent to block each exit. An intent is selected from the set of high-level intents as a decoy intent, and another intent different from the decoy intent is selected as the real intent.

[0119] The escape ship behavior estimation module 902 is used to set the behavior rules of the escape ship: the escape ship updates the belief distribution of the escape ship to various intentions in the high-level intention set according to the observable motion state of the pursuing ship, and selects the target exit according to the time cost of each exit and the interception risk cost determined based on the belief distribution.

[0120] The target estimation and heading construction module 903 is used by the pursuing ship to determine the current target exit of the escape ship based on the matching relationship between the heading status of the escape ship and the bearings of each exit; using the current target exit as the induced target exit, determining the pure pursuit direction of the pursuing ship pointing towards the escape ship, and the induced blocking direction of the pursuing ship pointing towards the induced target exit, and performing discrete sampling between the pure pursuit direction and the induced blocking direction to construct a set of candidate heading angles.

[0121] The drift constraint construction module 904 is used to obtain the single-step expected increment of the posterior ratio in the logarithmic domain based on the posterior ratio between the decoy intention and the true intention, construct a single-step drift amount based on the single-step expected increment, and set drift constraints for filtering the candidate heading angle set based on the single-step drift amount.

[0122] The heading selection and output module 905 is used to calculate the calculable drift lower bound of each candidate heading angle in the candidate heading angle set, retain the candidate heading angles that satisfy the drift constraint to form a feasible heading angle set, select the candidate heading angle with the smallest deviation from the pure pursuit direction as the output heading of the pursuing ship when the feasible heading angle set is not empty, use the pure pursuit direction as the output heading of the pursuing ship when the feasible heading angle set is empty, and generate the pursuing ship control command according to the output heading of the pursuing ship.

[0123] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A deception-based pursuit method for autonomous pursuit missions, characterized in that, include: S1. Define the high-level intention set of the pursuing ship, which includes the first intention to directly pursue the escaping ship and the second intention to block each exit respectively. Select one intent from the set of high-level intents as the decoy intent, and select another intent different from the decoy intent as the real intent; S2. Define the behavior pattern of the escape ship: The escape ship updates its belief distribution on various intentions in the high-level intention set according to the observable motion state of the pursuing ship, and selects the target exit according to the time cost of each exit and the interception risk cost determined based on the belief distribution. S3. The pursuing vessel determines the current target exit of the escape vessel based on the matching relationship between the heading status of the escape vessel and the bearings of each exit; the current target exit is used as the induced target exit, the pure pursuit direction of the pursuing vessel pointing towards the escape vessel and the induced blocking direction of the pursuing vessel pointing towards the induced target exit are determined, and discrete sampling is performed between the pure pursuit direction and the induced blocking direction to construct a candidate heading angle set; S4. Based on the posterior ratio between the decoy intention and the true intention, obtain the single-step expected increment of the posterior ratio in the logarithmic domain, construct a single-step drift amount based on the single-step expected increment, and set drift constraints for filtering the candidate heading angle set based on the single-step drift amount. S5. Calculate the calculable lower bound of the drift for each candidate heading angle in the candidate heading angle set, retain the candidate heading angles that satisfy the drift constraint to form a feasible heading angle set, select the candidate heading angle with the smallest deviation from the pure pursuit direction as the output heading of the pursuing ship when the feasible heading angle set is not empty, use the pure pursuit direction as the output heading of the pursuing ship, and generate the pursuing ship control command according to the output heading of the pursuing ship.

2. The deception pursuit method for autonomous pursuit missions according to claim 1, characterized in that, In step S2, the belief distribution includes the posterior probability maintained by the escape ship for each intention in the set of high-level intentions, and the sum of the posterior probabilities of all intentions is 1; the escape ship updates the belief distribution according to the observable motion state of the currently acquired pursuing ship in a Bayesian recursive manner.

3. The deception pursuit method for autonomous pursuit missions according to claim 1, characterized in that, In step S2, the target exit is selected based on the time cost of each exit and the interception risk cost determined based on the belief distribution. Specifically, this includes: The escape vessel calculates the total choice cost for each exit, which is the sum of the calculated time cost of the exit and the calculated interception risk cost of the exit under the current belief distribution. The interception risk cost is determined based on conditional risk functions under different intentions of the pursuing vessel. These conditional risk functions include: a direct pursuit risk function when the pursuing vessel executes its first intention; a blocking risk function for the calculated exit when the pursuing vessel executes its second intention for the calculated exit; and an indirect risk function for the calculated exit when the pursuing vessel executes its second intention for other exits besides the calculated exit. The indirect risk function has a reduction coefficient based on the spatial distance between exits, and the reduction coefficient is negatively correlated with the spatial distance between exits. The escape vessel selects the exit with the lowest total cost as its target exit.

4. The deception pursuit method for autonomous pursuit missions according to claim 3, characterized in that, The escape vessel uses the target exit as the current target exit. When the interception risk cost of the current target exit exceeds a preset threshold for a preset duration, it triggers exit replanning, recalculates the total selection cost of each exit, and updates the exit with the minimum total selection cost as the new target exit.

