Hierarchical asynchronous time constraint processing method for multi-agent explorer task planning

Abstract

The application discloses a layered asynchronous time constraint processing method for multi-agent detector task planning and belongs to the technical field of aerospace. The application is realized by the following method: a multi-node detector time planning problem model is established, a multi-agent simple time network equivalent expression is performed on the time constraint problem, a layered expression is performed on the multi-agent simple time network, and the agent calculation sequence of asynchronous processing is solved; according to the agent calculation sequence, constraint processing is performed on the current agent time constraint network based on an arc consistency algorithm, so that whether the current planning satisfies the time constraint consistency is judged. Through the layered asynchronous arc consistency time constraint processing of the multi-node detector, the unnecessary constraint detection is reduced by increasing the variable value domain update flag, the constraint detection number of a single agent is reduced, the constraint detection efficiency is improved, the fast processing of the complex time constraint in the autonomous planning process of the multi-node detector is realized, and the multi-agent detector task planning efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to a layered asynchronous time constraint processing method suitable for multi-agent probe task planning, belonging to the field of aerospace technology. BACKGROUND

[0002] Small celestial body exploration is of great significance for the study of the origin of the solar system. In recent years, NASA, JAXA and ESA have implemented many small celestial body exploration missions, and the exploration forms have developed from flyby exploration to in-situ exploration, sample return and other forms. Aiming at the two main problems of small celestial body landing at present: (1) the gravity of small celestial body is very weak and irregular, and the probe is easy to bounce and escape, (2) the surface soil properties are not completely known. The ground features are complex and diverse, with many gravel, grooves, etc., and the use of traditional rigid attachment strategy is easy to cause the probe to roll over and overturn. Therefore, some scholars propose a flexible multi-node probe, which contains multiple nodes connected by flexible mechanisms between nodes. Through the flexible mechanism to absorb the kinetic energy of the probe at the moment of landing, the purpose of suppressing landing rebound and improving landing stability is achieved.

[0003] Each node of the multi-node probe has independent computing power, and each node is regarded as an agent. Considering the long time delay of communication between small celestial bodies and the earth, in order to realize autonomous stable attachment on the surface of small celestial bodies, the multi-node probe needs to have autonomous task planning capability. In the planning process, most of them use simple time network (STN) to quantitatively represent time variables and time constraints, and on this basis, time constraint processing is carried out to judge whether the current time constraint meets the consistency requirement. The current time constraint processing method mainly includes path consistency and arc consistency. The most typical path consistency method is Floyd-Warshall algorithm, which calculates the shortest path of the whole network based on STN, and the time complexity is O(n 3 ), this method has many constraint detection numbers, and the calculation efficiency is low. Therefore, Bliek proposed a partial path consistency algorithm (PPC) which effectively reduces the constraint detection number. The ΔSTP algorithm combines PPC and the triangulation of STN graph to improve the calculation efficiency at the cost of storage resources, and the time complexity is O(t 2 ). 3The C algorithm enforces partial path consistency on the triangulated STN combined with the directed path consistency (PPC), and only needs to perform constraint processing twice for each triangle in the network, further improving the efficiency of time constraint consistency processing, and the time complexity is O(t). The arc consistency method tightens the time variable value domain through constraint propagation to determine whether the problem has a solution under the current constraint, and the main methods include AC-1 to AC-7, AC-2000, AC-2001, ACSTP and the like. The above time constraint processing methods are the core of the distributed processing method and its variants in the multi-agent field, such as DΔDPC, DΔPPC, DisACSTP and the like. In addition to improving the efficiency of time constraint processing in the multi-agent field, the privacy of the agent and the communication cost between the agents also need to be concerned. The path consistency based algorithm may increase new constraints in the calculation process, thereby increasing the communication overhead, and even causing the agents that do not need to communicate to communicate, which is not conducive to the privacy of the agent. In order to solve the problem, the distributed algorithm based on arc consistency effectively avoids the defect, but it does not judge the constraint propagation in the calculation process, and many invalid constraint detections are performed, causing the waste of computing resources and communication resources. The small celestial body multi-node probe has limited computing resources, and the landing task has high real-time requirements, so under the premise of meeting the privacy of the agent, improving the efficiency of time constraint processing in the planning process of the multi-node probe and reducing the communication information between the nodes is a complex problem. SUMMARY

[0004] The main purpose of the present application is to provide a hierarchical asynchronous time constraint processing method for multi-agent probe task planning, by establishing a multi-node probe time planning problem model, performing multi-agent simple time network equivalent expression on the time constraint problem, performing hierarchical representation on the multi-agent simple time network, solving the asynchronous processing agent calculation sequence, and then processing the current agent time constraint network based on the arc consistency algorithm on the basis of the sequence, reducing unnecessary constraint detection by increasing the variable value domain update flag, thereby improving the constraint detection efficiency, so as to realize the rapid processing of complex time constraints in the autonomous planning process of the multi-node probe, and improve the planning efficiency.

