Multi-agent collaborative search path planning method based on time sequence constraint

Abstract

The invention belongs to the technical field of task allocation, and particularly relates to a multi-agent collaborative search path planning method based on timing constraint. The method comprises the following steps: initializing an agent set and a to-be-searched target point set; a first search stage is started, and a global timer starts timing and sequentially executes alternative target point search operation on all agents; continuously sliding forward for a time window along with timing of the global timer, and when the alternative target point of any agent enters the time window, executing a planned path updating operation on the agent; and performing alternative target point search operation on all the intelligent agents in sequence until the target point set to be searched is an empty set or the timing of the global timer reaches the termination moment of the first search stage. According to the method, the corresponding agent is selected based on the time window to execute the search task, and the situation that the global search efficiency is low due to the fact that a certain agent falls into local optimal search is avoided.

Description

Technical Field

[0001] The present application belongs to the technical field of task allocation, and specifically relates to a multi-agent collaborative search path planning method based on timing constraints. Background Art

[0002] With the rapid development of artificial intelligence and information technology, unmanned system technology has become increasingly mature, and its autonomous and reliable working capabilities have also been significantly improved. On this basis, the task allocation and path planning technology between unmanned systems has also produced relatively more research results. In the military field, unmanned systems can not only achieve collaboration in the same field, but also cross-domain collaboration in different fields (sea, land, and air). A collaborative mechanism can be established between unmanned system devices, and through information sharing and integration, task collaboration and cooperation, functional complementation and energy efficiency can be achieved, thereby realizing the same-domain or cross-domain unmanned system devices to perform collaborative combat tasks; in the civilian field, the importance of unmanned systems in the fields of unmanned autonomous monitoring of the marine aquaculture industry, marine meteorological monitoring operations, maritime search and rescue, and marine resource exploration and development is becoming increasingly significant. At present, the research on the same-domain or cross-domain collaboration of unmanned systems can be roughly divided into collaborative track planning, collaborative formation, and collaborative take-off and landing according to the mission scenario.

[0003] In the field of collaborative planning of unmanned systems, the current mainstream research directions are the research on collaborative area coverage scanning technology and collaborative search path planning technology. The former is mainly aimed at the situation where the target to be searched in the area is unknown, while the latter is mainly to solve the problem of full coverage of known target points to be searched. At present, in the technical field of collaborative search path planning, the allocation of all target points to be searched is mainly achieved by allocating tasks to the agents in the unmanned system one by one and cyclically, so as to achieve full coverage of the target points to be searched and collaborative planning between unmanned systems. Although this method can realize the rotation, one-by-one planning and allocation of agents, the search efficiency of different agents is uneven whether in the same domain or across domains. This makes the collaborative planning not reasonably planned according to the specific situation of the agent, reduces the search efficiency, and prolongs the total time of the global task. Summary of the invention

[0004] The purpose of this application is to provide a multi-agent collaborative search path planning method that improves search efficiency and ensures the total time of the global task.

[0005] The embodiments of the present application can be implemented through the following technical solutions:

[0006] A multi-agent collaborative search path planning method based on time constraints includes the following steps:

[0007] S100: Initializing an agent set and a target point set to be searched, wherein the agent set includes at least two agents, and the target point set to be searched includes at least two target points to be searched;

[0008] S200: Start the first search phase, the global timer starts timing and performs the candidate target point search operation on all agents in turn;

[0009] S300: continuously sliding a time window forward as the global timer counts, and when the candidate target point of any agent enters the time window, performing a planning path update operation on the agent;

[0010] S400: re-execute the candidate target point search operation for all agents in sequence, and then return to execute step S300 until the set of target points to be searched is an empty set or the global timer reaches the end time of the first search phase.

[0011] Furthermore, for any agent, the candidate target point search operation includes:

[0012] Obtain the current position and estimated departure time of the agent, wherein the current position is the task starting position of the agent or the position of the most recently marked target point;

[0013] Based on the current position of the agent and under the constraints of the task constraints, the candidate target point of the agent is determined by searching the target point optimization function, and the time when the agent arrives at its candidate target point is determined based on the estimated departure time.

[0014] Furthermore, the planned path update operation includes the following steps:

[0015] C100: Determine whether any agent has reached its candidate target point and entered the time window. If yes, execute steps C200 to C400. If no, execute step C100 again after delaying the time step Δt.

[0016] C200: Mark the candidate target point as the determined target point of the agent, and extend the planned path of the agent from the last marked determined target point to the determined target point;

[0017] C300: taking the determined target point as the current position of the agent and updating the estimated departure time of the agent;

[0018] C400: Delete the determined target point from the set of target points to be searched.

[0019] Preferably, the agent includes at least one first type of agent and at least one second type of agent;

[0020] The search speed of the first type of intelligent agent is greater than the search speed of the second type of intelligent agent and the battery life of the first type of intelligent agent is less than the battery life of the second type of intelligent agent.

