CN111340348B - Distributed multi-agent task cooperation method based on linear time sequence logic - Google Patents
Distributed multi-agent task cooperation method based on linear time sequence logic Download PDFInfo
- Publication number
- CN111340348B CN111340348B CN202010108021.0A CN202010108021A CN111340348B CN 111340348 B CN111340348 B CN 111340348B CN 202010108021 A CN202010108021 A CN 202010108021A CN 111340348 B CN111340348 B CN 111340348B
- Authority
- CN
- China
- Prior art keywords
- agent
- task
- message
- action
- coupling
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/067—Enterprise or organisation modelling
Landscapes
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Economics (AREA)
- Game Theory and Decision Science (AREA)
- Educational Administration (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Numerical Control (AREA)
- Manipulator (AREA)
Abstract
The invention discloses a distributed multi-agent task cooperation method based on linear time sequence logic, which solves the problem of multi-agent task decoupling. Each intelligent agent constructs a self decoupling product type Buchi automatic machine by detecting the coupling edge, and constructs a self action sequence by the decoupling product type Buchi automatic machine; the end points of the coupling edges correspond to actions needing cooperation; when each intelligent agent independently executes the self action sequence by utilizing a decoupling product type Buchi automaton, judging whether the currently executed action and the corresponding trigger condition are in the coupling set, if so, judging that the currently executed action is the action needing cooperation, and requesting the cooperation intelligent agents to cooperate to make the action; when an agent fails, the agent responsible for inheritance inheriting the task of the failed agent is elected.
Description
Technical Field
The invention belongs to the field of artificial intelligence, and particularly relates to a distributed multi-agent task cooperation method based on linear time sequence logic.
Background
Linear Temporal Logic (Linear Temporal Logic) is a technical field with research prospects in the field of artificial intelligence at present. The linear sequential logic can model a series of complex tasks with sequential relationships in a programming language, and transfer the language through a graph model, so that the complex sequential relationships can be constrained through a series of conditional transfers. The method is applied in a wide range of fields due to its completeness. In the field of multi-agent, the multi-agent system has the characteristics of complex tasks and coupling in time sequence relation, so that some traditional methods are difficult to model, and linear time sequence logic can well model the complex tasks of the multi-agent system.
For the time sequence logic planning task in the multi-agent system, the existing solutions are as follows:
scheme 1: a centralized method for performing task planning in urban environment by applying linear sequential logic is proposed in literature (Yuhan Chen, Xu Chu Ding, all Stefanscu, and Calin Belta.A. formal approach to deployment of robust robots of in an urea like environment. in International Symposium on Distributed Autonomous robots System,2013.) each robot models the urban environment as a finite state transfer System and transfers its own tasks into corresponding Huchi robots to construct a product formula, Buchi. And then, each robot multiplies the product Buchi automata of all the robots, and an optimal path is planned for each robot through the finally obtained product automata of all the robots.
Scheme 2: the literature (Meng Guo and Dimos V diagnostics. Multi-agent plant configuration under 34(2):218, 235,2015.) constructs a hierarchical hybrid decision-control architecture for distributed multi-agent systems, which requires that each agent is assigned a linear timing logic formula as a task, and the agents cooperate with each other via a request-response communication model.
Scheme 3: document (Meng Guo and Dimos V dimalogonas. task and motion coordination for heterologous multiple systems with local coordination. IEEE Transactions on Automation Science and Engineering,14(2): 797-808, 2017.) this document distributively constructs an action sequence for each robot by predefining the coupling dependencies between the robot tasks in combination with the Dijkstra algorithm and guarantees synchronization of the cooperative actions by a request-response-confirmation communication model.
The prior art scheme is researched in the aspects of multi-agent task decoupling, information transmission during cooperation among multiple machines and processing strategy during node failure of a cluster part, but the effect is not ideal.
Disclosure of Invention
In view of this, the invention provides a distributed multi-agent task cooperation method based on linear sequential logic, which solves the problem of multi-agent task decoupling.
Furthermore, when part of the nodes in the cluster fail, other robots can react to the failure of the part of the robots, so that the cluster has robustness to the failure of the part of the nodes.
In order to solve the technical problem, the invention is realized as follows:
a distributed multi-agent task cooperation method based on linear time sequence logic comprises the following steps:
each intelligent agent constructs a self decoupling product type Buchi automaton and constructs a self action sequence through the decoupling product type Buchi automaton; the decoupling product type Guchi automaton has the following construction mode: each agent adopts a finite state transfer system for describing its working environmentAnd Buchi automaton describing its own tasksConstructing a product-type Buchi automaton PBA for guiding task execution, which is expressed asConstructing a LGBA of a B ü chi automaton of a generalized label, detecting a coupling edge in the LGBA, determining the coupling edge in the PBA according to the projection relation of the LBGA and the PBA, and recording the coupling edge and the trigger condition thereof into a coupling set; the end points of the coupling edges correspond to actions needing cooperation;
when each agent independently executes the action sequence of the agent by using a decoupling product type Buchi automaton, judging whether the currently executed action and the corresponding trigger condition are in the coupling set, if so, judging that the currently executed action is an action needing cooperation; at the moment, the action and the cooperation position which need to be cooperated are broadcasted to other intelligent agents, and the intelligent agent responsible for response makes a cooperation action;
when there is an agent p i When the intelligent agent fails, the failure event is broadcasted to other intelligent agents; enumerating Agents p responsible for inheritance j Inheriting a failed agent p i The task of (2).