[0005] The purpose of the present application is realized through the following technical solutions.

[0006] The hierarchical asynchronous time constraint processing method for multi-agent explorer task planning disclosed by the application firstly establishes a multi-node explorer time planning problem model, including node ID, node state variable, initial state, target state, executable activity variable and activity constraint, etc. Before time constraint processing, the activity time variable and time constraint are expressed as variable points and edges in a multi-agent simple time network MaSTN, and then an upper agent network is defined according to the shared variable relationship between agents, and the agent calculation order in asynchronous processing is solved according to the out-degree of the network agent and the number of shared variables of the agent. Based on the arc consistency algorithm, the current agent time constraint network is processed based on the order, unnecessary constraint detection is reduced by adding a variable value domain update flag, whether the activity time variable exists in the value domain is solved, and whether the time constraint is consistent in the planning process is judged according to the processing result. Since the constraint processing process reduces unnecessary constraint detection, the constraint detection efficiency can be improved, thereby realizing the rapid processing of complex time constraints in the autonomous planning process of the multi-node explorer and improving the planning efficiency.

[0007] The hierarchical asynchronous time constraint processing method for multi-agent explorer task planning disclosed by the application comprises the following steps:

[0008] Step one, a multi-node explorer time planning problem model is established, and the multi-node explorer time planning problem is defined as a six-tuple. The multi-node explorer time planning problem model includes node ID, node state variable, initial state, target state, executable activity time variable and activity time constraint.

[0009] The multi-node explorer time planning problem is defined as the following six-tuple

[0010] Π=[P,ST,O,C,s0,g] (1)

[0011] Wherein, P={a1,...,a m} represents a set of explorer node IDs, represents a set of state variables of m explorer nodes, represents a set of executable activity time variables of m explorer nodes, for any activity of any node has a start time point an end time point a duration that is represents the time constraint relationship between any two activities, s0 represents the initial state of the multi-node explorer, and g represents the target state.

[0012] Step two, the activity time variable and time constraint in step one are expressed as multi-agent simple time network, so that the time constraint processing problem is converted into a graph path problem, in order to solve step three and four.

[0013] Step 2.1: for single node internal activity time variable, time constraint, establish single agent simple time network STN.

[0014] For any two time variables u, w, define the time zero point z, and its value domain is I zu ,I zw If there is a time constraint relationship a ≤ w-u ≤ b, the constraint is expressed as interval I uw =[a, b], if u, w belong to the same activity, I uw represents the internal constraint of the activity, that is, the duration of the activity, if u, w belong to different activities, I uw represents the inter-activity constraint. The single node activity time variable and time constraint are expressed as variable points and edges in STN, that is, S = <V, D, C>, V represents the set of all time variables contained in the node, which includes the start time point, end time point and time zero point, D represents the value domain set of the time variable, which includes the weight of the edge between all variable points and time zero point, C represents the constraint relationship set between time variables, which includes the weight of the edge between variable points.

[0015] Step 2.2: on the basis of the agent simple time network obtained in step 2.1, consider the activity time constraint between nodes, establish multi-agent simple time network MaSTN, so that the time constraint processing problem is converted into a graph path problem, in order to solve the subsequent steps three and four.