[0021] Furthermore, after step S400, the following steps are also included:

[0022] S500: starting the second search phase, re-performing the candidate target point search operation on all first-category agents in turn, performing the parking operation on all second-category agents and obtaining their position information;

[0023] S600: continuously sliding a time window forward as the global timer counts, and when a candidate target point of any first-category agent enters the time window, performing a planning path update operation on the first-category agent;

[0024] S700: re-execute the candidate target point search operation for all first-category agents in sequence, and then return to execute step S600 until the set of target points to be searched is an empty set or the global timer reaches the end time of the second search phase.

[0025] Furthermore, the mission constraints include return time constraints;

[0026] When in the first search phase, the return time constraint condition is that the time it takes for the agent to return from its candidate target point to the starting position of the task is less than its endurance time;

[0027] When in the second search phase, the return time constraint condition is that the movement time of the first type of agent from its candidate target point to the second type of agent closest to it is less than its endurance time.

[0028] Preferably, when the set of target points to be searched is not an empty set, the first search phase is terminated when at least one first-type agent fails to search and determine a candidate target point for the first time;

[0029] When the set of target points to be searched is not an empty set and all first-type agents are unable to search and determine candidate target points, the second search phase terminates.

[0030] Furthermore, step S700 also includes the following steps:

[0031] When any first-category agent is unable to search and determine an alternative target point when the target set to be searched is not an empty set, it moves to the second-category agent closest to it.

[0032] Furthermore, the task constraints also include the following constraints:

[0033] a. Each target point to be searched can only be marked as a determined target point of one agent;

[0034] b. The agent that needs to update global information during the search can communicate with other agents directly or indirectly via other agents;

[0035] c. Collisions between agents are not allowed at any time.

[0036] Preferably, the candidate target point search operation is performed in sequence based on the search priority of each of the agents.

[0037] The embodiment of the present application provides a multi-agent collaborative search path planning method based on timing constraints, which has at least the following beneficial effects:

[0038] The multi-agent collaborative search path planning method based on timing constraints provided in the embodiment of the present application selects the corresponding agent to perform the search task based on the time window, thereby avoiding the situation where a certain agent falls into the local optimal search and causes low global search efficiency;

[0039] In this application, based on the diversity of search tasks, a first type of intelligent agent and a second type of intelligent agent are set to collaboratively complete the search task. Based on the performance difference between the intelligent agents, the search task is divided into a first search stage and a second search stage. The first type of intelligent agent performs the search task in both the first search stage and the second search stage. The second type of intelligent agent performs the search task in the first search stage and performs the task of picking up the first type of intelligent agent for return in the second search stage. This ensures that the first type of intelligent agent completes as many search tasks as possible within the endurance time, improves the global search efficiency, and also ensures that the first type of intelligent agent can be led back by the second type of intelligent agent with sufficient endurance time when the endurance time is insufficient.

[0040] In this application, the termination of the first search phase and the second search phase is achieved by limiting the return time constraint, which takes into account the actual operation of the agent and enables the agent to complete the search task in different search phases with maximum efficiency;

[0041] In this application, by setting the internal priority of the intelligent agent, the situation of competition and confusion caused by multiple intelligent agents being in the time window at the same time is avoided. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic diagram of the existing multi-agent collaborative search path planning method;

[0043] Figure 2 This is a flow chart of Embodiment 1 of the multi-agent collaborative search path planning method based on timing constraints in this application;

[0044] Figure 3A schematic diagram of path planning for a specific embodiment at time T0;

[0045] Figure 4 A schematic diagram of path planning for a specific embodiment at time T1;

[0046] Figure 5 It is a line graph showing the change of target search coverage with α;

[0047] Figure 6 It is a line graph showing the change of target weight coverage with α;

[0048] Figure 7 A flowchart of the operation of updating the planned path for the agent in this application;

[0049] Figure 8 A schematic diagram of path planning for a specific embodiment at time T2;

[0050] Fig. 9 A schematic diagram of path planning for a specific embodiment at time T3;

[0051] Fig.10 A schematic diagram of path planning for a specific embodiment at time T4;

[0052] Fig.11 This is a schematic diagram of the process in which different target points are planned as search paths for different agents during the global timer sliding process from T0 to T4;

[0053] Fig.12 A schematic diagram of the collaborative search path method of multiple agents in this application;

[0054] Fig.13 A schematic diagram showing a comparative analysis between the existing multi-agent collaborative search path planning method and the present application;

[0055] Fig.14 This is a flow chart of Embodiment 2 of the multi-agent collaborative search path planning method based on timing constraints in this application;

[0056] Fig.15 The specific embodiment is based on the simulation results of the multi-agent collaborative search path planning method in this application. DETAILED DESCRIPTION

[0057] Hereinafter, the present application will be further described based on preferred embodiments with reference to the accompanying drawings.

[0058] The vocabulary in this specification is used to illustrate the embodiments of the present application, but is not intended to limit the present application. Unless otherwise clearly specified and limited, the terms "disposed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, a direct connection, an indirect connection through an intermediate medium, or a communication between the two components. For those skilled in the art, the specific meanings of the above terms in this application can be specifically understood.

[0059] In addition, in the description of the embodiments of the present application, various components on the drawings are enlarged or reduced in size for ease of understanding, but this practice is not intended to limit the scope of protection of the present application.