Preferably, the recording process of the coupling set specifically includes:
marking nodes of the coupling tasks in the LGBA, and calling the nodes as coupling task nodes; marking child nodes of the coupling task nodes as the coupling task nodes; all edges containing coupled task nodes are marked as coupled edges (q) i ,q j ) (ii) a Will couple the edges (q) i ,q j ) Corresponding label lambda ij Set to true; the coupling task refers to a task completion state of a certain intelligent agent, and needs certain actions of other intelligent agents as conditions;
in building PBA, when an edge (q) that does not meet the branch condition occurs s ,q g ) Instead of deleting edges directly, the corresponding edge (q) in the LGBA is found according to the projection relationship of LGBA and PBA i ,q j ) Judging the found edge (q) i ,q j ) Tag lambda of ij Whether true; if true, the corresponding edge (q) in PBA is set s ,q g ) Determining as a coupled edge, and coupling the coupled edge (q) s ,q g ) And trigger condition g thereof sg Add to the coupling set of the PBA.
Preferably, the election agent p responsible for inheritance j Inheriting failed Agents p i The task of (1) is as follows:
all agents p receiving the broadcast of the failure event and having the same functionality as the failed agent j Constructing a united intelligent agent model; the united intelligent agent model is constructed as follows: constructing Agents p j Completing agent p i PBA of the task of (1), notedEstablishingToThe required inheritance task between the initial action nodes A is a jump action sequence; adding the jump action sequence toObtaining a combined agent model; wherein, the first and the second end of the pipe are connected with each other,for failed agent p i PBA of (3);
selecting an agent inheriting a task according to the cooperation cost indicated by the combined agent model; the agent inheriting the task executes the action sequence according to the joint agent model of the agent, and after the task is completed, the original decoupling product type Buchi automaton is switched back.
Preferably, the building of the united intelligent agent model comprises the following specific steps:
step b1, agent p j By describing their working environmentAnd describes an agent p i Of tasksConstruction of a decoupled product-based Guchi automaton
Step b2, assuming a failed agent p i Is currently in motionThe node in (1) isSlave nodeStarting to backtrack to a father node and searching for a forward node, wherein the backtrack reaches the initial node of the current taskAnd the process is finished,toAll the nodes form a set Q;
step b3, finding description failure agent p i Working environmentNeutralization ofContiguous adjacency points, denoted as point setsFind outNeutralization ofContiguous adjacency points, denoted as point setsWill point setAll ambient tags in (1) i And withCombined to form a plurality of new nodesForm set Y1; set pointsAll ambient tags in (1) j Combining with each node in Q to form a set Y2; will be provided withNode direction shown in the middle set Y2The nodes shown in the middle set Y1 are connected, and then the path searching operation is carried out to obtain the path informationThe combined agent model of the jump action sequence is added.
Preferably, the agent for election inheritance task is: and calculating the cost of the jump action sequence according to the joint agent model, and selecting the agent with the minimum cost as the agent for inheriting the task.
Preferably, the broadcasting the action and the cooperation position needing cooperation to other agents and the broadcasting the failure event to other agents are realized by adopting a gossip information transmission protocol:
establishing a Request message including k Reply message Reply k Acknowledgement message Confirm k Informing message Warning k The message group of (1);
when an agent performs an action requiring collaboration, a Request is issued to all other agents k A message; receiving the Request k The agent of the message calculates the cost of responding to the message and passes the cost through Reply k Informing other agents of completing Request through gossip protocol k Iteration of the message is carried out, and an agent with the minimum cost is found; finally passing through Confirm k End of message to Request k Responding to the message;
Warning k the information is used for the disabled agent to seek the inheritance task to other agents; sending Warning to other agents when an agent fails k Information; receiving Warning k The agent of the message firstly constructs a joint agent model, calculates the cost of inheriting the task to reach the task starting point according to the joint agent model, and passes the cost through Reply k Informing other agents of completing Request through gossip protocol k Iteration of the message is carried out, and the agent with the minimum cost is found to be used as the inheritance agent; finally passing through Confirm k End of message pair Warning k Responding to the message;
Request k message, Reply k Message, Confirm k Messages, Warning k The messages are all provided with version numbers so that the Request k Message and Warning k The message can realize the whole network broadcast and the election of the responder through the gossip information transmission protocol.