[0016] For multi-agent time network, the privacy of agent needs to be considered, and only shared information is exchanged between agents, so the variables and constraints in STN are distinguished. For the simple time network S i of agent i = <V i ,D i ,C i >, if there is a constraint c i ∈C i between any variable v i ∈V i and the variable of other agents, the variable is called shared variable, and is added to the shared variable set V i S If all constraints of variable v i belong to agent i, the variable is called private variable, and is added to the private variable set V i PThe other agent variables not belonging to agent i but having constraints with the variables of agent i are called external variables, and are added to the external variable set V i X The variables contained in agent i are called internal variables, and are added to the internal variable set V i L The variable set V i of agent i is then V i X ∪ V i L = V i X ∪ (V i S ∪ V i P ). Similarly, if a constraint c i ∈ C i has two end-point variables v i ,v i ∈ V X , it is added to the external constraint set If a constraint c i ∈ C i has two end-point variables v i ,v j ∈ V i L , it is added to the internal constraint set

[0017] Step three, abstractly represent the multi-agent simple time network obtained in step two at the agent level, and solve the agent computation order in asynchronous processing according to the out-degree of the agent in the network and the number of shared variables of the agent. Solve the agent hierarchical computation order set order in asynchronous processing, based on the multi-agent simple time network, through asynchronous computation of the agent local constraint network, some agents can use the updated external variable value domain for solving, and the internal variable value domain reduction is closer to the final value domain, thereby reducing the number of single-agent constraint detection.

[0018] Step 3.1: Abstractly represent the multi-agent simple time network obtained in step two at the agent level, and establish an upper agent network P = <A, E> according to the constraint relationship between agents, A = {1, 2,..., n} represents the agent ID set, E = {..., (i, j),...} represents the connection relationship set between agents, (i, j) represents that there is a constraint between agent i and j.

[0019] Step 3.2: Calculate the out-degree od of all agents according to the network P obtained in step 3.1, that is, the number of agents having constraints with agent i.

[0020]

[0021] wherein, if (i,j) E, then E(i)(j) = 1, otherwise E(i)(j) = 0.

[0022] Step 3.3: solving the hierarchical computation order set order of the asynchronous agents, placing the agent with the highest out-degree in the first layer of order and removing its ID from set A, then placing the agents in set A that have constraints with the agent in the first layer of order in the second layer of order and removing their IDs from set A, and so on until set A is empty, thereby obtaining the hierarchical computation order set order of the asynchronous agents. So that in step four, through the asynchronous computation of the agent local constraint network, some agents can use the updated external variable value domain to solve, and their internal variable value domain reduction is closer to the final value domain, thereby reducing the number of single agent constraint detections.

[0023] Step four, according to the order set order obtained in step three, based on the arc consistency algorithm, the time constraint network of agent i is processed, unnecessary constraint detection is reduced by adding variable value domain update flag, and then whether there is value domain of activity time variable is solved according to the variable value domain reduction in step two, whether the time constraint is consistent in the planning process is judged according to the constraint processing result.

[0024] Step 4.1: according to the set order obtained in step three, the information communication between agents. For all variables v i ∈V i L Add update flag sign v = true, only when the shared variable update flag sign v = true, the variable value domain is sent.

[0025] Step 4.1.1: if agent i is in the first layer of order, send the value domain of shared variable to the neighbor agent in this layer, receive the external variable value domain sent by the neighbor agent in this layer and the lower layer and add it to the current time network. Then execute step 4.2, send the updated variable value domain to the lower agent.

[0026] Step 4.1.2: if agent i is not in the first layer of order, send the value domain of shared variable to the neighbor agent in this layer and the upper layer, wait for the updated variable value domain sent by the upper layer, and then execute step 4.3.

[0027] Step 4.2: for all variables r i ∈V i XRecord the variable value range I sent by other intelligent agents. zr and the variable range I′ received in the previous iteration. zr In comparison, if I′ zr =I zr Then update the sign flag. r =false, otherwise no operation is performed. An empty list Q is created for the MaSTN of agent i. i .

[0028] Step 4.3: Reduce the feasible range of time variables through constraint propagation.

[0029] Step 4.3.1: Select the variable u whose range has not been calculated. i ∈V i L Its value range I zu Assigned to I′ zu .

[0030] Step 4.3.2: For variable w i ∈V i , if u i and w i There are constraints between them, and sign w =true, then calculate symbol This indicates summing over the endpoints of two intervals. Example: I zw =[a,b],I wu = [c, d], then Then the result was compared with I. zu Find the intersection until all variables w are found. i All solutions are traversed. If the intersection is empty, it means there is no feasible solution at present, the time constraint processing process ends, and the conclusion of constraint inconsistency is returned.