[0060] Figure 1 This is an existing multi-agent collaborative search path planning method. This method assigns the search target points to each agent one by one and in sequence. Figure 1 Describe the method, Figure 1 The target points to be searched in the algorithm include target point 1, target point 2, target point 3, target point 4 and target point 5, and the agents include agent X and agent Y. The existing method will assign target point 1, target point 3 and target point 5 to agent X in sequence, and assign target point 2 and target point 4 to agent Y in sequence. The time for agent X to complete the search task of target point 1 is 60 minutes, and the time for agent Y to complete the search task of target point 2 is 20 minutes. The time for agent X to complete the search task of target point 3 starting from target point 1 is 30 minutes, and the time for agent Y to complete the search task of target point 4 starting from target point 2 is 30 minutes. The time for agent X to complete the search task of target point 5 starting from target point 3 is 45 minutes. Figure 1 As shown: This method makes the total global task time of this time 135 minutes ((60 minutes + 30 minutes + 45 minutes) > (20 minutes + 30 minutes)), that is, the task time of agent X is the sum of the task time of target point 1, target point 3 and target point 5, which is 135 minutes, and the task time of agent Y is the task time of target point 2 and target point 4, which is 50 minutes.

[0061] by Figure 1 Taking the analysis as an example, it can be seen that during the last 10 minutes of the completion of target point 1 by agent X and the entire search process of target points 3 and 5, agent Y was in an idle state. For the entire search process, the search ability and efficiency of agent Y were greatly wasted. In addition, in the process of path planning, it was simply allocated according to the order of the target points, and the impact of the search time on the continuous operation of the agent was not considered. It can be seen that this method has the problems of low search efficiency and long total time of the global task.

[0062] based on Figure 1 The problem of the multi-agent collaborative search path method in the present invention is that the present invention provides a multi-agent collaborative search path planning method based on timing constraints. The multi-agent collaborative search path planning method based on timing constraints in the present invention will be described in detail below. Figure 2 As shown, the method comprises the following steps:

[0063] S100: Initializing an agent set and a target point set P to be searched, wherein the agent set includes at least two agents, and the target point set P to be searched includes at least two target points to be searched;

[0064] Where P = {P 1 ,P 2 ,···,P Q}, Q is the maximum number of target points to be searched at the beginning of the search task, Q ≥ 2.

[0065] Specifically, Figure 3 As an example to illustrate the steps of the search path planning method in this application, the agent set includes two agents, agent X and agent Y, and the target point set to be searched P = {P 1 , P 2 , P 3 , P 4 , P 5}, the number of target points and agents set is the same as the prior art described above, to facilitate comparative analysis with the prior art.

[0066] In some preferred embodiments of the present application, the collaborative search path planning method can be applied to a large area where there are a large number of discretely distributed suspicious, dangerous, etc. areas that need to be searched. In view of the problem that the importance of target points in the large area is different, in order to ensure efficient search of different target points, in step S100, a weight set W of the target points to be searched is also initialized, and the weights of the weight set W of the target points to be searched correspond to the target points to be searched in the target point set P to be searched, and the weight of each target point is set based on its importance;

[0067] Where W = {W 1 ,W 2 ,···,W Q}, the weight value is between 1 and 10, and the larger the value, the more important the target point to be searched is.

[0068] Specifically, Figure 3 For example, the weight set W of the target point to be searched includes the target point to be searched P 1 , P 2 , P 3 , P 4 , P 5 One-to-one weight W1 , W 2 , W 3 , W 4 , W 5 , which corresponds to the importance of the target point to be searched.

[0069] S200: Start the first search phase, the global timer Timer starts timing and performs the candidate target point search operation on all agents in turn;

[0070] Furthermore, when the search task starts, the global timer Timer starts timing, and the clock frequency of each agent itself is synchronized with the global timer Timer to ensure the stability of operation and the efficiency of data processing.

[0071] Specifically, for any agent, the search operation for candidate target points includes:

[0072] The current position and estimated departure time of the agent are obtained, wherein the current position is the starting position of the agent's task or the position of its most recently marked target point.

[0073] Based on the current position of the agent, under the constraints of the task constraints, the agent's alternative target point is determined through the target point optimization function search, and the time when the agent arrives at its alternative target point is determined based on the estimated departure time.

[0074] by Figure 3 and Figure 4 For example, the five-pointed star in the figure represents the starting position of the task, the triangle represents the candidate target point, the circle represents the determined target point, and the square represents the target point to be searched in the set of target points to be searched that has not been determined as the candidate target point. That is, the points in the set of target points to be searched are the triangles and squares in the figure, the blue line represents agent X, and the red line represents agent Y. After the candidate target point is marked as the determined target point, the blue dotted line or red dotted line will become a blue solid line or a red solid line.