5. The deception pursuit method for autonomous pursuit missions according to claim 1, characterized in that, In step S3, the pursuing vessel determines the current target exit of the escape vessel based on the matching relationship between the escape vessel's course status and the bearings of each exit, specifically including: Obtain the current heading angle of the escape vessel and the azimuth angle of the escape vessel to each exit; Calculate the heading angle deviation between the current heading angle and each of the azimuth angles; Each of the heading angle deviations is converted into a matching score between the escape vessel's heading and the corresponding exit bearing, and the heading angle deviation and the matching score are negatively correlated. The exit with the highest matching score is determined as the current target exit of the escape vessel.

6. The deception pursuit method for autonomous pursuit missions according to claim 5, characterized in that, After determining the current target exit of the escape vessel, the process also includes: Multiple consecutive time-series determination results are smoothed using an exponential moving average method to obtain a smoothed exit determination result. The pursuing vessel is also equipped with a duration threshold. Only when the same smoothed exit determination result is held continuously for a period of time exceeding the duration threshold will the pursuing vessel output the smoothed exit determination result as the updated current target exit of the escaping vessel.

7. The deception pursuit method for autonomous pursuit missions according to claim 1, characterized in that, In step S3, discrete sampling is performed between the pure pursuit direction and the induced blocking direction to construct a candidate heading angle set, specifically including: The pure pursuit direction is taken as the first boundary direction of the candidate heading angle set, and the induced blocking direction is taken as the second boundary direction of the candidate heading angle set. Within the angular range between the first boundary direction and the second boundary direction, uniform discrete sampling is performed according to a preset number of samples to obtain multiple sampling directions; The first boundary direction, the second boundary direction, and each of the sampling directions are collectively used as candidate heading angles, and the candidate heading angles form a candidate heading angle set.

8. The deception pursuit method for autonomous pursuit missions according to claim 1, characterized in that, S4 specifically includes: The posterior ratio is defined as the ratio of the probability of the escape vessel's belief in the decoy's intention to the probability of the escape vessel's belief in the true intention; Obtain the single-step expected increment of the posterior ratio in the logarithmic domain, and use the single-step expected increment as the single-step drift. When the single-step drift is positive, the current course of the pursuing vessel promotes the escape vessel's belief in the decoy's intention; when the single-step drift is negative, the current course of the pursuing vessel inhibits the escape vessel's belief in the decoy's intention. The drift constraint is that the single-step drift amount is greater than a preset drift threshold.

9. The deception pursuit method for autonomous pursuit missions according to claim 1, characterized in that, The pursuing ship outputs a heading that is either one of the candidate heading angles in the set of candidate heading angles or the pure pursuing direction; The pursuing ship control commands include at least one of the following: desired course command, desired yaw rate command, and desired speed command; the pursuing ship generates a desired yaw rate based on the course control deviation between the current course and the output course of the pursuing ship, and adjusts the desired longitudinal speed based on the turning amplitude corresponding to the course control deviation.

10. A deception-based pursuit system for autonomous pursuit missions, characterized in that: include: The intent definition module is used to define the high-level intent set of the pursuing ship, which includes a first intent to directly pursue the escaping ship and a second intent to block each exit respectively; Select one intent from the set of high-level intents as the decoy intent, and select another intent different from the decoy intent as the real intent; The escape ship behavior estimation module is used to set the behavior rules of the escape ship: the escape ship updates the belief distribution of the escape ship to various intentions in the high-level intention set according to the observable motion state of the pursuing ship, and selects the target exit according to the time cost of each exit and the interception risk cost determined based on the belief distribution; The target estimation and course construction module is used by the pursuing ship to determine the current target exit of the escape ship based on the matching relationship between the course status of the escape ship and the bearings of each exit. Using the current target exit as the induced target exit, determine the pure pursuit direction of the pursuing vessel towards the escaping vessel, and the induced blocking direction of the pursuing vessel towards the induced target exit. Discrete sampling is performed between the pure pursuit direction and the induced blocking direction to construct a candidate heading angle set. The drift constraint construction module is used to obtain the single-step expected increment of the posterior ratio in the logarithmic domain based on the posterior ratio between the decoy intention and the true intention, construct a single-step drift amount based on the single-step expected increment, and set drift constraints for filtering the candidate heading angle set based on the single-step drift amount. The heading selection and output module is used to calculate the calculable lower bound of the drift of each candidate heading angle in the candidate heading angle set, retain the candidate heading angles that satisfy the drift constraint to form a feasible heading angle set, select the candidate heading angle with the smallest deviation from the pure pursuit direction as the output heading of the pursuing ship when the feasible heading angle set is not empty, use the pure pursuit direction as the output heading of the pursuing ship when the feasible heading angle set is empty, and generate the pursuing ship control command according to the output heading of the pursuing ship.

Images