Preferably, the message group is:
wherein k is an agent identification number; sigma d An action requiring collaboration; pi h A location requiring cooperation; t is a unit of m Estimated time to reach a position needing cooperation for the requesting agent;indicating whether agent k determined to respond to the other agent's request information,responding to a Request for agent k k The time that the message is expected to take,indicating whether the requested action has been confirmed to be completed,predicting a time for the responding agent to pass through the collaboration zone; a. the k In order to disable the capabilities of the agent,the message sequence numbers of the messages are respectively, the sequence numbers are version numbers when the messages are generated, and the sequence numbers symbolize the sequence of the message generation.
Has the beneficial effects that:
(1) decoupling of the coupling tasks. When the Buchi automaton for guiding the execution of the tasks is constructed, the LGBA is used for detecting the coupling edge, and the coupling edge is added into each local decoupling product Buchi automaton, so that the process of action planning can be distributed, and the planning time complexity is reduced.
(2) And (4) designing a combined robot model. By backtracking the current state of the failure node and constructing a combined robot model, other robots can complete the succession of tasks on the premise of not destroying the time sequence relation of the tasks, and the cluster can continue to complete established tasks under the condition that partial nodes fail. The invention can greatly reduce the computational complexity of the process and improve the capability of the cluster for dealing with partial node failure.
(3) The combination of the communication model and gossip. In order to ensure the synchronization of the cooperative action, a request-response-confirmation model in the literature is combined with the gossip protocol, and the gossip protocol suitable for the scene is designed, so that the speed of consistent convergence of multiple robots is increased.
Drawings
FIG. 1 is a schematic diagram of a task collaboration method for multiple robots under normal working conditions;
FIG. 2 is a schematic diagram of a task collaboration method in the case of robot failure.
Detailed Description
The following description will be given taking a robot as an agent.
Referring to fig. 1 and 2, the distributed multi-robot task collaboration method based on linear sequential logic of the present invention includes the following steps:
● under normal working conditions, each robot independently constructs a self-decoupling product type Buchi automaton and constructs a self-action sequence through the automaton. The decoupled product Buchi automaton here is the result of adding a coupling set to the product Buchi automaton. The coupling set records the coupling edge and the trigger condition thereof, and the end point of the coupling edge corresponds to the action needing cooperation.
● when each robot independently executes the local action sequence by using a decoupling product type Buchi automaton, judging whether the currently executed action and the corresponding trigger condition are in the coupling set, if so, the currently executed action is the action needing cooperation; at the moment, the action and the cooperation position which need to be cooperated are broadcasted to other robots, and the robot in charge of response makes a cooperative action. If not, the original operation is continued.
● when a robot fails, the failure event is broadcast to other robots. And selecting the robot in charge of inheritance by the robot receiving the failure event, wherein the robot in charge of inheritance adopts the built combined robot model to inherit the task of the failure robot. In the invention, the broadcasting cooperation event and the failure event to other robots are propagated by utilizing the gossip protocol in the cluster, and iteration is carried out until convergence. In addition, the election operation may be implemented by priority or by judgment that the spatial distance is shortest. The embodiment of the invention provides a judgment mode based on a combined robot model and cooperation cost.
Therefore, the invention establishes a new theoretical framework for constructing the product automata and further deeply considers the specific information transmission process when the cooperation is carried out among multiple machines. Firstly, the scheme distributes the process of action planning by detecting the coupling edge and adding the coupling edge into each local decoupling product Buchi automaton, decouples tasks, and turns the planning process to local, so that the calculation complexity is low. Secondly, by constructing a joint agent model, other agents can react to the failure of part of agents, so that the cluster has robustness to the failure of part of nodes. In addition, by adopting the gossip protocol, the time for the multiple machines to converge to be consistent is short, and the amount of transmitted information is small.
The following describes the construction of a decoupling product type Buchi automaton, the processing of a cooperation action, the task inheritance realization of a failure robot and the design of a communication model based on a gossip protocol in detail.
Construction of (I) coupling task decoupling and decoupling Buchi automaton
(1) Finite state transition system
In order to determine the position transition relationship of the robot in the environment, the environment needs to be modeled first, and in the distributed mission planning framework, the environment is modeled as a Finite state transition system (FTS). FTS is defined as follows:
definition 1. finite state transfer system (FTS) consists of one tuple:
wherein:
Π={π 1 ,π 2 ,...,π N representing all areas of the rasterized working environment, wherein the areas are N areas; pi is also known as an environmental label, or state;
AP represents an atomic proposition describing a task that is not repartitionable;
L:Π→2 AP attributes representing task atom propositions that the grid region has;
State pi i Is denoted as Post (pi) i )={π j ∈Π|π i → c π j }. The sequence of movements of the robot can be represented by a sequence of infinite states, τ ═ pi 1 π 2 .., wherein, n i ∈Post(π i-1 )。
(2) Non-deterministic Buchi automaton
Forming task expressions by combining atomic propositions AP using Linear Temporal Logic (LTL) languageWith respect to each LTL expressionThere must be a corresponding automatic B ü chi machine (NBA) which is recorded as
wherein Q is the set of states in the automaton,representing the initial set of states in the automaton, 2 AP Representing a set of alphabets consisting of task atom propositions, delta: Qx 2 AP →2 Q Representing the transition relationships between states in an automaton, Q F Representing a set of acceptable states in the automaton.