[0031] Step 4.3.3: Update the value range I in the previous step. zu with I′ zu Compare, if I′ zu =I zu Then update the sign flag. v =false, and set variable u i Add to list Q i , if I′ zu ≠I zu This indicates that the variable's value range has been updated. If variable u i In list Q i In the middle, the variable u will be... i From list Q i Remove from the list; otherwise, no action will be taken.

[0032] Step 4.3.4: return to step 4.3.1 until all variables u i ∈V i L are traversed.

[0033] Step 4.4: when all agents in this round complete the constraint processing, calculate the number of variables in each agent set V i L and the list Q i , if all agents satisfy V i L and the number of variables in Q i are equal, the calculation is completed, and the consistent network is returned, if they are not equal, return to step 4.3 to perform a new round of iteration.

[0034] Further comprising step five: reducing unnecessary constraint detection in the multi-node probe autonomous planning process through the constraint processing of steps one to four, so as to realize the rapid processing of complex time constraints in the multi-node probe autonomous planning process, and improve the planning efficiency.

[0035] Advantages:

[0036] 1. The hierarchical asynchronous time constraint processing method suitable for multi-agent probe task planning disclosed in the present application, by establishing a multi-node probe time planning problem model, including node ID, initial state, target state, executable activity variable and activity constraint, etc. The activity time variable and time constraint are expressed as variable points and edges in the multi-agent simple time network MaSTN before the time constraint processing, and then the upper agent network is defined according to the shared variable relationship between agents, and the agent calculation order in asynchronous processing is solved according to the out-degree of the agents in the network. Further, based on the arc consistency algorithm, the current agent time constraint network is processed based on the order to solve whether the activity time variable exists a value range, and whether the time constraint in the planning process is consistent is judged according to the processing result. Thus, the complex time constraints in the multi-node probe autonomous planning process are rapidly processed, and the planning efficiency is improved.

[0037] 2. The hierarchical asynchronous time constraint processing method suitable for multi-agent probe task planning disclosed in the present application is based on the multi-agent simple time network, and through asynchronous calculation of the agent local constraint network, some agents can use the updated external variable value range for solving, and the internal variable value range reduction is closer to the final result. Since information is only sent to other agents when the variable value range changes, this method reduces the number of communication information between agents, and can effectively solve the problem of waste of communication resources in the existing distributed arc consistency algorithm.

[0038] 3, The layered asynchronous time constraint processing method suitable for multi-agent detector task planning is disclosed, value range update marks are added to time variables in the multi-agent simple time network established in step two, and whether the variable is processed is determined according to true and false values of the update marks in the constraint processing process, the variable value range that is not updated in the last iteration is not calculated, unnecessary constraint detection is reduced, the limited computing resources of the multi-node detector are effectively saved, and the constraint processing efficiency is improved. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 is a flow chart of the layered asynchronous time constraint processing method of multi-agent detector task planning

[0040] Figure 2 is a schematic diagram of a single-agent simple time network.

[0041] Figure 3 is a schematic diagram of a multi-agent simple time network.

[0042] Figure 4 is a constraint relationship diagram of the embodiment adopted by the present application. DETAILED DESCRIPTION

[0043] In order to better illustrate the purposes and advantages of the present application, the content of the application is further illustrated below in combination with the drawings and embodiments.

[0044] In order to verify the feasibility of the method, as shown in the table, a three-node detector is selected as an example, the method of the present application is used to judge the consistency of the time constraints in the planning process, the layered asynchronous fast processing of the time constraints is realized, and the planning efficiency is improved. Figure 4

[0045] As shown in the table, the layered asynchronous time constraint processing method suitable for multi-agent detector task planning disclosed in the embodiment is specifically implemented as follows: Figure 4

[0046] Step one, a multi-node detector time planning problem model is established, and the multi-node detector time planning problem is defined as a six-tuple. The multi-node detector time planning problem model includes node ID, node state variable, initial state, target state, executable activity time variable and activity time constraint.

[0047] The three-node detector time planning problem is defined as the following six-tuple

[0048] Π=[P,ST,O,C,s0,g] (3)

[0049] P={1,2,3} represents a detector node ID set, ST represents a set of state variables of the three nodes of the detector,​​ denotes the set of activity variables executable by the three nodes of the probe, for node 2 with start time point a s , end time point a e , duration d a = [1, 1], i.e. denotes the time constraint relationship between activity 1 and activity 2, s0denotes the initial state of the multi-node probe, g denotes the target state.