[0075] in, Figure 3 The schematic diagram of path planning at the beginning of the search task (T0) shows that agents X and Y determine the target point P to be searched based on the task constraints and the target point optimization function. 1 and P 2 is the candidate target point. At this time, the current positions of Agent X and Agent Y are the starting positions of the task, and the estimated departure time is the time when the global timer Timer is 0; Figure 4 The schematic diagram of path planning during the search task (T1, T1>T0) is shown. 2 Marked as the target point, agent Y based on P 2 And determine P through task constraints and target optimization function search3 is the candidate target point. At this time, the current position of agent Y is P 2 , the estimated departure time is when agent Y completes the current position (P 2 ) of the search task, based on the estimated departure time and P 2 With P 3 The distance between them determines the agent Y's arrival at the alternative target point P 3 It can be seen from this specific embodiment that the candidate target point, the determined target point, the current position, the estimated departure time, the target point to be searched, etc. are all variables and are not constant at different times during the search process.

[0076] In some specific embodiments of the present application, the task constraints include the following constraints:

[0077] a. Each target point to be searched can only be marked as a determined target point of one agent;

[0078] b. Agents that need to update global information during the search can communicate with other agents directly or indirectly through other agents;

[0079] c. Collisions between agents are not allowed at any time.

[0080] In some specific embodiments of the present application, the target optimization function is calculated by the following formula:

[0081]

[0082] Among them, α is an adjustable parameter, and its value range is 0 to 1, which reflects the importance of time consumption and search target point weight in the search task. W is the time it takes for the agent to search from the current target point to the i-th target point to be searched, i is the weight corresponding to the search target point i, k is the kth agent, and C is the number of agents.

[0083] In some preferred embodiments of the present application, the value range of the adjustable parameter α is 0.5 to 0.9. Figure 5 and Figure 6 The target search coverage rate η for different numbers of target points to be searched is shown respectively P With the target weight coverage η w The curve comparison chart with the change of adjustable parameter α shows that a larger target search coverage η can be obtained when the α value is 0.5-0.9 P and target weight coverage η w .

[0084] Among them, the target search coverage η Pis the ratio of the number of searched targets to the total number of targets, and the target weight coverage η w It is the ratio of the sum of the weights of the searched targets to the sum of the weights of all targets, that is:

[0085]

[0086] Among them, X i Represents the target point P i Has it been searched? If it has been searched, then X i =1, otherwise X i =0.

[0087] S300: Continuously slide forward a time window T as the global timer Timer is counting wl , when the candidate target point of any agent enters the time window, that is, the time for any agent to reach its candidate target point is between (Timer, Timer+T wl ) interval, perform the planning path update operation on the agent;

[0088] Specifically, Figure 7 As shown, the planning path update operation includes the following steps:

[0089] C100: Determine whether any agent has reached its candidate target point and entered the time window. If yes, execute steps C200 to C400. If no, execute step C100 again after a delay of time step Δt.

[0090] C200: Mark the candidate target point as the determined target point of the agent, and extend the planned path of the agent from the last marked determined target point or the task starting position to the determined target point;

[0091] C300: taking the determined target point as the current position of the agent and updating the estimated departure time of the agent; that is, taking the arrival time of the agent from the last marked determined target point or the task starting position to the determined target point as the estimated departure time of the agent from the current position (the determined target point);

[0092] C400: Delete the determined target point from the set P of target points to be searched.

[0093] by Figure 3 , Figure 4 as well as Fig.11 For example, Figure 3 , Figure 4 They are the path planning schematic diagrams at different times. Fig.11 The figure shows the process of planning different target points as search paths for different agents during the sliding process of the global timer Timer. Fig.11 The two moments T0 and T1 in Figure 3 , Figure 4 Correspondingly, when the target point to be searched is a candidate target point, it is connected to the task starting position or the last marked determined target point by a dotted line. When the candidate target point is marked as the determined target point of the corresponding agent, the dotted line between it and the task starting position or the last marked determined target point becomes a solid line, and the solid lines are connected to represent the search path of the corresponding agent. In this embodiment, the initialized set of target points to be searched P = {P 1 , P 2 , P 3 , P 4 , P 5}.

[0094] Specifically, see Figure 3 and Fig.11 Explain the path planning at time T0 (when the search task starts). At the beginning of the search task, Agent X and Agent Y determine P based on the task constraints and the target optimization function. 1 , P 2 As the time window slides, it moves to the right on the horizontal axis, such as Fig.11 As shown, at time T1, the P of agent Y 2 The moment of entering the time window is Figure 4 As shown, P 2 The candidate target point is marked as the determined target point of agent Y, and the planned path of agent Y extends from the starting position of the task (indicated by the letter O) to P 2 , the search path of agent Y is updated to OP 2 , the current position of agent Y becomes P 2 , and will complete P 2 The search time is taken as the 2 The estimated departure time of departure, P 2 Delete from the target point set P to be searched, and the target point set P to be searched becomes {P 1 , P 3 , P 4 , P 5}, and agent X's P 1 If the time window does not enter the time window at the moment T2, the time window is extended by the time step Δt until it reaches the moment T2.

[0095] In some preferred embodiments of the present application, in order to prevent multiple agents from competing and causing confusion in the time sliding window at the same time, it is necessary to determine the internal priorities of multiple agents before the search task begins to ensure the efficiency and order of the search task. Furthermore, the search operation for candidate target points is performed in sequence based on the priority of each agent.