(3) Multiplication type Buchi automaton
The product-based Buchi automaton is a combination of FTS and NBA and thus contains both environmental and task state information.
wherein Q ∈ Q * Nodes in NBA represent task states.
If and only if (π) i ,π j ) E → and q n ∈δ(q m ,L(π i ));q n Is a node, representing the action performed, L (π) i ) Is a trigger condition;
Q 0 * ={<π,q>|π∈Π 0 ,q∈Q 0 is the initial state set; .
W * (<π i ,q m >,<π j ,q n >)=W(π i ,π j )
wherein<π j ,q n >∈δ(<π i ,q m >)。
The coupling task refers to a task completion state of a certain robot, certain actions of other robots are required as conditions, and in order to distribute a task planning process, the task needs to be decoupled in a process of constructing a product type Buchi automaton. Thus increasing the flag bit τ in PBA, formingIs denoted by τ denotesIs not a coupled edge. The set of flag bits is called a coupling set. The coupling edge refers to a transfer relationship related to a coupling task in the PBA of each robot, and generally, such a transfer relationship cannot independently complete conversion on the condition that other robots simultaneously perform corresponding actions.
The detection idea of the coupling edge is as follows: constructing a LGBA of a B ü chi automaton of a generalized label, detecting a coupling edge in the LGBA, determining the coupling edge in the PBA according to the projection relation of the LBGA and the PBA, and recording the coupling edge in the PBA and trigger conditions thereof into a coupling set; the endpoints of the coupled edges correspond to actions that require cooperation.
The coupling edge detection process is specifically realized as follows:
and constructing a broad sense label Buchi automaton LGBA of the task. Here, LGBA may be reconstructed or LGBA formed during PBA construction may be extracted. And marking nodes where the coupling tasks are located in the LGBA, and calling the nodes as coupling task nodes. And traversing the child nodes of the coupling task in the LGBA, and marking all the child nodes of all the coupling task nodes as the coupling task nodes. An edge comprising one or two coupled task nodes is denoted as a coupled edge (q) i ,q j ). Where q is the end point of an edge. Will couple the edges (q) i ,q j ) Corresponding label lambda ij Set to true. The label can borrow the label in the LGBA, and can also newly set a label.
In the process of building PBA, when the PBA is pairedAndwhile performing multiplication, traverseAnd detecting whether the environment label of the subsequent node meets the transfer condition or not by each edge in the PBA, and if so, obtaining the edge in the PBA. In the existing processing procedure, when an edge (q) not meeting the transfer condition appears s ,q g ) And when the data is needed, the data is directly deleted. And the present invention is directed to these edges (q) s ,q g ) Not deleting directly, but finding out the sum (q) in LGBA according to the projection relation of LGBA and PBA s ,q g ) Corresponding edge (q) i ,q j ) Judging the found edge (q) i ,q j ) Tag lambda of ij Whether true; if true, the corresponding edge (q) in PBA is set s ,q g ) Determining as a coupled edge, coupling the edges (q) s ,q g ) And trigger condition g thereof sg Added to the coupling set of the PBA.
In the embodiment of the invention, the flag bit tau is constructed into a two-dimensional array, the rows and the columns respectively correspond to the end points of the edges, and the elements in the array record the trigger conditions. The flag bit τ is equivalent to recording the coupling edge and its trigger condition. Then, when the coupling edge (q) is determined s ,q g ) Then, the corresponding trigger condition g can be set sg Element τ (q) recorded to τ s ,q g ) In (1).
In the step, the coupling edge is detected and added into each local decoupling product Buchi automaton, so that the action planning process can be distributed, and the planning time complexity is reduced.
Design of combined robot model
In a complex environment, the robot may have a function damage phenomenon due to its own equipment, for example, three robots D A ,D B ,D C ,D A Is assigned to D B Task of opening the door, D B The goods need to be transported through the door D C Is assigned the task of cleaning the house, if D is the same A If a problem occurs, the door cannot be opened, the entire cluster cannot complete the task, and if D C The robot can be at D A Inheriting D temporarily when a problem occurs A The robot cluster can complete the task. In order to solve the problem of failure of part of nodes possibly occurring in a cluster and improve the robustness of the method, the invention designs an integration scheme of a failure robot. Let failed robot be p i When there is an agent p i When the intelligent agent fails, the failure event is broadcasted to other intelligent agents; all agents p receiving the broadcast of the invalidation event and having the same functionality as the invalidating agent j And constructing a combined robot model, and selecting the robot inheritance task with the minimum cooperation cost according to the cooperation cost indicated by the combined robot model, so that the robot can inherit the disabled robot task under the condition of minimum time complexity.