[0050] Step two, the activity time variables and time constraints in step one are expressed as a multi-agent simple time network, so that the time constraint processing problem is converted into a graph path problem, so as to facilitate subsequent steps three and four. The activity time variables have variable value ranges.

[0051] Step 2.1: For single-node internal activity time variables and time constraints, a single-agent simple time network STN is established.

[0052] For two time variables a s , a e , define the time zero point z, and their value ranges are I zas , I zae , respectively. There is a time constraint relationship 1≤a e -a s ≤1, so the constraint is expressed as the interval I asbs = [1, 1], a s , a e belongs to the same activity, so I asbs represents the internal constraint of the activity, i.e. the duration of the activity. For variables a s , b s , belong to different activities, then I asbs = [1, ∞] represents the inter-activity constraint. The activity time variables and time constraints of the single node are expressed as variable points and edges in the STN (such as Figure 2 ), i.e. S = <V, D, C>, V denotes the set of all time variables contained in the node, the set of time variables includes the start time point of the activity, the end time point and the time zero point, D denotes the set of value ranges of the time variables, the set of value ranges includes the weights of the edges between the variable points and the time zero point, and C denotes the set of constraint relationships between the time variables, the set includes the weights of the edges between the variable points.

[0053] Step 2.2: On the basis of the agent simple time network obtained in step 2.1, the activity time constraints between nodes are considered, and a multi-agent simple time network MaSTN is established, so that the time constraint processing problem is converted into a graph path problem, so as to facilitate subsequent steps three and four.

[0054] For multi-agent time network, the privacy of agent should be considered, and only shared information is exchanged between agents, so the variables and constraints in STN are distinguished (such as Figure 3 ). Taking agent 2 as an example, for the simple time network S2 of agent 2, variable a s ∈V2 and variable f s ∈V3 of agent 3 have a constraint [0, 0], variable a s is called a shared variable, and is added to the shared variable set All constraints of variable a e belong to agent 2, so the variable is called a private variable, and is added to the private variable set For other agent variables that do not belong to agent 2 but have constraints with the variables of agent 2, they are called external variables (such as variable f s ), and are added to the external variable set All variables contained in agent 2 are called internal variables, and are added to the internal variable set Then the variable set of agent 2 is Similarly, constraint c asfs ∈C2, one end variable is added to the external constraint set Constraint c asae ∈C2, two end variables are added to the internal constraint set

[0055] The time constraint variable and time constraint used in the embodiment of the application are shown in Table 1.

[0056] Table 1 Time constraint variable and time constraint of the embodiment

[0057]

[0058]

[0059] Step three, abstractly representing the multi-agent simple time network obtained in step two at the agent layer, and solving the agent calculation sequence in asynchronous processing according to the agent out-degree of the network and the number of shared variables of the agent. The agent hierarchical calculation sequence set order in asynchronous processing is solved, and based on the multi-agent simple time network, through asynchronous calculation of the local constraint network of the agent, some agents can use the updated external variable value domain for solving, and the internal variable value domain reduction is closer to the final value domain, so that the number of single-agent constraint detection is reduced.

[0060] Step 3.1: Abstract the multi-agent simple time network obtained in step two to agent level, and build the upper agent network P = <A, E> according to the constraint relationship between agents, A = {1, 2, 3} represents the agent ID set, and E = {(1, 2), (2, 1), (2, 3), (3, 2)} represents the connection relationship set between agents.

[0061] Step 3.2: Calculate the out-degree od of all agents according to the network P, that is, the number of agents that have constraints with agent i.

[0062]

[0063] Wherein, if (i, j) E, then E(i)(j) = 1, otherwise E(i)(j) = 0.

[0064] According to the set E, the out-degree of agents 1, 2, and 3 is 1, 2, and 1 respectively.

[0065] Step 3.3: Solve the asynchronous agent hierarchical calculation order set order, place the agent with the highest out-degree, agent 2, in the first layer of order, and remove its ID from the set A, and then place the agents with constraint relationship with the first layer agent, agent 1 and agent 3, in the second layer of order, and remove the IDs of these agents from the set A, at this time the set A is empty set, thereby obtaining the asynchronous agent hierarchical calculation order set order. In order to reduce the number of single agent constraint detection in step four through asynchronous calculation of agent local constraint network, some agents can use the updated external variable value domain for solving, and the internal variable value domain reduction is closer to the final value domain.