[0096] In some specific embodiments of the present application, the time window T wl The width of the time window T is no greater than the delay time step Δt, and the time window T wl The length of is small, which avoids the situation where two candidate target points of an agent appear at the same time in the same time window. And based on priority considerations, if the time window t wl If the width is greater than the delay time step Δt, it may cause the high-priority agent to appear in the time sliding window several times in a row, causing the low-priority agent to lose the opportunity to select the target point to be searched.

[0097] In some preferred embodiments of the present application, when there is a weight set W of target points to be searched, the weight corresponding to the determined target point needs to be deleted from the weight set W of target points to be searched.

[0098] S400: re-execute the candidate target point search operation for all agents in sequence, and then return to execute step S300 until the set of target points to be searched P is an empty set or the global timer Timer reaches the end time of the first search phase.

[0099] Among them, if the set of target points to be searched P is an empty set, it means that all the target points to be searched have been searched.

[0100] The following will be combined Figure 3 , Figure 4 , Figure 8-Figure 11 The following describes the loop execution process of this step and step S300. Fig.11 At different moments in Figure 3 , Figure 4 , Figure 8 , Fig. 9 , Fig.10 Corresponding. Figure 4 and Fig.11 As shown, at time T1, P 2 After being deleted from the target point set P to be searched, based on the current position P of the agent Y 2 and the estimated departure time, and through the task constraints and the target optimization function, select the target point set to be searched {P 1 , P 3 , P 4 , P 5}Determine P 3 is the candidate target point of agent Y, and calculates the time to reach P 3 Estimated time of arrival; Figure 8 and Fig.11 As shown, at time T2, when agent X's P 1 The moment and P of agent Y 3 At the same time, entering the time window, P 1The candidate target point is marked as the determined target point of agent X, and the planned path of agent X extends from the starting position of the task to P 1 , the search path of agent X is updated to OP 1 , the current position of agent X becomes P 1 , P 3 The candidate target point is marked as the determined target point of agent Y, and the planning path of agent Y starts from the last marked determined target point P 2 Extend to P 3 , the search path of agent Y is updated to OP 2 -P 3 , the current position of agent Y becomes P 3 , and will complete P 1 , P 3 The search time is respectively as 1 , P 3 The estimated departure time of departure, P 1 , P 3 Delete from the set of target points to be searched P, based on the current position and estimated departure time of agents X and Y, and through task constraints and target optimization function, select from the set of target points to be searched {P 4 , P 5} respectively determine P 5 , P 4 is the candidate target point of agent X and Y, and calculates the arrival point P 5 , P 4 Estimated time of arrival; Fig. 9 and Fig.11 As shown, at time T3, the specific execution process will not be described in detail, and the execution process at times T0, T1 and T2 can be referred to. The result at this time is to convert P 4 Marked as the determined target point of agent Y, the search path of agent Y is updated to OP 2 -P 3 -P 4 , wait for the search target point set P to become {P 5}; At time T4, the result at this time is to convert P 5 Marked as the determined target point of agent X, the search path of agent X is updated to OP 1 -P 5 , wait for the search target point set P to become an empty set. From this we can see that Fig.11 The specific embodiment in the embodiment is the case where all the target points to be searched have been searched and the first search stage is terminated.

[0101] In some preferred embodiments of the present application, there are objective facts that there are many target points to be searched and the search situation is complex, and multiple intelligent agents need to act on a search task at the same time. There must be performance differences between the multiple intelligent agents. The following considerations are based on the endurance time and search speed between different types of intelligent agents to achieve coordinated cooperation between different types of intelligent agents.

[0102] Specifically, the intelligent agent includes at least one first-class intelligent agent and at least one second-class intelligent agent, the search speed of the first-class intelligent agent is greater than the search speed of the second-class intelligent agent and the endurance of the first-class intelligent agent is less than the endurance of the second-class intelligent agent. Based on the respective characteristics of the first-class intelligent agent and the second-class intelligent agent, the search path of each intelligent agent is collaboratively planned at different stages of the search task to maximize the overall search efficiency.

[0103] Furthermore, in step S100, a first type of agent set A and a second type of agent set B are initialized based on the first type of agent and the second type of agent, respectively. 1 ,A 2 ,A 3 ,···,A M}, M is the number of the first type of agents, M≥1, the second type of agent set B={B 1 ,B 2 ,B 3 ,···,B N}, N is the number of the second type of agents, N≥1. In some specific embodiments of the present application, the first type of agent is a drone, and the second type of agent is an unmanned boat.

[0104] Furthermore, since the first type of intelligent agent has a fast search speed but a short endurance time, and the second type of intelligent agent has a long endurance time but a slow search speed, it is necessary to ensure that the first type of intelligent agent performs the search task within the endurance time and meets the task constraints, so as to increase the number and weight of the searched target points within the maximum endurance time.

[0105] Based on this reason, the search task also includes a second search phase. In the first search phase, the endurance of the first and second intelligent agents is relatively sufficient, and both the first and second intelligent agents perform the search task. If the set of target points to be searched P becomes an empty set, that is, all the target points to be searched have been searched, then this search task only has the first search phase; if the set of target points to be searched P is not an empty set, that is, not all the target points to be searched have been searched, when at least one of the first intelligent agents cannot determine the candidate target point for the first time, the first search phase ends, and the first and second intelligent agents enter the second search phase.

[0106] Furthermore, if Fig.14As shown, after step S400, the following steps are also included:

[0107] S500: starting the second search phase, re-performing the candidate target point search operation on all first-category agents in turn, performing the parking operation on all second-category agents and obtaining their position information;

[0108] That is, the search task in this embodiment also includes a second search phase, which is started after the first search phase is terminated. In the second search phase, the first type of intelligent agent continues to perform the search task within the maximum endurance time, and the second type of intelligent agent is parked at the last determined target point in the first search phase to wait for the first type of intelligent agent that needs to return. And the search operation of the alternative target point here is basically the same as the search operation of the alternative target point in the previous text, the only difference is that the object of the search operation of the alternative target point here is the first type of intelligent agent.

[0109] S600: Continuously slide forward a time window T as the global timer Timer is counting wl , when the candidate target point of any first-class agent enters the time window, that is, the time for any first-class agent to reach its candidate target point is (Timer, Timer+T wl ) interval, a planning path update operation is performed on the first type of intelligent agent; the planning path update operation here is basically the same as the planning path update operation in the previous text, the only difference is that the object of the planning path update operation here is the first type of intelligent agent.

[0110] S700: re-execute the candidate target point search operation for all first-category agents in sequence, and then return to execute step S600 until the set of target points to be searched P is an empty set or the global timer Timer reaches the end time of the second search phase.

[0111] The following will describe the situation where the global timer Timer reaches the end time of the first search phase and the search task is completed.

[0112] In some preferred embodiments of the present application, during the execution of the search task, the return time of the intelligent agent must also be considered, and the return time is considered as a constraint condition of the first search stage to ensure that the intelligent agent can return smoothly while performing the search task.

[0113] Specifically, the task constraints also include a return time constraint. When the search task is in the first search phase, the return constraint is that the time it takes for the first type of agent to return from its alternative target point to the task start position is less than its endurance time, that is, it is necessary to ensure that it can smoothly return to the task start position after the alternative target point is completed. The return position of all agents in this application is the task start position. In actual applications, the return position can be set based on actual needs, and whether the return constraint is met is evaluated based on the specific return position.

[0114] Furthermore, when the set of target points to be searched P is not an empty set, the first search phase terminates when at least one first-category agent fails to search and determine the candidate target points for the first time. That is, when the search of the target points to be searched has not been completed, at least one first-category agent does not meet the return time constraint and / or the above constraints a, b, c, so that the agent does not meet the conditions for continuing to perform the search task and cannot determine the candidate target points for the search. At this time, the first search phase terminates. Specifically, taking 9 as an example, Fig. 9 Agent X from P 1 To P 5 The time is 35 minutes. Assuming that agent Y completes P 4 After searching, the agent X can return normally. 5 The time to return to the starting point of the task is 80 minutes, and the endurance of agent X is 90 minutes. Obviously, 90 minutes < (35 minutes + 80 minutes), that is, agent X completes P 5 After searching, it is impossible to return. Although the search target points have not been completely searched, the first search stage is still terminated. This application realizes the termination of the first search stage through the restriction of constraints, taking into account the actual operation of the intelligent agent, so that the intelligent agent can complete the search task in the first search stage with the greatest efficiency.

[0115] Furthermore, when in the second search phase, the return constraint condition also needs to be considered. The return time constraint condition is that the movement time of the first-class agent from its candidate target point to the second-class agent closest to it is less than the endurance time, that is, it is necessary to ensure that the first-class agent can smoothly land on the nearest second-class agent after completing the search for the candidate target point. When the endurance time of the first-class agent is not enough to support it to complete the candidate target point and reach the nearest second-class agent, the first-class agent does not meet the return time constraint condition. When it is unable to search and determine the candidate target point, it needs to move to the second-class agent closest to it. The search task of the first-class agent is completed, and it lands on the nearest second-class agent to wait for return.

[0116] Furthermore, when the set P of target points to be searched is not an empty set, when all first-category agents are unable to search and determine the candidate target points, the second search phase terminates. That is, when the search for the target points to be searched is not completed, all first-category agents do not meet the return time constraint and / or the above-mentioned constraints a, b, c, that is, all first-category agents do not meet the conditions for continuing to perform the search task and cannot determine the candidate target points for the search, and the second search phase terminates at this time. The present application realizes the termination of the second search phase by limiting the constraints, taking into account the actual operation of the agents, so that the first-category agents can complete the search task in the second search phase with the greatest efficiency.