The construction idea of the combined robot model is as follows: construction robot p j Completing robot p i PBA of the task ofEstablishingToThe required inheritance task between the initial action nodes A is a jump action sequence; adding the jump action sequence toObtaining a combined robot model; wherein the content of the first and second substances,for a disabled robot p i PBA of (3).
The construction method of the combined robot model comprises the following specific steps:
step b1, marking the robot for constructing the combined robot model as p j . Robot p j By describing their working environmentAnd describe the disabled robot p i Of tasksConstructing a decoupling product type Buchi automaton according to step 1 The build has been completed in a previous step, where no repeat build is needed.
Step b2, assume failed robot p i Is currently inThe node in (1) isIn thatThe method comprises the steps of backtracking from a node to a father node and searching for the previous nodes, wherein the previous nodes are all relevant to the current task. Backtracking until reaching the starting node of the current taskAnd the process is finished, so that the process is finished,toAll nodes of (a) constitute a set Q. The Q contains the complete task, the start nodeAnd is also a place where the robot inheriting the task needs to access. Since the starting task is coded into a special identifier, the task starting node can be identified by means of the special identifier.
Step b3, finding description failure robot p i Working environmentNeutralization ofContiguous adjacency points, denoted as point setsFind outNeutralization ofContiguous adjacency points, denoted as point setsThe adjacent points are twoThe closest point, i.e. the point that can be reached by a one-step jump. Set pointsAll ambient tags in (1) i Andcombined together by permutation and combination to formMultiple new nodesForming set Y1. All points in set Y1 areIn (1). Will point setAll the environmental labels in (1) j The nodes in Q are combined and arranged to form a set Y2. All points in the set Y2 areIn (1). Will be provided withNode direction shown in the middle set Y2The nodes shown in the set Y1 make a connection, the connection having a direction, through which the connection is establishedToWhile simultaneously adding the jump action sequence toIn (1). And then, performing path searching operation to obtain a combined robot model. The path search operation may employ Dijkstra's algorithm. Through the above calculation, generate the slaveTransferring toWay in which tasks are inheritedAnd the task timing relation is not violated at the same time.
Step b4, selecting the robot with inherited tasks: the cost of the jump action sequence, i.e. the cost of the robot jumping from the current position to the task initial action node a, which may be for example time, or other information, is calculated according to the joint robot model. And selecting the robot with the minimum cost as the robot for inheriting the task.
Step B5, the robot inheriting the task executes the action sequence according to the self combined robot model, and after the task is finished, the original decoupling product type Buchi automaton is switched back
The combined robot model can enable other robots to complete inheritance of tasks on the premise of not destroying the time sequence relation of the tasks, so that the cluster can continue to complete established tasks under the condition that partial nodes fail. Meanwhile, the computational complexity of the process is greatly reduced, and the capability of the cluster for dealing with partial node failures is improved.
Design of communication model
In the framework, in order to avoid the situation that the robot runs a planned path locally and the collaboration is not synchronous, a request-response-confirmation communication model based on gossip is adopted. (Meng Guo and Dimos V collaborative tasks. task and movement correlation for a heterologous multiagent system with local coordinated tasks. IEEE Transactions on Automation Science and Engineering,14(2):797 and 808,2017.) A request-response-validation model is proposed that guarantees the completion of collaborative tasks during a robot task by requesting, responding and validating three types of messages. However, the model is too ideal in practical application, and the process that information propagates in the cluster to reach convergence is not considered. The invention considers the message transmission in the multi-robot cluster in more detail by combining the communication model and the gossip information transmission protocol, and accelerates the message convergence. At the same time increase Warning k A message. Under gossip protocol, the format of each messageComprises the following steps:
wherein, Request k For request messages, Reply k In response to the message, Confirm k To acknowledge messages, Warning k To inform the message. k is the number of the robot. Sigma d Filling the trigger condition of the action for the action needing cooperation; pi h Filling the environment tags in the positions needing cooperation; t is m And estimating the time of the robot to reach the position needing cooperation for the requesting robot.Indicates whether the robot k determines to respond to the Request of other robots k The message, the element may be an array, the location of the array corresponds to a different robot, and 1/0 at the location indicates whether the corresponding robot responds to the request.Responding to Request for local robot k k The time that the message is expected to take,in order to confirm whether the assistance action is completed,the time for the responding robot to pass through the cooperation area is estimated. Burning k Messages are prepared for the combined robot model, sent out for a disabled robot, seeking other robots to inherit their own tasks, A k In order to disable the capabilities of the robot, the legacy robot is required to have the same capabilities;the message sequence numbers of the messages are respectively the version numbers of the messages when the messages are generated, and the sequence numbers symbolize the sequence of the message generation.