[0066] Step four, according to the order set order obtained in step three, constraint processing is performed on the agent i time constraint network based on the arc consistency algorithm, unnecessary constraint detection is reduced by adding variable value domain update flag, and then it is judged whether the time constraint is consistent in the planning process according to the constraint processing result whether the value domain of the activity time variable exists or not by solving the variable value domain reduction in step two.

[0067] Step 4.1: According to the set order obtained in step three, information communication is carried out between agents. For all variables v i ∈V i L Add update flag sign v = true, only when the shared variable update flag sign v = true, the variable value domain is sent.

[0068] Step 4.1.1: Agent 2 is in the order first layer, it sends the value domain of shared variable to its neighbor agent in this layer, since there is no neighbor agent in this layer, it does not send, receives the external variable value domain I zxe and I zfs from agent 1 and agent 3 in the lower layer, and adds them to the current time network. Then step 4.3 is performed to send the updated variable value domain to the lower agent.

[0069] Step 4.1.2: Agent 1 and agent 3 are in the order second layer, there is no neighbor agent in this layer, so it sends the value domain of shared variable x e and f s to the upper agent 1, and waits for the updated variable value domain I zas and I zbs from the upper layer, then step 4.2 is performed.

[0070] Step 4.2: For all variables r i ∈ V i of agent i, record the variable value domain I X sent by other agent, and compare it with the variable value domain I' zr received in the last iteration, if I' ze = I ze , then update the sign ze = false, otherwise do nothing. An empty list Q e is established for agent i's MaSTN.

[0071] Step 4.3: Reduce the feasible value domain of time variable through constraint propagation. Take agent 2 as an example.

[0072] Step 4.3.1: Choose a variable whose value domain is not calculated. Here we choose a s , and assign its value domain I zas = [8, 11] to I' zas .

[0073] Step 4.3.2: For all variables w i ∈ V i , if there is a constraint between a s and w i , and sign ae = true, then calculate sign denotes the sum of the endpoints of two intervals. For example, I zae = [8, 12], I aeas = [-1, -1], then and then compare the result with I zas .zas Intersection, until all variables w are traversed i The value domain after intersection is updated as I zas =[8,9.5].

[0074] Step 4.3.3: Compare the updated value domain I zas with I' zas , if I' zas =I zas , update the sign as =false, and add variable a s to the list Q2, if I' zas ≠I zas , it means the variable value domain is updated, if variable a s is in the list Q2, remove variable a s from the list Q2, otherwise, do nothing.

[0075] Obviously, I' zas ≠I zas , at this time, the list Q2 does not contain variable a s , so nothing is done.

[0076] Step 4.3.4: Return to step 4.3.1 until all variables are traversed.

[0077] Step 4.4: When all agents complete the constraint processing in this round, calculate the number of variables in each agent set V i L and the list Q i , if all agents satisfy V i L and the number of variables in Q i are equal, the calculation is completed, and the consistent network is returned, if they are not equal, return to step 4.3 to perform a new round of iteration. After the calculation, the value domain of each variable is shown in Table 2.

[0078] Table 2: Time variable value domain calculation results

[0079]

[0080]

[0081] It also includes step five: through the constraint processing of steps one to four, unnecessary constraint detection in the multi-node probe autonomous planning process is reduced, so as to realize the rapid processing of complex time constraints in the multi-node probe autonomous planning process, and improve the planning efficiency.

[0082] According to the above steps, for the time constraint network in the multi-agent planning process, the hierarchical asynchronous constraint consistency judgment method is adopted, part of the agents can use the updated external variable value domain for solving, the variable value domain is closer to the final result compared with the synchronous method, and the constraints that do not need to be repeatedly calculated are removed by adding the update flag to the variable, the constraint detection number of a single agent is reduced, the communication information number between the agents is reduced, the time constraint consistency in the multi-node detector planning process is quickly judged, and the planning speed is improved.

[0083] According to the method, after the multi-agent hierarchical asynchronous constraint processing is performed on the adopted embodiment, the constraint processing number, the constraint processing time and the communication information number are shown in Table 3. One constraint processing is performed, one communication information number is sent by the agent i to the agent j.