[0117] Combine the following Figure 1 , Fig.12 and- Fig.13 Agent X and Agent Y are shown searching for P 1 , P 2 , P 3 , P 4 , P 5 Take an example to illustrate the effect of this application. Figure 1 , Fig.12 and Fig.13 As shown by Figure 1 , Fig.12 and Fig.13 It can be seen that the present application can save 40 minutes compared with the existing method. The effect of shortening the total time is more obvious in the case of a large range and many target points. The method in the present application can significantly shorten the total time of the global task and significantly improve the search efficiency, thereby making the total time saving effect of the global task more significant when performing a large-scale search task, and the search efficiency will be greatly improved. In summary, the present application sets a time sliding window so that the alternative target point that slides into the time window is used as the path to determine the target point and add the corresponding intelligent agent, and selects the corresponding intelligent agent to perform the search task based on the time window, thereby avoiding the situation where a certain intelligent agent falls into the local optimal search and causes low global search efficiency, thereby improving the search efficiency and ensuring the total time of the global task. [Specific embodiment]

[0119] In some specific embodiments of the present application, in order to calculate the time cost involved in the above target optimization function, it is necessary to obtain the following variables, and the following are the definitions of the variables:

[0120] Variable 1: The agent’s actual navigation path d ij

[0121] The actual navigation path of the agent d ij The agent moves from the target point P i To the target point P j The Euclidean distance between them is calculated by the following formula:

[0122]

[0123] Among them, x i 、x j The target points P are i , P j The horizontal axis, y i ,y j The target points P are i , P j The vertical coordinate of .

[0124] Variable 2: Steering angle θ ijp

[0125] Steering angle θ ijp The agent moves from the target point P i Navigate to the target point P j Then sail to P p The steering angle is calculated based on the cosine theorem using the following formula:

[0126]

[0127] in, From the target point P i To the target point P j The vector of From the target point P j To the target point P p vector, steering angle θ ijp The range of is [0,π].

[0128] Variable 3: Search time T ijp

[0129] Search time T ij The intelligent agent determines the target point as P i From the target point P j Navigate to the target point P p The time spent is calculated based on formula (4) and formula (5) by the following formula:

[0130]

[0131] Among them, V and ω are the linear velocity of the agent when sailing in a straight line and the angular velocity when turning, respectively.

[0132] Variable 4: The current total search time T of the kth agent a (k)

[0133] Current total search time T a (k) is calculated based on formula (6) by the following formula:

[0134]

[0135] Where C is the number of agents, Indicates whether the kth agent is from P j Sail to P p , in the planning path of the kth agent, if there is a j Sail to P p ,but otherwise, That is, when the candidate target point P p If the time window is not entered or the sum of the current search time and the return time is greater than the corresponding agent's endurance, When the candidate target point P p When the time window is entered or the sum of the current search time and the return time is less than the corresponding agent's endurance,

[0136] Variable 5: Return time T of the kth first-class agent r (k)

[0137] When the kth first-class agent returns, the specific return point position of the first-class agent is first calculated based on the position of the second-class agent that chooses to land, and then its return time T is calculated according to formula (6): r (k).

[0138] The calculation process of the return point position is as follows: Assume that the last determined target point position before the first type of intelligent agent returns is P A (x A ,y A ), the position of the second type of agent is P S (x S ,y S ), then the return point P L (x L ,y L ) is calculated as follows:

[0139]

[0140] Among them, V 1 is the linear velocity of the first type of agent returning home, V 2 is the linear velocity of the second type of agent.

[0141] Furthermore, when the search task is in the second search phase, the first type of agent can continue to perform the search task only if the return time constraint is met, that is,

[0142]

[0143] Among them, T max (k) represents the maximum flight time of the kth first-class agent. That is, when the search task is in the second search phase, if the first-class agent meets the return time constraint and the set of target points to be searched P is not empty, the first-class agent will continue to perform the search task, otherwise it will return with the corresponding second-class agent.

[0144] Variable 6: Waiting and operation time T before the kth agent sets out o (k)

[0145] Before departure, both the first and second types of agents need operators to perform pre-departure inspection and preparation operations. Assume that the inspection and preparation operation time of the operator is T s , then the waiting and operation time before the kth agent sets out is T o (k) is:

[0146]

[0147] Variable 7: Total task time T of the kth agent c (k)

[0148] The total task time T of the kth agent c (k) is the waiting time and operation time before departure T o (k) The sum of the total search time and the return time. If the agent is a second-class agent, there is no return time, that is:

[0149]

[0150] Variable 8: Total time of global task T

[0151] The global total task time T is the maximum total task time of all first-class agents, that is:

[0152] T=max{T c (1),T c (2),…,T c (M)} (13);

[0153] Fig.15It is shown that in a specific embodiment, in order to verify the search efficiency of the multi-agent collaborative search path planning method based on timing constraints in this application, this embodiment is based on the search target point in a 30km*30km sea area. Experiments are conducted, 100 target points to be searched are randomly generated in the target sea area, and the collaborative search task is completed by two types of agents, unmanned aerial vehicles (UAVs) and unmanned boats (USVs). To simplify the problem, we ignore the process of UAVs reaching the specified navigation altitude from takeoff, and focus on the collaborative search path planning process between UAVs and USVs. When determining the number of UAVs and USVs for collaborative search, it is necessary to consider factors such as the requirements of the task itself and the maximum search range that a single VSV and UAV can cover. After comprehensive consideration, we chose 3 USVs and three UAVs to form a collaborative search system in this experiment. The target optimization parameter α is set to 0.7, and the relevant parameter information of USVs and UAVs is shown in Table 1.

[0154] Table 1 USV and UAV status parameter information

[0155]

[0156] like Fig.15 As shown, the multi-agent collaborative search path planning method in this application plans appropriate search paths for three USVs and three UAVs. After starting from the base, the UAVs fly in three different directions respectively. Based on the above variables 1-variable 8 and according to the optimization of formula (1), the optimal candidate target point is selected each time, and the candidate target point is transformed into the determined target point based on the time window and the planned path is updated. The UAV completes the search of most of the target points to be searched within the endurance time; due to the slow navigation speed of the USV, it can only search for fewer target points to be searched. After the end of the first search phase, the position is conducive to the return of the UAV with insufficient endurance time. Compared with the return of the VAV to the shore base, the nearby reception of the USV greatly reduces the time spent on the return of the VAV and improves the overall search efficiency.

[0157] The above is a detailed introduction to the specific implementation methods of the present application. For those skilled in the art, several improvements and modifications may be made to the present application without departing from the principles of the present application. These improvements and modifications also fall within the scope of protection of the claims of the present application.

Claims

1. A multi-agent collaborative search path planning method based on timing constraints, characterized by: The following steps are involved: S100: Initializing an agent set and a target point set to be searched, wherein the agent set includes at least two agents, and the target point set to be searched includes at least two target points to be searched; S200: Start the first search phase, the global timer starts timing and performs the candidate target point search operation on all agents in turn; S300: continuously sliding a time window forward as the global timer counts, and when the candidate target point of any agent enters the time window, performing a planning path update operation on the agent; S400: re-execute the candidate target point search operation for all agents in sequence, and then return to execute step S300 until the set of target points to be searched is an empty set or the global timer reaches the end time of the first search phase.

2. The multi-agent collaborative search path planning method based on timing constraints according to claim 1 is characterized in that: For any agent, the candidate target point search operation includes: Obtain the current position and estimated departure time of the agent, wherein the current position is the task starting position of the agent or the position of the most recently marked target point; Based on the current position of the agent and under the constraints of the task constraints, the candidate target point of the agent is determined by searching the target point optimization function, and the time when the agent arrives at its candidate target point is determined based on the estimated departure time.

3. According to the method of multi-agent collaborative search path planning based on timing constraints in claim 2, it is characterized in that: The planned path update operation includes the following steps: C100: Determine whether any agent reaches the candidate target point and enters the time window. If yes, execute steps C200 to C400. If no, execute step C100 again after delaying the time step Δt. C200: Mark the candidate target point as the determined target point of the agent, and extend the planned path of the agent from the last marked determined target point or the task starting position to the determined target point; C300: taking the determined target point as the current position of the agent and updating the estimated departure time of the agent; C400: Delete the determined target point from the set of target points to be searched.

4. According to the method of multi-agent collaborative search path planning based on timing constraints in claim 2, it is characterized in that: The intelligent agents include at least one first type of intelligent agent and at least one second type of intelligent agent; The search speed of the first type of intelligent agent is greater than the search speed of the second type of intelligent agent and the battery life of the first type of intelligent agent is less than the battery life of the second type of intelligent agent.

5. According to the method of multi-agent collaborative search path planning based on timing constraints in claim 4, it is characterized in that: After step S400, the following steps are also included: S500: starting the second search phase, re-performing the candidate target point search operation on all first-category agents in turn, performing the parking operation on all second-category agents and obtaining their position information; S600: continuously sliding a time window forward as the global timer counts, and when the candidate target point of any first-category agent enters the time window, performing a planning path update operation on the first-category agent; S700: re-execute the candidate target point search operation for all first-category agents in sequence, and then return to execute step S600 until the set of target points to be searched is an empty set or the global timer reaches the end time of the second search phase.

6. The multi-agent collaborative search path planning method based on timing constraints according to claim 5 is characterized in that: The mission constraints include return time constraints; When in the first search phase, the return time constraint condition is that the time it takes for the agent to return from its candidate target point to the starting position of the task is less than its endurance time; When in the second search phase, the return time constraint condition is that the movement time of the first type of agent from its candidate target point to the second type of agent closest to it is less than its endurance time.

7. The multi-agent collaborative search path planning method based on timing constraints according to claim 6 is characterized in that: When the set of target points to be searched is not an empty set, when at least one first-type agent fails to search and determine the candidate target point for the first time, the first search phase is terminated; When the set of target points to be searched is not an empty set and all first-type agents are unable to search and determine candidate target points, the second search phase terminates.

8. The multi-agent collaborative search path planning method based on timing constraints according to claim 6 is characterized in that: Step S700 also includes the following steps: When any first-category intelligent agent is unable to search and determine an alternative target point when the set of target points to be searched is not an empty set, it moves to the second-category intelligent agent closest to it.

9. The multi-agent collaborative search path planning method based on timing constraints according to claim 6 is characterized in that: The task constraints also include the following constraints: a. Each target point to be searched can only be marked as a determined target point of one agent; b. The agent that needs to update global information during the search can communicate with other agents directly or indirectly via other agents; c. Collisions between agents are not allowed at any time.

10. The multi-agent collaborative search path planning method based on timing constraints according to claim 1 is characterized in that: The candidate target point search operation is performed in sequence based on the search priority of each of the agents.

Images