When the action needing cooperation is executed, the robot constructs a Request message, and the action sigma needing cooperation is added into the Request message d And a cooperation position pi h And calculates the time of the self-arrival at the position needing cooperation. And transmitting the Request message in the cluster by using a gossip protocol, iterating until convergence, and making a robot in charge of response perform a cooperative action.
The iterative process of the Request message is as follows: each robot continuously transmits the Request message received by the robot to the neighbor node; when each node receives a Request message, if the node does not generate a Reply message, estimating an estimated cooperation Cost, namely estimating the estimated time spent by the node in response to the Request message; the Cost is compared with t in other received replies d By comparison, if Cost is small, it is more appropriate to respond to the request, and all 0's are generatedUpdating the corresponding position of the self to be 1, and filling the Cost into the corresponding positionIn generating a new Reply message, ofSet to be in all Reply messagesThe maximum value of (2) plus 1, and propagation is carried out; if a Reply message is generated by itself and the message is stored in the robot, judging other received Reply messagesWhether the message is larger than the message sent by the Reply message; if receivedIf the message size is larger than the preset value, updating the content in the Reply message stored by the message by using the content in the received Reply message; if receivedAnd if the value is small, the treatment is not carried out. In the end of this process,andis updated as information about the robot responding to the request. When new, less costly collaborators appear, only all are the sameThe robot updates the messages synchronously, thereby greatly reducing the communication traffic in the network. When gossip protocol is iterated to all robotsThe method for judging whether the consistency is reached is that the messages are continuously received for n times for each robotIf the communication Request does not change, the robot responding to the Request is generated, and the communication iteration process is finished, so that the robot responds.
And then, when the triggering condition of the transfer relationship of the robot proposing the Request message is met, the robot proposing the Request message generates a Confirm message, the message is propagated in the cluster, and the received robot compares the cooperative action, deletes the Request message corresponding to the Confirm message and all the generated Reply messages corresponding to the Request. And then back to the task before responding.
The response mechanism of the Warning message is consistent with the Request message, A k The same robot processes the message. When the robot fails, transmitting the warming information to other robots; the robot receiving the surfing message firstly constructs a combined robot model, calculates the cost of the inheritance task reaching the task starting point according to the combined robot model, informs other robots of the cost through Reply, completes the whole network iteration of the Request message through a gossip protocol, and finds the robot with the minimum cost as the inheritance robot. Reply i And σ in the Confirm message d No recognizable numerical value such as-1 is required to be filled in or otherwise filled in.
In summary, the above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (6)
1. A distributed multi-agent task cooperation method based on linear time sequence logic is characterized by comprising the following steps:
each agent constructs its own decoupling product formulaAutomaton and by the decoupled product formThe automaton constructs a self-body action sequence; the decoupled product formThe automaton has the following construction mode: each agent adopts the function of describing its working environmentLimited state transfer systemAnd describing self-tasksAutomatic machineConstructing a product formula directing task executionAutomaton PBA, denoted asConstructing generalized labelsAn automatic machine LGBA, detecting a coupling edge in the LGBA, determining the coupling edge in the PBA according to the projection relation between the LBGA and the PBA, and recording the coupling edge and the trigger condition thereof into a coupling set; the end points of the coupling edges correspond to actions needing cooperation;
using decoupled product equations at each agentWhen the automaton independently executes the action sequence of the automaton, judging whether the currently executed action and the corresponding trigger condition are in the coupling set, if so, judging that the currently executed action is an action needing cooperation; at the moment, the action and the cooperation position which need to be cooperated are broadcasted to other intelligent agents, and the intelligent agent responsible for response makes a cooperation action;
when there is an agent p i When the intelligent agent fails, the failure event is broadcasted to other intelligent agents; electing agent p responsible for inheritance j Inheriting a failed agent p i The task of (1);
the election agent p responsible for inheritance j Inheriting a failed agent p i The task of (1) is as follows:
all agents p receiving the broadcast of the failure event and having the same functionality as the failed agent j Constructing a united intelligent agent model; the united intelligent agent model is constructed as follows: building Agents p j Completing agent p i PBA of the task of (1), notedEstablishingToThe required inheritance task between the initial action nodes A is a jump action sequence; adding the sequence of jump actions toObtaining a joint agent model; wherein, the first and the second end of the pipe are connected with each other,for failed agent p i PBA of (3);
selecting an agent of an inherited task according to the cooperation cost indicated by the combined agent model; the agent inheriting the task executes the action sequence according to the joint agent model of the agent, and after the task is completed, the agent is switched back to the original decoupling product formulaAn automaton.