[0084] Table 3 Constraint processing number, constraint processing time and communication information number

[0085] Method Synchronous calculation Layered asynchronous calculation Constraint processing number 72 42 Constraint processing time 107.3 ms 98.9 ms Communication information number 24 pieces 14 pieces

[0086] According to the data in the table, compared with the distributed synchronous calculation method based on arc consistency, the hierarchical asynchronous arc consistency time constraint processing method reduces the constraint processing number, improves the constraint processing efficiency, reduces the communication information number between the agents, and can effectively speed up the planning speed.

[0087] The above specific description further describes the purpose, technical scheme and beneficial effects of the application, and it should be understood that the above description is only a specific embodiment of the application and is not used to limit the protection scope of the application, and any modification, equivalent replacement, improvement, etc. within the spirit and principles of the application should be included in the protection scope of the application.

Claims

1. A hierarchical asynchronous time constraint handling method for multi-agent explorer task planning, characterized in that: Comprising the following steps, Step one, establish a multi-node detector time planning problem model, and define the multi-node detector time planning problem as a six-tuple; the multi-node detector time planning problem model includes node ID, node state variable, initial state, target state, executable activity time variable and activity time constraint; Step two, express the activity time variable and time constraint in step one as a multi-agent simple time network, thereby converting the time constraint processing problem into a graph path problem, so as to facilitate solving in steps three and four; the activity time variable has a variable value range; The implementation method of step two is, Step 2.1: For single-node internal activity time variable and time constraint, establish a single-agent simple time network STN; For any two time variables u, w, define time zero z, then the value domain of u, w is I zu , zw ; if there is a time constraint relationship a ≤ w - u ≤ b, then the constraint is expressed as interval I uw = [a, b], if u, w belong to the same activity, then I uw represents the internal constraint of the activity, that is, the activity duration, if u, w belong to different activities, then I uw represents the inter-activity constraint; the time variables and time constraints of a single node activity are expressed as variable points and edges in the STN, that is, S = <V, D, C>, V represents the set of all time variables contained in the node, the set of time variables includes the start time point, the end time point and the time zero point, D represents the set of value domains of the time variables, the set of value domains includes the weights of the edges between the variable points and the time zero point, and C represents the set of constraint relationships between the time variables, the set includes the weights of the edges between the variable points. Step 2.2: Based on the agent simple time network obtained in step 2.1, consider the activity time constraints between nodes, establish a multi-agent simple time network MaSTN, thereby converting the time constraint processing problem into a graph path problem, so as to facilitate solving in steps three and four; For multi-agent temporal networks, the privacy of agents should be considered, and agents only exchange shared information. Therefore, the variables and constraints in STN are distinguished. For the simple temporal network STN i of agent i i , i , i , if any variable v i ∈V i has constraints c i ∈C i with other agents' variables, the variable is called a shared variable and is added to the shared variable set V i S . If all constraints of variable v i belong to agent i, the variable is called a private variable and is added to the private variable set V i P . If the variable of other agents has constraints with the variable of agent i, the variable is called an external variable and is added to the external variable set V i X . All variables contained in agent i are called internal variables and are added to the internal variable set V i L . Then, the variable set V i of agent i is V i X ∪V i L = V i X ∪(V i S ∪V i P ). Similarly, if the constraint c i ∈C i has one end variable v i ∈V i X , it is added to the external constraint set . If the constraint c i ∈C i has two end variables v i ,v j ∈V i L , it is added to the internal constraint set Step three, abstractly represent the multi-agent simple time network obtained in step two at the agent layer, and solve the agent calculation order in asynchronous processing according to the out-degree of the network and the number of shared variables of the agent; solving the asynchronous agent hierarchical calculation order set order, based on the multi-agent simple time network, through asynchronous calculation of the agent local constraint network, some agents can use the updated external variable value range for solving, and the internal variable value range reduction is closer to the final value range, thereby reducing the number of single-agent constraint detections; Step four, based on the order set order obtained in step three, process the time constraint network of agent i based on the arc consistency algorithm, reduce unnecessary constraint detection by adding a variable value range update flag, and then solve whether there is a value range for the activity time variable according to the variable value range reduction in step two, and judge whether the time constraint is consistent in the planning process according to the constraint processing result.

2. The layered asynchronous time constraint handling method for multi-agent probe mission planning of claim 1, wherein: Further comprising step five, reducing unnecessary constraint detection in the autonomous planning process of the multi-node detector through constraint processing in steps one to four, thereby realizing fast processing of complex time constraints in the autonomous planning process of the multi-node detector and improving planning efficiency.

3. The layered asynchronous time constraint handling method for multi-agent probe mission planning of claim 2, wherein: The implementation method of step one is, The multi-node detector time planning problem is defined as the following six-tuple Π=[P,ST,O,C,s0,g] (1) Where, P = {a1,...,a} m } represents the set of detector node IDs. Let represent the set of state variables for m nodes of the detector. Let represent the set of active time variables that can be executed by m nodes of the detector, and for any activity of any node. With start time point End time Duration Right now Let s0 represent the time constraint relationship between any two activities, s0 represent the initial state of the multi-node detector, and g represent the target state.

4. The layered asynchronous time constraint handling method for multi-agent probe mission planning of claim 3, wherein: The implementation method of step three is, Step 3.1: Abstractly represent the multi-agent simple time network obtained in step two at the agent layer, and establish an upper agent network P = <A, E> according to the constraint relationship between agents, A = {1, 2,..., n} represents the agent ID set, E = {..., (i, j),...} represents the connection relationship set between agents, (i, j) represents that there is a constraint between agent i and j; Step 3.2: Calculate the out-degree od of all agents according to the network P obtained in step 3.1, that is, the number of agents that have constraints with agent i; Wherein, if (i, j) E, then E(i)(j) = 1, otherwise E(i)(j) = 0; Step 3.3: solving the hierarchical computation order set of the asynchronous agents, the agent with the highest out-degree is placed in the first layer of the order set, and its ID is removed from set A, then the agents in set A which have constraints with the agent in the first layer are placed in the second layer of the order set, and their IDs are removed from set A, and so on until set A is empty, thus obtaining the hierarchical computation order set of the asynchronous agents; so that in step four, through the asynchronous computation of the agent local constraint network, some agents can use the updated external variable value domain to solve, and their internal variable value domain reduction is closer to the final value domain, thus reducing the number of single agent constraint detection.

5. The layered asynchronous time constraint handling method for multi-agent probe mission planning of claim 3, wherein: The implementation method of step four is, Step 4.1: Inter-agent information communication is performed according to the set order obtained in Step three; all variables v i ∈ V i L Add update flag sign v = true, only when the shared variable update flag sign v = true, the variable value domain is sent; Step 4.1.1: if the agent i is in the first layer of the order set, send the value domain of the shared variable to the neighbor agents in the layer, receive the external variable value domain sent by the neighbor agents in the layer and the lower layer and add it to the current time network; Then step 4.2 is executed, and the updated variable value domain is sent to the lower layer agent; Step 4.1.2: if the agent i is not in the first layer of the order set, send the value domain of the shared variable to the neighbor agents in the layer and the upper layer, wait for the updated variable value domain sent by the upper layer, and then execute step 4.3; Step 4.2: For all variables r i ∈ V i X , record the variable value domain I zr sent by other agents and compare it with the variable value domain I' zr received in the last iteration, if I' zr = I zr , update the flag sign r = false, otherwise do nothing; create an empty list Q i for the MaSTN of agent i; Step 4.3: reducing the feasible value domain of the time variable through constraint propagation; Step 4.3.1: Select the variable u for which the range is not computed i ∈ V i L , and assign its range I zu to I' zu ; Step 4.3.2: for variable w i ∈ V i , if there is a constraint between u i and w i , and sign w = true, then compute sign denotes the sum of the endpoints of the two intervals, respectively; Example: I zw = [a, b], I wu = [c, d], then The result is then intersected with I zu until all variables w i are exhausted; if the intersection is empty, then there is no feasible solution, and the time constraint process is ended, returning the conclusion that the constraints are inconsistent. Step 4.3.3: Compare the updated domain I zu with I' zu , if I' zu = I zu , update the sign v = false, and add the variable u i to the list Q i , if I' zu ≠ I zu , it means the domain of the variable is updated, if the variable u i is in the list Q i , remove the variable u i from the list Q i , otherwise, do nothing. Step 4.3.4: Go back to step 4.3.1 until all variables u i ∈ V i L are traversed; Step 4.4: When all agents in this round finish the constraint processing, calculate the number of variables in each agent set V i L and the list Q i , if the number of variables in V i L and Q i are equal, return the consistent network, if not, return to step 4.3 to perform a new round of iteration.

Images