2. The method according to claim 1, characterized in that the recording process of the coupling set is specifically:
nodes of the coupling tasks are marked in the LGBA and are called as coupling task nodes; marking the child nodes of the coupling task node as the coupling task node; all edges containing the nodes of the coupled task are recordedIs a coupled edge (q) i ,q j ) (ii) a Will couple the edges (q) i ,q j ) Corresponding label lambda ij Set to true; the coupling task refers to a task completion state of a certain intelligent agent, and needs certain actions of other intelligent agents as conditions;
in building PBA, when an edge (q) that does not meet the branch condition occurs s ,q g ) Instead of deleting the edge directly, the corresponding edge (q) in the LGBA is found according to the projection relation between the LGBA and the PBA i ,q j ) Judging the found edge (q) i ,q j ) Tag lambda of ij Whether true; if true, the corresponding edge (q) in PBA is set s ,q g ) Determining as a coupled edge, and coupling the coupled edge (q) s ,q g ) And trigger condition g thereof sg Add to the coupling set of the PBA.
3. The method of claim 1, wherein the building of the federated agent model comprises the specific steps of:
step b1, agent p j By describing their working environmentAnd describe agent p i Of tasksConstructing decoupled product formAutomatic machine
Step b2, assuming failed agent p i Is currently in motionThe node in (1) isSlave nodeStarting to backtrack to a father node and searching for a previous node, wherein the backtrack reaches the initial node of the current taskAnd the process is finished, so that the process is finished,toAll the nodes form a set Q;
step b3, finding out description failure intelligent agent p i Working environmentNeutralization ofContiguous adjacency points, denoted as point setsFind outNeutralization ofContiguous adjacency points, denoted as point setsWill point setAll ambient tags in (1) i Andcombined to form a plurality of new nodesForm set Y1; will point setAll the environmental labels in (1) j Combined with each node in Q to form a set Y2; will be provided withNode direction shown in the middle set Y2The nodes shown in the middle set Y1 are connected, and then the path searching operation is carried out to obtain the path informationThe combined intelligent agent model of the jump action sequence is added.
4. The method of claim 1, wherein the agent that elects the inherited task is: and calculating the cost of the jump action sequence according to the joint agent model, and selecting the agent with the minimum cost as the agent for inheriting the task.
5. The method of claim 1, wherein the broadcasting actions and collaboration locations requiring collaboration to other agents and the broadcasting failure events to other agents are implemented using gossip messaging protocols:
establishing a Request message including k Reply message Reply k Acknowledgement message Confirm k Informing message Warning k The message group of (1);
when the intelligent agentWhen the action needing cooperation is executed, the Request is sent to all other agents k A message; receiving a Request k The agent of the message calculates the cost of responding to the message and passes the cost through Reply k Informing other agents of completing Request through gossip protocol k Iteration of the message is carried out, and an agent with the minimum cost is found; finally passing through Confirm k End of message to Request k Responding to the message;
Warning k the information is used for the disabled agent to seek the inheritance task to other agents; sending Warning to other agents when an agent fails k Information; receiving Warning k The agent of the message firstly constructs a joint agent model, calculates the cost of inheriting the task to reach the task starting point according to the joint agent model, and passes the cost through Reply k Informing other agents of completing Request through gossip protocol k Iteration of the message is carried out, and the agent with the minimum cost is found to be used as the inheritance agent; finally passing through Confirm k End of message pair Warning k Responding to the message;
Request k message, Reply k Message, Confirm k Messages, Warning k The messages are all provided with version numbers so that the Request k Message and Warning k The message realizes the whole network broadcast and the election of the responder through the gossip information transmission protocol.
6. The method of claim 5, wherein the set of messages is:
wherein k is an agent identification number; sigma d An action requiring collaboration; pi h A location requiring cooperation; t is a unit of m Estimated time to reach a position needing cooperation for the requesting agent;indicating whether agent k determined to respond to the request information of the other agents,responding to a Request for agent k k The time that the message is expected to take,indicating whether the requested action has been confirmed to be completed,predicting a time for the responding agent to pass through the collaboration zone; a. the k In order to disable the capabilities of the agent,the message sequence numbers of the messages are respectively the version numbers of the messages when the messages are generated, and the sequence numbers symbolize the sequence of the message generation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010108021.0A CN111340348B (en) | 2020-02-21 | 2020-02-21 | Distributed multi-agent task cooperation method based on linear time sequence logic |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010108021.0A CN111340348B (en) | 2020-02-21 | 2020-02-21 | Distributed multi-agent task cooperation method based on linear time sequence logic |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111340348A CN111340348A (en) | 2020-06-26 |
CN111340348B true CN111340348B (en) | 2022-07-26 |
Family
ID=71184161
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010108021.0A Active CN111340348B (en) | 2020-02-21 | 2020-02-21 | Distributed multi-agent task cooperation method based on linear time sequence logic |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111340348B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113255967A (en) * | 2021-04-28 | 2021-08-13 | 北京理工大学 | Task planning method and device based on end point backtracking under signal time sequence logic constraint |
CN115576278B (en) * | 2022-09-30 | 2023-08-04 | 常州大学 | Multi-agent multi-task layered continuous control method based on temporal equilibrium analysis |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
FR2968809A1 (en) * | 2010-12-09 | 2012-06-15 | Commissariat Energie Atomique | METHOD AND APPARATUS FOR GENERATING A CONTROL SYSTEM FROM SPECIFIC REDUCED BEHAVIOR |
CN109657868A (en) * | 2018-12-26 | 2019-04-19 | 北京理工大学 | A kind of probabilistic programming recognition methods of task sequential logic constraint |
CN110014428A (en) * | 2019-04-23 | 2019-07-16 | 北京理工大学 | A kind of sequential logic mission planning method based on intensified learning |
-
2020
- 2020-02-21 CN CN202010108021.0A patent/CN111340348B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
FR2968809A1 (en) * | 2010-12-09 | 2012-06-15 | Commissariat Energie Atomique | METHOD AND APPARATUS FOR GENERATING A CONTROL SYSTEM FROM SPECIFIC REDUCED BEHAVIOR |
CN109657868A (en) * | 2018-12-26 | 2019-04-19 | 北京理工大学 | A kind of probabilistic programming recognition methods of task sequential logic constraint |
CN110014428A (en) * | 2019-04-23 | 2019-07-16 | 北京理工大学 | A kind of sequential logic mission planning method based on intensified learning |
Non-Patent Citations (3)
Title |
---|
Distributed Cooperative Fault Detection for Multiagent Systems A Mixed H infinity H2 Optimization Approach;Hao Fang et al.;《IEEE Transactions on Industrial Electronics》;20180208;第65卷(第8期);全文 * |
Distributed Cooperative Fault Detection for Multi-agent Systems with Time-Varying Delay;Hao Fang et al.;《2018 Annual American Control Conference》;20180816;全文 * |
基于时序活动逻辑的复杂系统多Agent动态协作模型;蒋伟进等;《计算机学报》;20130515;第32卷(第05期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN111340348A (en) | 2020-06-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111340348B (en) | Distributed multi-agent task cooperation method based on linear time sequence logic | |
Xuan et al. | Communication decisions in multi-agent cooperation: Model and experiments | |
Kantaros et al. | Global planning for multi-robot communication networks in complex environments | |
Qasim et al. | An ant colony optimization based approach for minimum cost coverage on 3-D grid in wireless sensor networks | |
Yan et al. | Formation trajectory tracking control of discrete-time multi-AUV in a weak communication environment | |
Aragues et al. | Distributed consensus algorithms for merging feature-based maps with limited communication | |
Shiroma et al. | CoMutaR: A framework for multi-robot coordination and task allocation | |
Guo et al. | Task and motion coordination for heterogeneous multiagent systems with loosely coupled local tasks | |
Seraj et al. | Learning Efficient Diverse Communication for Cooperative Heterogeneous Teaming. | |
Capitán et al. | Decentralized delayed-state information filter (DDSIF): A new approach for cooperative decentralized tracking | |
Ong et al. | Cooperative collision avoidance via proximal message passing | |
Li et al. | Early validation of cyber–physical space systems via multi-concerns integration | |
Mahajan et al. | Identifying tractable decentralized control problems on the basis of information structure | |
Liu et al. | Distributed asynchronous extended target tracking using random matrix | |
CN117251825B (en) | Multi-sensor data fusion platform for new energy power station | |
Capitan et al. | Delayed-state information filter for cooperative decentralized tracking | |
Gulzar et al. | Cucker-Smale Model Based Consensus Algorithm for Enhanced Convergence in Communication Network of Distributed Energy Systems | |
CN115759199B (en) | Multi-robot environment exploration method and system based on hierarchical graph neural network | |
Phan-Minh | Contract-based design: Theories and applications | |
Huang et al. | Failure-robust multi-robot tasks planning under linear temporal logic specifications | |
Tošić et al. | Distributed protocols for multi-agent coalition formation: a negotiation perspective | |
Kim et al. | DisCoF+: Asynchronous DisCoF with flexible decoupling for cooperative pathfinding in distributed systems | |
Li et al. | Labeled multi-Bernoulli filter based group target tracking using SDE and graph theory | |
Klaesson et al. | Planning and optimization for multi-robot planetary cave exploration under intermittent connectivity constraints | |
Mi et al. | A mobile robot controller using reinforcement learning under scLTL specifications with uncertainties |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |