WO2006118193A1

WO2006118193A1 - Agent and distributed restriction supplementing method

Info

Publication number: WO2006118193A1
Application number: PCT/JP2006/308836
Authority: WO
Inventors: Yasuki Iizuka; Takashi Shimojima
Original assignee: Matsushita Electric Industrial Co., Ltd.
Priority date: 2005-04-27
Filing date: 2006-04-27
Publication date: 2006-11-09
Also published as: JPWO2006118193A1

Abstract

An agent comprises a variable storage section storing variable data, a restriction storage section storing restriction data, a variable change inhibition period storage section storing variable change inhibition period data, an improvement easiness creating section for creating improvement easiness data representing the easiness for the variable data to satisfy restriction data, a communication section for communicating the improvement easiness data with neighboring agents, a neighbor situation storage section storing the improvement easiness data on the neighboring agents, a constriction solving section for comparing the improvement easiness data and that on the neighboring agents, changing variable data to solve restriction violation according to the variable change inhibition period data, and updating the variable change inhibition period data. Such many agents cooperate asynchronously and solve restriction violation. The whole group of agents obtains a solution more quickly without falling into a local optimum solution.

Description

Agents and distributed constraint satisfaction methods

Technical field

TECHNICAL FIELD [0001] The present invention relates to an agent that communicates with neighboring agents and obtains a solution for satisfying a constraint condition in an asynchronous manner, and an agent distributed constraint satisfaction method. Background art

[0002] As a device that creates a plan that satisfies the constraints specified between one plan and other plans, such as production plans for factories, nurses' work schedules for hospitals, and logistics plans, Satisfaction devices are known. More specifically, the conventional constraint satisfaction device is composed of computer hardware and software capabilities. Figure 29 shows the configuration of a computer that constitutes a conventional constraint satisfaction device. In FIG. 29, a computer 900 includes a CPU 901, a memory 902, a display unit 903, an input unit 904, a secondary storage unit 905, and a network interface 906. Further, the secondary storage unit 905 stores initial values of variables, constraint data, and software for controlling each unit.

[0003] Further, it is known that a constraint satisfaction device can generally apply its operation principle to a constraint satisfaction problem. This can be explained, for example, in Japanese Patent Laid-Open No. 11-316682 (pages 4-5) and Mitsuru Ishizuka, “Course of Information Science Core Curriculum, Knowledge Representation and Fast Reasoning” (Maruzen, 1996, p. 20, 103). — 119).

[0004] This constraint satisfaction problem is explained as follows. In other words, the constraint satisfaction problem is a set of m variables xl, x2,..., Xm and the possible values of each variable Dl, D2,..., Dm, and a set of constraints between each variable When P = {pi, p2, · · ·, pr} exists, it is to find a combination in which the value of the variable satisfies all of this constraint. In other words, in predicate logic, it is to find a set of variables when expression 1 is true.

[0005] [Equation 1]

("*, ^ X ₂ } * * *

[0006] Algorithms and search methods for constraint satisfaction problems have already been proposed. Known methods include a knock track search method that searches for all combinations of solutions, a hill-climbing method that uses knowledge information for a problem called heuristics, a best-first search, and an A * (A 'star) algorithm. Since heuristic search progresses in the direction of improving the evaluation value of the solution, it reaches the solution at high speed, but if it falls into the local optimal solution, it cannot escape, and there is a possibility that it cannot reach the true solution. Also known are methods called stochastic search, such as simulated 'annealing and genetic algorithms. In addition, tabu search, which is a generalized local search method, is known. Tabu search has a mechanism that avoids searching for recently selected neighborhood solutions, and makes searching for neighborhoods efficient. This includes, for example, Japanese Patent Laid-Open No. 11-195 066 (page 9, Fig. 2) and Mitsuru Ishizuka "Information Science Core Curriculum Course, Knowledge Representation and Fast Inference" (Maruzen, 1996, p. 20, 103). — 119), Sadick · Μ Site, 1 other author, Hiroshi Shiraishi, “Latest method of combinatorial optimization algorithm” (Maruzen, 2002, p. 163), Rina's Deciter, “Constraint” 'Processing' ((USA), Morgan 'Power Ufman' Publicishers, 2003).

[0007] Since the constraint satisfaction problem can be expressed using a constraint network diagram, the constraint network diagram will be described here.

FIG. 30 is a diagram for explaining a constraint network diagram expressing the constraint satisfaction problem. In Fig. 30, node 1001 indicates the variable of the constraint satisfaction problem, and arc 1002 indicates the constraint relationship between the variables. The constraint satisfaction problem can be expressed in such a constraint network diagram without losing generality because the constraint between η variables can be expressed by converting it into a binary constraint. For example, an example will be described in which each node takes one of the values “black” and “white” and the nodes at both ends of each arc cannot have the same value.

FIG. 31A is a constraint network diagram showing an example of an initial state of a constraint satisfaction problem, and FIG. 31B is a constraint network diagram showing an example of a state in which a solution of the constraint satisfaction problem is defined. In FIG. 31A and FIG. 31 ノード, a node with a “black” value is shown in black, and a node with a “white” value is shown in white. In FIG. 31A, the values of node Χ203 and node Χ206 are both “black”, and a constraint violation has occurred between them. Also, a constraint violation has occurred between node Χ204 and node Χ205, which have the same value “white”. Figure 31B shows the solution to this problem when the constraints between all nodes are met. One of them shows a fixed state. In the following description, such a restricted network diagram is used as appropriate.

[0010] In particular, among the constraint satisfaction devices, between a distributed plan such as a network resource allocation plan, a production plan including a large number of operations, or a work stoppage plan for power system equipment, and other plans. As a device that asynchronously cooperates with other devices to create a plan that satisfies the specified constraints, a distributed constraint satisfaction device composed of multiple agents is known. It is known that the operating principle of the agents that make up this distributed constraint satisfaction device can be generally applied to distributed constraint satisfaction problems in which the variables and constraints of the constraint satisfaction problem are distributed.

Here, a constraint network diagram expressing the distributed constraint satisfaction problem will be described. A distributed constraint satisfaction problem can also be expressed as a constraint network diagram in the same way as the constraint satisfaction problem because the constraint between n variables can be expressed as a binary constraint. FIG. 32 is a diagram for explaining a constraint network diagram expressing the distributed constraint satisfaction problem. In FIG. 32, 1003a, 1003b, and 1003c are agents, a node 1001 is a variable that the agent 1003a has, and an arc 1002 indicates a constraint relationship between these variables. In this way, each agent targets some variables in the distributed constraint satisfaction problem and resolves the constraints asynchronously with agents that have neighboring constraint relationships with the target.

[0012] Here, an example will be described in which a production plan is made so that individual companies such as multiple independent companies and factories that have been signed through delivery contracts make profits, and the profits are increased as a whole. . Plans are prepared independently and in parallel by each operator, and the plans are coordinated between each company. This overall production planning can be generally described as a distributed constraint satisfaction problem. In Fig. 33, business operators F1 to F7 individually formulate production plans for parts A to E and semi-finished products F, such as the number of products produced, inventories, and the number of deliveries. Adjust the overall plan. At this time, it is necessary to satisfy each constraint condition among business operators such as delivery date, price, production period, and production capacity. However, in order for business operator F1 to create each plan without leaking confidential information of business operator F3 and business operator F5 to business operator F4, it is not necessary to collect information on distributed variables and constraints in one place. We need a way to solve the problem. In this way, in the distributed constraint satisfaction problem, communication costs increase when trying to communicate information about constraints, and variable data that is internal data is sent to other agents. It is not practical to collect and solve information about the distributed constraint satisfaction problem in one place because of problems such as having to conceal it.

[0013] It is known that it is very difficult to apply an existing algorithm or search algorithm for solving a constraint satisfaction problem to a distributed constraint satisfaction problem as it is. As a method for searching for a backtrack in a distributed environment, an asynchronous distributed backtrack method is known. Knock track search is a simple method that remembers the branch point during the search and returns to the branch point to search for another route if a contradiction occurs during the search. In contrast, the asynchronous distributed backtracking method requires a complex configuration that uses and separates two types of messages and multiple memories to achieve this in a distributed environment. In addition, this asynchronous distributed backtracking method is a full solution search using backtracking in the same way as the knocktracking search, and therefore the search takes a very long time. In addition, each agent must maintain a large amount of constraint violation information called “no good”. This algorithm is, for example, disclosed in Japanese Patent Application Laid-Open No. 11-316682 (pages 4-5), Makoto Yokoo, and three others, “Formulation and Solution of Distributed Cooperative Problem Solving by Satisfying Distributed Constraints” (The Institute of Electronics, Information and Communication Engineers) Journal D—I, 1992, Vol. 75, No. 8, p. 704—713), Makoto Yokoo, and 1 other, “Distribution constraint satisfaction algorithm: review” (autonomous agents and multi-agent systems) (USA), 2000, No. 3, IV, p. 189-212).

[0014] Also, an asynchronous weak commitment search method that improves the asynchronous distributed backtracking method is known. Asynchronous weak commitment search is fast by introducing the concept of priority, but it needs to hold and exchange large amounts of constraint violation information. In addition, costs such as searching for the information and generating further constraint violation information depending on conditions are necessary. Moreover, it is necessary to distribute the restriction information between specific agents to the third agent, and security between agents cannot be maintained. This algorithm is, for example, Lina 'Decita's' Constraint' Processing '(USA), Morgan' Kaufman's' Publishers', 2003, and Yokoo Makoto 'Asynchronous Weak Commitment Search Method' (1st Constraint Programming International Conference on the Principles and Practices of Science (CP-95)), (USA), 1995, p. 407-422)).

[0015] In addition, as an algorithm that does not require a knock track and does not have constraint violation information, A diffuse breakout algorithm is known. However, under certain conditions, this algorithm may fall into a loop and fail to obtain a solution. In addition, because the constraints are weighted, it is not possible to deal with dynamic situations in which the constraints change while seeking a solution. This includes, for example, Japanese Patent Laid-Open No. 9-297689 (page 4, Fig. 1), Makoto Yokoo, and one other, "Distributed breakout: Iteratively improved distributed constraint satisfaction algorithm" (Journal of Information Processing Society of Japan, 19 1998, No. 39, No. 6, p. 1889-1897) etc.

[0016] In addition, the conventional agent, like the conventional constraint satisfaction device, more specifically includes the hardware and software of the computer. In particular, multiple agents communicate asynchronously over the network and work together to resolve constraint violations.

[0017] In the conventional agent described above, each agent tries to obtain a variable for eliminating the constraint violation state between agents asynchronously. Therefore, there is a problem that there is a high possibility of falling into a local optimal solution and a possibility of falling into an infinite loop of processing in which some agents change values in order.

Disclosure of the invention

[0018] The present invention solves such a problem, and when obtaining a variable for eliminating the constraint violation state between agents asynchronously, it does not attempt to resolve the constraint violation with only some of the agents. By solving constraint violations with many agents, we provide an agent in which the entire set of agents reaches the solution earlier as a result of not falling into a local optimal solution or an infinite loop.

[0019] In order to solve the above-described problem, the agent of the present invention is a variable storage that stores variable data indicating a current value of the solution to be obtained in an agent in which a plurality of agents obtain the solution in cooperation in an asynchronous manner. Part, a constraint storage unit for storing constraint data indicating a combination of values of variable data and variable data stored by neighboring agents, and a variable for storing variable change prohibition period data indicating a period during which the change of variable data is prohibited Generated by the change prohibition period storage unit, the improvement degree generation unit that generates the improvement degree data indicating the degree of ease with which the variable data of the own agent satisfies the constraint data, and the variable data and improvement degree generation unit Communication unit that transmits / receives improvement degree data to / from neighboring agents, improvement degree data and variable data obtained from neighboring agents Configuration to The neighborhood situation storage unit that stores the neighboring situation data, the improvement degree data generated by the improvement degree creation unit, and the improvement degree data of the neighboring agents stored in the neighborhood situation storage unit are compared, and According to the change prohibition period data, the variable data is changed to a value that satisfies the combination of the constraint data so that the constraint violation with the variable data of the neighboring agent stored in the neighborhood status storage unit is resolved, and the variable A constraint resolution unit that sets the change prohibition period data to be a predetermined period.

[0020] Therefore, by prohibiting the continuous change of variables for a certain period under a certain condition, the constraint violations of many agents can be resolved just by eliminating the constraint violations of some agents. As a result, the entire set of agents can reach the solution more quickly as a result of not falling into the local optimal solution.

In the agent of the present invention, the improvement ease data includes at least the number of constraints indicating the number of constraint data, the number of constraint violations indicating the number of variable data violating the constraint data, and the variable data When the value is changed, the structure has one of the possible improvement numbers indicating the number of constraint data that can resolve the violation status, and the ease of improvement generation unit calculates the total number of constraint data for the variable data. If the variable data and the variable data of the neighboring agent satisfy the combination of the values of the constraint data, the total number of constraint data is obtained as the constraint violation number, and the variable data and the variable data of the neighboring agent Among the constraint data that does not satisfy the constraint data value combination, the variable data is not included in the variable change prohibition period data, and the variable data value is changed. Therefore, the total number of combinations that can be changed to satisfy the combination of constraint data values is obtained, and the attribute value of the improvement ease data is generated as the number that can be improved.

[0022] Therefore, it becomes possible to determine whether or not the agent should change the variable data based on the degree to which the variable data satisfies the constraint data. As a result, the entire set of agents may reach the solution earlier. it can.

[0023] The agent of the present invention further includes an improvement trend storage unit that stores improvement trend data indicating past transitions of the degree to which the variable data satisfies the constraint data, and the constraint resolution unit includes: Based on the number of constraint violations included in the variable data, the improvement trend data stored in the improvement trend storage is additionally updated, and the length of the variable change prohibition period is determined according to the improvement trend data. Then, after changing the variable data, the variable change prohibition period data stored in the variable change prohibition period storage unit is updated.

[0024] Therefore, the agent can adjust the variable change prohibition period of the variable data according to the improvement tendency of the constraint violation and prohibit the change of the variable only for the period adapted to the problem. The possibility of falling into an infinite loop of only a part of the agents or falling is reduced, and as a result, the entire set of agents can reach the solution faster.

[0025] Further, in the agent of the present invention, the improvement level generator generates the improvement level data at every first fixed time interval, and the communication unit transmits the improvement level data to neighboring agents. Configure.

[0026] Therefore, the agent can periodically acquire the improvement degree of neighboring agents, and determines whether or not to change the variable based on the more recent improvement status, and as a result, the entire set of agents is determined. Can reach the solution faster.

[0027] Also, in the agent of the present invention, the improvement ease generation unit generates improvement ease data at every second fixed time interval, and the constraint resolution unit determines the neighborhood change according to the variable change prohibition period data. The variable data is changed so as to eliminate the constraint violation with the variable data stored in the agent, and the variable change prohibition period data stored in the variable change prohibition period storage unit is updated.

[0028] Therefore, the agent can periodically check his / her improvement level, change the variable based on his / her latest improvement status, and as a result, the entire set of agents reaches the solution earlier. be able to.

[0029] Further, the agent of the present invention is configured such that the first constant time interval is set smaller than the second constant time interval.

[0030] Therefore, the frequency at which the agent checks its improvement level can be lower than the frequency at which the agent sends and receives the improvement level of the neighboring agent, and the improvement level can be checked based on the latest situation of the neighboring agent. As a result, the entire set of agents can reach the solution faster.

[0031] Further, the agent of the present invention provides a variable change indicating a period during which change of variable data is prohibited. The prohibited period is set to be k times the second fixed time interval (k is an integer).

[0032] For this reason, it becomes possible for neighboring agents to judge the degree of improvement of variable data in a state where the change of variable data is prohibited after the agent has changed its own variable data. It is possible to resolve constraint violations with many agents that violate only, and to reduce the possibility of falling into an infinite loop with only some agents, and as a result, the entire set of agents can reach the solution faster.

[0033] Further, in the agent of the present invention, when the constraint solving unit determines whether or not to change the variable data, improvement of the neighboring agent stored in the improvement degree data of the own agent and the neighboring state storage unit. Comparing with ease data, there are at least the greatest number of possible improvements, cases where there is no agent that can change the variable data by violating the constraint other than its own agent, and when the number of constraint violations is the largest. It is configured so that it is determined that the variable data will be changed when one of the cases where the number of constraints is the smallest.

[0034] Therefore, it is possible to determine whether or not the agent should change the variable data based on a plurality of degrees that the variable data satisfies the constraint data. As a result, the degree to which the entire set of agents reaches the solution can be adjusted.

[0035] In addition, the agent of the present invention is configured such that, when changing the variable data, the constraint solving unit selects a variable value satisfying the constraint from the constraint data and updates the variable data in the variable storage unit. .

[0036] Therefore, the agent can select a value from which the constraint is removed from the combination list, and a value that can properly eliminate the constraint is selected reliably. As a result, the entire set of agents can be solved more quickly. Can be reached.

[0037] Further, in the agent of the present invention, the constraint resolution unit has the total ml of constraint violations in the latest third constant time interval of the improvement trend data and the number of constraint violations in the previous third constant time interval. When ml <mO, the current set value of the variable change prohibition period is shortened, and when ml ≥ mO, it is determined that there is no improvement trend. It is configured to update the variable change prohibition period data by increasing the current setting value of the change prohibition period. [0038] Therefore, the agent can adjust the variable change prohibition period of the variable data according to the actual improvement trend of the number of constraint violations, and prohibit the change of the variable only for the period adapted to the problem. As a whole set of agents can reach the solution faster

[0039] According to the agent of the present invention, by prohibiting the continuous change of a variable for a certain period under a certain condition, only a part of the agents does not try to resolve the constraint violation, and many By solving the constraint violation with the agent, the set of agents reaches the solution faster as a result of falling into the local optimal solution or infinite loop.

[0040] In an agent that obtains a solution in cooperation with a plurality of agents asynchronously, the agent stores a variable storage unit that stores variable data indicating a current value of the solution to be obtained, and variable data and neighboring agents store the variable data. A constraint storage unit that stores constraint data indicating a combination of values with variable data, a variable change prohibition period storage unit that stores a period during which variable data is prohibited to be changed, a variable change prohibition period storage unit that stores data, and variable data of the own agent An easy-to-improvement generator that generates easy-to-improvement data indicating the degree of ease for satisfying the constraint data, and a communication unit that transmits and receives variable data and the easy-to-improvement data generated by the easy-to-improvement generator A proximity situation storage unit that stores improvement degree data and variable data obtained from neighboring agents, and an improvement degree generation unit. Compared to the improvement ease data of neighboring agents stored in the neighborhood situation storage unit and the improvement ease data created, the neighborhood stored in the neighborhood situation storage unit according to the variable change prohibition period data Constraint resolution by changing the variable data to a value that satisfies the combination of constraint data so that the constraint violation with the agent's variable data is resolved, and setting the variable change prohibition period data to a predetermined period A part.

[0041] The distributed constraint satisfaction method is a method in which each agent has variable data indicating the current value of a solution to be obtained, constraint data indicating a combination of variable data and variable data of neighboring agents, and variable data. In the distributed constraint satisfaction method, multiple agents cooperate to obtain a solution of variable data in which all constraint relationships between variable data are established, each of which has variable change prohibition period data indicating a period during which the change of the value of the variable is prohibited. Each agent can easily improve its own variable data to show the degree of ease for satisfying the constraint data A generation step for generating degree data, a step for asynchronously transmitting / receiving variable data and the improvement degree data generated in the generation step to each neighboring agent, and the variable data and improvement degree data as own improvement degree data Compared with the improvement ease data of each neighboring agent, the decision step for determining whether or not to change its own variable data, and its own variable data when it is determined to change its own variable data at the decision step Change the data to a value that satisfies the combination of constraint data and notify neighboring agents. Change step and variable change prohibition period when variable data is changed in the change step. Change prohibition to set data to a predetermined period. Steps.

Brief Description of Drawings

FIG. 1 is a configuration diagram of an agent according to a first embodiment of the present invention.

FIG. 2A is a diagram showing a structure of variable data stored in the variable storage unit of the agent according to the first exemplary embodiment of the present invention.

FIG. 2B is a diagram showing a structure of constraint data stored in the constraint storage unit of the agent according to the first exemplary embodiment of the present invention.

FIG. 2C is a diagram showing a structure of ease of improvement data stored in the neighborhood state storage unit of the agent according to the first exemplary embodiment of the present invention.

FIG. 2D is a diagram showing a structure of variable change period data stored in the variable change prohibition period storage unit of the agent according to the first exemplary embodiment of the present invention.

FIG. 3 is a flowchart showing the operation of the agent according to the first exemplary embodiment of the present invention.

FIG. 4 is a diagram showing a setting example of a variable change prohibition period of the agent according to the first embodiment of the present invention.

FIG. 5 is a flowchart showing determination processing for changing an agent variable according to the first exemplary embodiment of the present invention.

FIG. 6 is a block diagram of the task assignment device according to the first exemplary embodiment of the present invention.

FIG. 7 is a diagram showing a task problem restriction network in which the task assignment device according to the first exemplary embodiment of the present invention creates a plan.

[FIG. 8A] FIG. 8A is a diagram showing a constraint network in the initial state of the task assignment device according to the first embodiment of the present invention. It is a figure which shows a network.

[8B] FIG. 8B is a diagram showing the constraint network in the planning state of the task assignment device according to the first exemplary embodiment of the present invention.

[8C] FIG. 8C is a diagram showing a constrained network in the final state of the task assignment device according to the first exemplary embodiment of the present invention.

FIG. 9 is a diagram showing a configuration of an agent according to the second embodiment of the present invention.

FIG. 10 is a flowchart showing the operation of the agent according to the second exemplary embodiment of the present invention.

FIG. 11A is a diagram showing a structure of agent improvement tendency data according to the second embodiment of the present invention.

[FIG. 11B] FIG. 11B is a diagram for explaining the storage operation of agent improvement tendency data according to the second embodiment of the present invention.

[FIG. 11C] FIG. 11C is a diagram for explaining the operation of determining the improvement tendency of the agent according to the second embodiment of the present invention.

12) FIG. 12 is a flowchart showing an operation of adjusting the variable change prohibition period of the agent according to the second embodiment of the present invention.

FIG. 13 is a configuration diagram of a schedule adjustment apparatus equipped with an agent according to the second embodiment of the present invention.

[14] FIG. 14 is a diagram for explaining a scheduling problem in which the schedule adjusting apparatus according to the second embodiment of the present invention creates a plan.

[15] FIG. 15 is a diagram showing a schedule problem restriction network in which the schedule adjustment apparatus according to the second embodiment of the present invention creates a plan.

[16] FIG. 16 is a diagram for explaining an initial operation of the schedule adjusting apparatus according to the second embodiment of the present invention.

FIG. 17 is a diagram for explaining the operation after TP adjustment of the schedule adjustment device according to the second exemplary embodiment of the present invention.

FIG. 18 is a configuration diagram of an agent according to the third embodiment of the present invention.

[19] FIG. 19 is a collaborative work robot equipped with the agent according to the third embodiment of the present invention. FIG.

20] FIG. 20 is a diagram for explaining a movement problem in which the movement planning apparatus according to the third embodiment of the present invention creates a plan.

21] FIG. 21 is a diagram showing a movement problem restriction network in which the movement planning apparatus according to the third embodiment of the present invention creates a plan.

FIG. 22] FIG. 22 is an explanatory diagram of the mobile operator of the cooperative work robot according to the third embodiment of the present invention.

[23] FIG. 23 is a diagram for explaining an initial state of the movement planning apparatus according to the third embodiment of the present invention.

24] FIG. 24 is a diagram for explaining an intermediate state of the movement planning apparatus according to the third embodiment of the present invention.

25] FIG. 25 is a diagram for explaining the final operation of the movement planning apparatus according to the third embodiment of the present invention.

FIG. 26 is a diagram showing an example of a problem constraint network used in the experiment.

FIG. 27 is a diagram showing a comparison of the average number of cycles until reaching the solution of the experimental result.

[FIG. 28] FIG. 28 is a diagram showing comparison of solution arrival rates of experimental results.

[29] FIG. 29 is a block diagram of a computer constituting a conventional constraint satisfaction device.

[30] FIG. 30 is a diagram for explaining a constraint network expressing a constraint satisfaction problem. [31A] Figure 31A shows a constraint network showing an example of the initial state of the constraint satisfaction problem.

[31B] Figure 31B is a diagram showing a constraint network showing an example of a state in which the solution of the constraint satisfaction problem is fixed.

圆 32] Fig. 32 is a diagram for explaining a constraint network diagram expressing the distributed constraint satisfaction problem

FIG. 33 is a diagram for explaining an example of a dispersion constraint satisfaction problem.

Explanation of symbols

100, 100a, 100b, 100c, 200, 200a, 200b, 200c, 200n, 700 agents 101 Variable storage

102 Constraint memory

103 Neighborhood Status Memory

104 Variable change prohibition period storage section

105, 205 Constraint Resolution Department

106 Communication Department

107 Improvement level generator

208 Improvement Trend Memory

210 Improvement trend data

701 Plan coordinate storage

703 Neighborhood Plan Coordinate Storage Unit

800, 811, 812 Robot

801 External detector

802 Moving part

803 Collision avoidance part

813 Obstacle

814, 815 goals

900 calculator

BEST MODE FOR CARRYING OUT THE INVENTION

[0044] (Embodiment 1)

Hereinafter, a task assignment device in which a plurality of agents according to the first embodiment of the present invention are connected will be described with reference to the drawings. The agent according to the first embodiment prohibits the change of the variable continuously for a certain period under a certain condition, so that it is possible to eliminate the constraint violation of many agents by simply eliminating the constraint violation of some agents. The solution is to eliminate the reaction, and as a result of falling into the local optimal solution, the entire set of agents reaches the solution faster.

First, the configuration and operation of the agent according to the first embodiment will be described. FIG. 1 is a configuration diagram of an agent according to the first embodiment of the present invention. In Figure 1, The agent 100 includes a variable storage unit 101, a constraint storage unit 102, a variable change prohibition period storage unit 104, an improvement ease generation unit 107, a communication unit 106, a neighborhood situation storage unit 103, and a constraint resolution unit 105. ing.

[0046] The variable storage unit 101 stores variable data indicating the current value of the solution to be obtained. The constraint storage unit 102 stores constraint data indicating a combination of values of variable data and variable data stored in neighboring agents. The variable change prohibition period storage unit 104 stores variable change prohibition period data indicating a period during which variable data change is prohibited. The improvement degree generation unit 107 generates improvement degree data indicating the degree of ease with which the variable data of the own agent satisfies the constraint data. The communication unit 106 transmits / receives the variable data and the improvement degree data generated by the improvement degree generation unit 107 to / from neighboring agents. The neighborhood situation storage unit 103 stores neighborhood situation data composed of improvement ease data and variable data acquired from neighboring agents. The constraint resolving unit 105 compares the improvement ease data generated by the improvement degree generation unit 107 with the improvement ease data of the neighboring agent stored in the neighborhood state storage unit 103, and sets it as variable change prohibition period data. In response, the variable data is changed to a value satisfying the combination of the constraint data so as to eliminate the constraint violation with the neighboring agent variable data stored in the neighborhood status storage unit 103. At the same time, the constraint resolution unit 105 sets the variable change prohibition period data to be a predetermined period.

Next, the structure of data stored in the agent 100 will be described. FIG. 2A is a diagram showing a structure of variable data stored in the variable storage unit 101 of the agent according to the first exemplary embodiment of the present invention. In FIG. 2A, the variable data 121 stored in the variable storage unit 101 has a table structure in which variable names and their values are records. Figure 2A shows that home agent A has one variable X and its value is 1.

FIG. 2B is a diagram showing a structure of constraint data stored in the constraint storage unit 102 of the agent according to the first exemplary embodiment of the present invention. In FIG. 2B, the constraint data 122 stored in the constraint storage unit 102 has a table structure in which a record indicating a combination of a variable name of its own variable name, a neighboring agent variable name, and a possible value of those values is a record. . In Fig. 2B, self-agent A has a constraint related to variable X as a constraint related to variable X. Indicates that there is a constraint that the value combination (x, y) must be (1, 1) or (2, 2) or (3, 3). It is also shown that there is a constraint that the value combination force 1) or (2, 2) or (3, 3) with the variable y of agent C. An agent that has a variable with respect to this variable X is called a neighbor agent.

FIG. 2C is a diagram showing a structure of variable change period data stored in the variable change prohibition period storage unit 104 of the agent according to the first exemplary embodiment of the present invention. In FIG. 2C, the variable change prohibition period data 123 stored in the variable change prohibition period storage unit 104 includes a variable name, time information indicating the start and end times of the period during which the change of the variable value is prohibited. Is a table structure in which is a record. In FIG. 2C, it is shown that the value of variable X of own agent A is prohibited until 00:00:00.

[0050] FIG. 2D is a diagram showing a structure of ease of improvement data stored in the neighborhood state storage unit 103 of the agent according to the first exemplary embodiment of the present invention. In FIG. 2D, the neighbor situation data stored in the neighbor situation storage unit 103 has a table structure in which the variable names and values of neighboring agents that have a constraint relationship with the variable data and the improvement ease data 125 are records. The In addition, the ease-of-improvement data 125 includes the constraint number 1 indicating the number of constraint data, the constraint violation number m indicating the number of variable data that violates the constraint data, and the restriction data when the variable data value is changed. It consists of a number n that can be improved to indicate the number of constraint data that can resolve the violation state.

[0051] For variable X of own agent A and neighboring agent B that has constraints, the value of variable y is 2, and the number of constraints 1 in the improvement ease data 125 is 1, the number of constraint violations m is 1, and can be improved It shows that the number n is 1. Similarly, for the neighboring agent C, the value z of the variable z is S3, and the number of constraints 1 in the improvement ease data 125 is 1, the number of constraint violations m is 1, and the number of possible improvements _n force Si. . In FIG. 2D, the self-agent improvement ease data 125 is also configured to store the self-agent A improvement ease data 125 in a separate storage means. I don't mind!

Next, the operation of the agent will be described. FIG. 3 is a flowchart showing the operation of the agent according to the first exemplary embodiment of the present invention. In Fig. 3, the time interval T1 (timer event Evl) for sending the improvement ease data to the neighboring agent as a parameter. Then, the time interval T2 (timer event Ev2) for improving operation is used. This time interval T1 is an example of a first constant time interval. The time interval T2 is an example of a second constant time interval.

[0053] First, the constraint resolution unit 105 generates a timer event Evl activated at a time interval T1, a timer event Ev2 activated at a time interval T2, and a message reception event Ev3 by receiving a message from a neighboring agent. Eventually it waits until one event occurs (step S401). Next, the constraint solving unit 105 instructs the start of the next process according to the type of event that has occurred (step S402).

A case where the event is a timer event Evl will be described. The improvement ease generation unit 107 determines whether or not the change of the value of the variable is within the period during which the change of the variable value is prohibited from the variable change prohibition period data stored in the variable change prohibition period storage unit 104. If it is out of the period, the improvement ease generation unit 107 uses the variable data stored in the variable storage unit 101 and the constraint data stored in the constraint storage unit 102 to have 1 constraint, m constraint violations, The improvement ease data composed of the improvement possible number n is generated, and if it is within the period, it is stored that the value of the variable cannot be changed (step S403). At this time, the improvement ease generation unit 107 generates the attribute value of the improvement ease data for the variable data stored in the variable data table 121 as follows. The improvement level generation unit 107 calculates the total number of records including the variable name of the variable data from the constraint data 122 stored in the constraint storage unit 102 and sets the number of constraints to 1. And the variable data of neighboring agents are included in the value combination, and the total number of ヽ records is calculated as the constraint violation number m. The improvement ease generation unit 107 further includes that the variable data value is not stored in the variable change prohibition period data 123 in the record, and the value of the constraint data is changed by changing the variable data value. The total number of records that can be combined is calculated, and the number that can be improved is n. The time interval T1 is preferably shorter than the time interval T2 described later. Neighboring agents can always be notified of the latest situation where the time interval T1 is sufficiently short.

[0055] Subsequently, the communication unit 106 transmits the improvement ease data to all of the neighboring agents by message communication, and returns to step S401. However, the communication unit 106 uses the If the change of the variable is prohibited and it is determined that the value of the variable cannot be changed within the specified period, a message indicating that the change cannot be made is transmitted (step S404).

[0056] Next, the case where the event is the message reception event Ev3 will be described. The communication unit 106 receives a message from a nearby agent (step S405). The communication unit 106 stores the variable data and ease of improvement data of neighboring agents included in the received message in the neighboring state storage unit 103, and returns to step S401 (step S406).

Next, a case where the event is a timer event Ev2 will be described. The improvement ease generation unit 107 determines whether the change of the variable value is prohibited from the variable change prohibition period data 123 stored in the variable change prohibition period storage unit 104, similarly to the processing in step S403. Further, the ease-of-improvement generation unit 107 uses the variable data 121 stored in the variable storage unit 101 and the constraint data 122 stored in the constraint storage unit 102 to obtain a constraint number 1, a constraint violation number m, and an improvement possible number n. Generated improvement ease data (step S407). However, the neighbor agent name in the self-improvement degree data 125 of the self agent can be distinguished from the data of other agents as a value representing the self agent.

[0058] Next, the constraint solving unit 105 records the improvement degree data 125 generated by the improvement degree generation unit 107 in step 407 in the neighborhood state storage unit 103, and the improvement degree of each neighboring agent. Compare with data 125 to determine if the value of the variable should be changed. If it is determined to be changed, the process proceeds to step S409. If it is determined not to change, this event processing is terminated, and the process returns to step S401 (step S408). Details of this determination method will be described later.

[0059] Subsequently, the constraint resolution unit 105 selects one of the variables that can be taken by the agent from the possible combinations of the constraint data 122 stored in the constraint storage unit 102, and obtains the value of the variable. The current value of the variable data 121 stored in the variable storage unit 101 is updated. As a result, the ease-of-improvement generation unit 107 generates the ease-of-improvement data with the updated variable data values in the same manner as the processing in step S403, and the communication unit 106 generates the variable data and the ease of improvement generated. The degree data is notified to each neighboring agent by a message (step S409).

[0060] Furthermore, the constraint resolution unit 105 stores the variable change prohibition period in the variable change prohibition period storage unit 104 so that the value of the variable whose value has been changed in step S408 cannot be changed for a certain period. The current time is set as the start time, the time after a lapse of a certain period is stored as the end time, and the process returns to step S401 (step S410).

In step S410, in the present embodiment, the variable is set to be prohibited from changing for a certain period, and the period at this time is k times the time interval T2 (k is an integer). For example, k = 2, that is, 2 times. Also, when this period has passed, the prohibition setting of the variable change prohibition period storage unit 104 is canceled.

It should be noted that this operation flow operates asynchronously independent of neighboring agents. For this reason, messages are sent at any time between neighboring agents that operate asynchronously between agents. In the present embodiment, when a message is received, the message reception event Ev3 is stored in the event queue by the event processing mechanism, and even if a step other than step S401 is being processed, the process proceeds to step S401. This is so that events can be detected. Even if a timer event occurs during message processing in step S405, it can be detected and processed in the same way.

Next, a method for setting the variable change prohibition period in step S410 will be described. In this embodiment, the method of prohibiting any change in the value of a variable (prohibition method 1) is used.However, there is a method of prohibiting a variable from being set to a certain value (prohibition method 2) or a certain value strength. It can also be realized as a method that prohibits changing to a value (prohibition method 3), or a method that prohibits the above combinations or other specific changes (prohibition method 4).

[0064] With this prohibition method, for example, the range of variable X, that is, the set of values taken by variable X is set to {1, 2, 3}, and the value of variable X is changed from "1" to "2" In this case, the operation during the prohibition period according to the prohibition method described above is as follows.

In the prohibition method 1, changing the value of the variable X is completely prohibited, and the value of the variable X cannot be changed for a certain period. The variable change prohibition period is set for each variable.

[0066] In the prohibition method 2, a force that prohibits the return of the value of the variable X to “1” for a certain period or a change of the value of the variable X to “2” is prohibited for a certain period. In this case, changing the value of variable X to “3” is not prohibited even during the prohibited period. The variable change prohibition period is set for each variable value.

[0067] In prohibition method 3, change of the value of variable X from "1" to "2" is prohibited. In this case, "1" or Even during the period when the change from “1” to “2” is prohibited, the change from “1” to “3” and the change from “3” to “2” are not prohibited. The variable change prohibition period is set by distinguishing the direction of variable change.

In prohibition method 4, change of the value of variable X from “3” to “2”, “2” to “1”, and “1” to “3” is prohibited. In this case, the value is limited to changes in the ascending order, such as “1” to “2”, “2” to “3”, and “3” to “1”.

[0069] Next, an example of setting a variable change prohibition period of an agent that works according to the present embodiment will be described. FIG. 4 is a diagram illustrating a setting example of the variable change prohibition period of the agent according to the first embodiment of the present invention. In FIG. 4, after changing the value of variable A from “1” to “2”, a variable change prohibition period 151 is set (at time 153) to prohibit the change from “1” to “2”. Furthermore, during the variable change prohibition period 151, the value of variable A is changed from “2” to “1”, and variable change prohibition period 152 from variable “2” to “1” is set (at time 154). ) Each variable change prohibition period is set independently, and even if it is a variable change prohibition period from `` 1 '' to `` 2 '', it is a variable change prohibition period from `` 2 '' to `` 1 ''. Otherwise, changing from “2” to “1” is not prohibited. After the variable change prohibition period ends, the prohibition setting is canceled (at time 155 and time 156).

Next, the details of the determination processing of the constraint solving unit 105 that changes the agent variables that are relevant to the present embodiment will be described. FIG. 5 is a flowchart showing the determination processing of the constraint solving unit 105 that changes the agent variable according to the first embodiment of the present invention. The determination process shows the detailed operation of step S408 in the flowchart showing the operation of the agent in FIG. In FIG. 5, the constraint solving unit 105 always determines that the value of the variable is not changed (S509), determines that the value of the variable is changed (S510), and ends.

[0071] In step S501, the constraint resolution unit 105 sets the variable data value of the constraint data from the variable data 121 stored in the variable storage unit 101 and the constraint data 122 stored in the constraint storage unit 102. In the combination, the record is searched to determine whether there is any variable data that violates the constraint. If there is no variable data that violates the constraint, the constraint resolution unit 105 determines that the variable is not changed (S509), and ends.

[0072] In step S502, the variable force variable violates the constraint examined in step S501. If the variable change prohibition period data 123 stored in the change prohibition period storage unit 104 is included in the prohibition period and the change of the variable is prohibited, the constraint resolution unit 105 must change the value of the variable. Judge (S509) and end.

[0073] In step S503, the constraint resolving unit 105 calculates the ease of improvement data composed of the number of constraints 1, the number of constraint violations m calculated by the improvement ease generator 107, and the number of possible improvements n, and the neighborhood status memory. The improvement ease data 125 of the neighboring agent stored in the part 103 is compared. If the possible improvement number n is greater than the possible improvement number n of any neighboring agent, the constraint solving unit 105 determines to change the value of the variable (S510), and ends. On the contrary, if the improvement possible number n is smaller than the improvement possible number n of any neighboring agent, the constraint solving unit 105 determines that the variable is not changed (S509), and ends. In cases other than the above, that is, when the improvement possible number n is the same as the maximum improvement number n of neighboring agents, the process proceeds to step S504. At this time, the case where the improvement possible number n is 0 is also included with all neighboring agents.

[0074] In step S504, the constraint solving unit 105 includes the ease of improvement data composed of the number of constraints 1, the number of constraint violations m, and the number of possible improvements n calculated by the ease of improvement generation unit 107 and the neighborhood status storage unit 103. From the data on the ease of improvement of the neighboring agents stored in (the number of constraints 1, the number of constraint violations m, the number of possible improvements _n ) 125 Check if there is an agent that can change the variable from the number n that can be improved. If the target number of agents is 0, the constraint resolution unit 105 determines to change the variable (S510), and the process ends. If the number of target agents is 1 or more, the process proceeds to step S505.

[0075] In step S505, the constraint resolution unit 105 compares the number of constraint violations m with the number of constraint violations m of the agent targeted in step S504. Determines that the value of the variable is to be changed (S510) and ends. If the number of constraint violations of any neighboring agent is smaller than m, it is determined that the variable should not be changed (S509), and the process ends. If the constraint violation number m is the same as the maximum constraint violation number m of the neighboring agent, the process proceeds to step S506.

[0076] In step S506, the constraint resolution unit 105 compares the constraint number 1 with the constraint number 1 of the agent targeted in step S504, and the constraint number of any neighboring agent is small. If so, it is determined to change the variable (S510), and the process ends. If the number of constraints of any neighboring agent is too large, it is determined that the variable is not changed (S509), and the process ends. If the restriction number 1 is the same as the minimum one of the neighboring agent restriction numbers 1, the process proceeds to step S507.

[0077] In step S507, the constraint resolving unit 105 includes an agent that includes the agent and the agent that is the target in step S504. Determine.

In step S508, if the constraint solving unit 105 determines that the self agent is to change the variable, the constraint solving unit 105 determines to change the variable, and ends. If it is not determined that the variable is to be changed, it is determined that the variable is not changed, and the process ends.

[0079] Note that the probabilistic determination of agents in step S508 may be a method of calculating according to data that agents can refer to in common, such as time information and ranking among agents.

[0080] Further, in this probabilistic determination method, it is assumed that both variables are determined with an average probability distribution with no bias in the determination result, and at least one of the variables is continuously excellent for a long time on average. It shall not be changed first. For example, the case where only the wrong judgment method is used in which the node with the younger node name is always changed with priority is excluded.

[0081] In addition, according to the processing procedure for determining whether or not it is sufficient to change a variable, in the present embodiment, an example of the power used as the processing procedure from step S501 to step S508 is shown. The types of steps to be judged and the order of judgment are not limited to this.

Next, a description will be given of an example in which a plan is created by a task assignment device in which a plurality of agents that are useful for the present embodiment are connected. FIG. 6 is a configuration diagram of the task assignment device according to the first exemplary embodiment of the present invention. In FIG. 6, the agent 100a, agent 100b,..., Agent 100c in the task assignment device 300 are connected via a wired or wireless network and can communicate with each other. In the present embodiment, agent 100, agent 100b,..., And agent 100c have the same configuration as agent 100 shown in FIG. If agent 100a is its own agent, agent 100b and the like are neighboring agents. In addition, variable storage unit 101, constraints The storage unit 102 and the variable change prohibition period storage unit 104 will be described assuming that initial values necessary for the operation are set in advance, but an initial setting device for setting the initial value of each agent is provided separately through the network. It can also be implemented as a configuration set for each agent.

[0083] Next, a specific operation of the task assignment device in which a plurality of agents related to the present embodiment are connected will be described. Here, multiple agents that take charge of either “black task” or “white task” each have one variable that indicates the type of task, and the initial value of state power. An embodiment will be described in which task allocation plans are created in cooperation with each other asynchronously.

FIG. 7 is a constrained network diagram in which the task assignment device according to the first exemplary embodiment of the present invention creates a plan. In FIG. 7, each node shows a variable xl of agent XI, a variable X 2 of X2,..., A variable x7 of X7. Each node takes one of the values “black” indicating “black task” and “white” indicating “white task”. In addition, each arc represents a constraint between each variable. Here, it is assumed that neighboring agents connected by an arc cannot handle the same task. In other words, nodes connected by constraints cannot have the same value. For example, node xl and node x4 must not be “white” or “black” at the same time. In the state shown in Figure 7, the values of node x4 and node x5 are “black” at the same time, and a constraint violation has occurred.

Note that FIG. 7 shows the constraint relationship between variables for each agent in the task assignment device, and does not show an actual network connection configuration for connecting the agents. However, as shown in Fig. 6, it is not always necessary to configure communication paths between all agents. It is only necessary to configure communication paths between agents that have at least a variable in a constraint relationship.

[0086] Next, an example will be described in which the restriction between the variables of each agent is eliminated by a task assignment device in which a plurality of agents that are useful for the present embodiment are connected. FIG. 8A is a constrained network diagram of an initial state of the task assignment device according to the first exemplary embodiment of the present invention. FIG. 8B is a constraint network diagram in a planning state of the task assignment device according to the first exemplary embodiment of the present invention. FIG. 8C shows the final state of the task assignment device according to the first exemplary embodiment of the present invention. FIG. In FIG. 8A to FIG. 8C, the ease-of-improvement data for each node is shown as “(number of improvement possible nZ constraint violation number mZ constraint number 1)” after the symbol xj representing each node. In addition, here, the decision whether to change the variable or not is changed when the number n that can be improved is larger. If this is the same, the one with the larger number of constraint violations m changes, and if this is the same, the agent with the smaller number of constraints 1 changes the value of the variable. In addition, regarding the method of prohibiting variable changes, change of variable values is prohibited.

[0087] FIG. 8A shows an initial state. Here, a constraint violation occurs between node x4 and node x5. Nodes xl, x2, x3, x6, and x7 do not violate the constraint and the number of constraints is 1, so all the ease of improvement data is (OZOZ1). Node χ4 has one constraint violation with node χ5, and if node χ4 changes to white, the constraint violation with node χ5 is resolved, but the constraint between nodes xl, χ2, and χ3 Since a violation occurs, the number η that can be improved (in this case, a number greater than or equal to 0) is zero. Therefore, the improvement ease data of the node χ4 is (0/1/4). Similarly, the improvement ease data of node χ5 is (OZ1Z3). This information is exchanged between neighboring agents through step S403, step S404, step S405, and step S406 in FIG.

[0088] Constraint violation node x4 and variable x5 have almost the same conditions Force Constraint number 1 is smaller in node x5, so node x4 is determined not to change value, and node x5 changes value It is determined to be. Therefore, the node x5 changes the value to “white” and becomes the state shown in FIG. 8B.

[0089] In the state of FIG. 8B, node x5 can improve two constraint violations between node x6 and node x7 by changing the value from "white" to "black". Since this is a violation, the number of possible improvements 1 is 1. Therefore, the ease of improvement data for node x5 is (1Z2Z3). The improvement degree data of nodes x6 and x7 are both (1Z1Z1). The number of possible improvements n is the same, but the number of constraints 1 is larger for node x5. If the value of node x5 is changed from “white” to “black” here, it returns to the state shown in Fig. 8A and falls into an infinite loop of the local optimal solution.

[0090] However, node x5 is in the variable change prohibition period because the value was changed earlier, and the value cannot be changed. Therefore, node x6 and node x7 change their values in step S504. It is determined. Since node x6 and node x7 are not directly connected, both can change the value at the same time.

[0091] Finally, variables x6 and x7 change their values, resulting in the state of FIG. 8C. Figure 8C shows the final state with all constraints satisfied.

[0092] In this way, all agents were assigned to satisfy all white, black, and black task power constraints. In other words, all the constraints were satisfied and a task allocation plan was created by the task allocation device.

Note that the agent in the present embodiment exchanges information with neighboring agents at the time interval T1 and improves the state at the time interval T2. However, these operations are executed at the same timing. May be. In other words, the information exchange with the neighboring agent and the judgment of improvement may be operated continuously.

[0094] By adopting such a configuration, in this embodiment, the agent is prohibited from changing the variable continuously, and the same variable cannot be changed to the same value for a certain period. As a result, the possibility of falling into a local optimal solution decreases, and as a result, the solution can reach the solution faster as a whole.

[0095] (Embodiment 2)

Next, a schedule adjustment device connecting a plurality of agents according to the second embodiment of the present invention will be described.

[0096] The agent working on the present embodiment eliminates restriction violations of some agents by prohibiting continuous change of variables in a period corresponding to the resolution situation of variable constraint violations. The goal is for the entire set of agents to reach the solution even faster as a result of the local optimal solution and the infinite loop.

First, the configuration and operation of the agent according to the second embodiment will be described. FIG. 9 is a configuration diagram of an agent according to the second embodiment of the present invention. In Fig. 9, Agent 200 stores the improvement trend data indicating the past transition of the degree to which the force variable data that has almost the same configuration as Agent 100 shown in Fig. 1 meets the constraint data. 1 is different from FIG. 1 in that a trend storage unit 208 is further provided. Further, when the constraint resolution unit 205 changes the value of the variable data stored in the variable storage unit 101, the latest improvement trend. 1 is different from FIG. 1 in that the direction data is generated and stored in the improvement trend storage unit 208, and the length of the variable change prohibition period of the variable data is adjusted according to the improvement trend data. Further, the improvement trend data stored in the improvement trend storage unit 208 is a table structure of transition data of the variable name and the number of constraint violations m of the variable. The improvement trend data is history data of the number of constraint violations including the past several times. Details of the data structure of past changes will be described later.

Next, the operation of each agent will be described. FIG. 10 is a flowchart showing the operation of the agent according to the second exemplary embodiment of the present invention. In FIG. 10, the operation of the agent 200 is different from that in FIG. 3 in that force steps S601 and S602 that are substantially the same as the steps of the agent 100 shown in FIG. 3 are added. In addition to the time interval T1 and the time interval T2, the time interval T4, which is a period for determining the improvement tendency of variables, and the current set value TP of the variable change prohibition period are used as parameters. This time interval T4 is an example of a third constant time interval. The time interval T4 is a period corresponding to several times the time interval T2, but in the present embodiment, the time interval T4 is described as being set to 5 times the time interval T2. In addition, it is desirable that the current set value TP of the variable change prohibition period is k times the time interval T2 (k is an integer) because of the relationship between the synchronization timing of the variable update operation and the determination operation. The set value in will be described later.

[0099] In step S601, the constraint solving unit 205 records the number of constraint violations m calculated by the ease-of-improvement generation unit 107 in the improvement trend storage unit 208 as current improvement trend data. The number of constraint violations, m, is set so that a record of about twice the time interval T4 remains.

Here, the improvement trend data recording process will be described. In this embodiment, the constraint violation number m is used as improvement trend data. FIG. 11A is a diagram showing a structure of agent improvement tendency data according to the second embodiment of the present invention. In FIG. 11A, improvement trend data 210 has a structure that records the number of constraint violations m for past variables, and stores the current number of constraint violations m in the rightmost column. In the left column, the constraint violation number m recorded at the previous time is stored, and in the left column, the constraint violation number m recorded two times before is stored. In this way, it is twice as long as T4 time, that is, 10 times as long as T2. The period is recorded. FIG. 11B is a diagram explaining an agent improvement tendency data storage operation according to the second embodiment of the present invention. In Fig. 11B, when the number of constraint violations is recorded as new data in the improvement trend data 210, the previous records are shifted one by one, and the current number of constraint violations is recorded in the rightmost column. It is recorded to keep the time order according to the past number of constraint violations.

[0101] In step S602, the constraint solving unit 205 determines an improvement trend from the history of the number of constraint violations m recorded in the improvement trend storage unit 208. The constraint resolution unit 205 compares the total number of constraint violations in the latest time interval T4 with the total number of constraint violations in the previous time interval T4, and determines the improvement trend. This determination changes the value of the current setting value TP during the variable change prohibition period.

[0102] Here, the operation for changing the value of the current set value TP during the variable change prohibition period will be described.

[0103] FIG. 11C is a diagram illustrating an operation of determining an improvement tendency of an agent according to the second embodiment of the present invention. In FIG. 11C, in the improvement trend data 210, the total ml of the constraint violations in the recent time interval T4 and the total mO of the constraint violations in the previous time interval T4 are compared. If ml> mO, it is judged as “not improved”, and if ml ≦ mO, it is judged as “improved”. In response to this determination, the current set value TP of the variable change prohibition period is changed. Here, the total number of previous constraint violations is 6, the total number of recent constraint violations is 5, and the improvement trend is judged as “improved”.

Next, a detailed operation of the adjustment process of the constraint solving unit 205 that adjusts the current set value TP in the variable change prohibition period will be described. FIG. 12 is a flowchart showing an operation of adjusting the variable change prohibition period of the agent according to the second embodiment of the present invention. In Fig. 12, the current set value TP of the variable change prohibition period takes a value within a predetermined range from the minimum value TP-min to the maximum value TP-max.

[0105] In step S701, the constraint resolution unit 205 determines an improvement tendency from the number of constraint violations m stored in the improvement trend storage unit 208. If it is determined that the improvement is improved, the constraint resolution unit 205 proceeds to step S704. move on. If it is determined that “Improved !, NA! /,”, The process proceeds to step S702. In step S702, if TP is greater than TP-max, the process proceeds to step S703. Otherwise, exit without changing TP.

[0107] In step S703, TP is incremented by a predetermined value, and the process ends.

If TP> TP—min in step S704, the process proceeds to step S705. Otherwise, exit without changing TP.

[0109] In step S705, the predetermined value is reduced by the TP force and the process is terminated.

[0110] Next, an example in which a plan is created by a schedule adjustment device that connects a plurality of agents that are relevant to the present embodiment will be described. FIG. 13 is a configuration diagram of a schedule adjustment apparatus equipped with an agent that can make full use of Embodiment 2 of the present invention. In FIG. 13, the schedule adjustment apparatus 301 is composed of a plurality of agents 200a, 200b, 200c,.禾 IJ user 入出力 inputs / outputs the schedule to Agent 200a, User B to Agent 200b, User C to Agent 200c, and User n to Agent 200η via the input / output unit. Each agent adjusts the schedule with other users based on the entered schedule. For example, the user 入力 inputs his / her desired schedule to the schedule adjustment device on which the agent 200a is installed, and instructs the start of conference adjustment and schedule adjustment with other participants. Agent 20 Oa force S The plan created by coordinating with other agents 200b, 200c, and 200η is received as a schedule result from the schedule adjustment device.

[0111] Next, a specific operation of the schedule adjustment apparatus in which a plurality of agents that are useful for the present embodiment are connected will be described. Here, the three agent 200a, agent 200b, and agent 200c that adjust the conference date of user Α, user Β, and user C coordinate the first conference date of user A and user B. In addition, the operation for adjusting the second meeting date between User B and User at the same time will be described.

FIG. 14 is a diagram for explaining a schedule problem in which the schedule adjustment device according to the second embodiment of the present invention creates a plan. In Figure 14, as initial conditions, the settable days for each user are 1 and 2 days for 禾 IJ user A, 2 days, 3 and 4 days for 禾 IJ user B, and は IJ user C for 2 days and 3 days, and the initial setting of the desired date to hold each user's meeting is 2 days. Show. In creating this schedule plan, each user cannot set up two meetings at the same time. In addition, each user does not disclose the conference setting date to other users from the beginning from the viewpoint of privacy. Also, User B does not disclose to User C that he / she will set up a first meeting with User A. Similarly, User B does not disclose to User A that he will have a second meeting with User C.

FIG. 15 is a restriction network diagram of a schedule problem in which the schedule adjustment apparatus according to the second embodiment of the present invention creates a plan. In FIG. 15, each node shows variables of the agent 200a, the agent 200b, and the agent 200c. Agent 2 OOa has node x8 indicating the conference setting date of user A, and node x8 takes “1” or “2”. The agent 200b includes a node x9 indicating a date when the user B sets up a conference with the user A, and a node xlO indicating a date when the user B sets up a conference with the user C. Node x9 and node xlO each take one of the values “2”, “3”, and “4”. The agent 200c has a node xl l indicating the date when the user IJ C sets up a meeting with the user B. The node xl l takes “2” or “3”. The constraints between the nodes here are "x8 = x9", "x9 ≠ xl0", "xlO = xl l"

[0114] In the present embodiment, in order to explain the state in which the current set value TP of the variable change prohibition period is adjusted, the operation in step S408 for determining whether or not to change the variable value is performed. Do as follows. In the probabilistic decision process (step S507) when the improvement ease data values are exactly the same for each of the two nodes connected by the constraint, the node with the younger node name, for example, node x9, becomes node xlO. It shall be determined to change the value with priority. However, it is possible that such a younger node will be permanently changed in preference and may be judged in this way for a limited period of time. In the present embodiment, in order to explain the operation in such a state, it is determined that a large number of changes are biased toward one node at least in a period within twice the time interval T4. In addition, the initial value of the current set value TP of the variable change prohibition period of each variable is equivalent to the same time as the time interval T2. Furthermore, in this embodiment, the node value change prohibition method is to change the value back to the value before the change after changing the value. It is prohibited during the prohibited period. However, if there is a possible value other than the value before the change, the change to that value is not prohibited.

[0115] Next, an example of an operation for eliminating the constraint violation between variables of each agent while adjusting the current setting value TP of the variable change prohibition period with the schedule adjustment device connected to multiple agents according to the second embodiment explain. First, after starting the operation, it falls into the local optimal solution, and the improvement of the variable does not progress. The constraint resolution unit 205 determines “Improved, no,” and starts adjusting the current set value TP of the variable change prohibition period. The first operation example up to is described. FIG. 16 is a diagram for explaining an initial operation of the schedule adjusting apparatus according to the second embodiment of the present invention. In Fig. 16, each line shows the value of the variable at each time. In addition, the ease-of-improvement data is indicated after each variable value as “(number of improvement possible nZ constraint violation number mZ constraint number 1)”.

[0116] At time 0, that is, in the initial state, agent 200a, agent 200b, and agent 200c have the values of variable x8, variable x9, variable χ10, and variable xl l all set to "2" A constraint violation has occurred between variable x9 and variable xlO. Here, each agent generates improvement ease data, transmits and receives it, and determines whether to change the value of the variable. Agent 200b changes the value of x9, which has priority, to "3" because variable x9 and variable xlO have the same improvement ease data. Agent 200a and Agent 200c are determined not to change the value of the variable.

[0117] At time 1, the variable x9 can be changed to the force value "3" and the value "4", which are prohibited from changing to the value "2". In addition, there is a constraint violation between variable x8 and variable x9. Comparing the improvement degree data of variable x8 and variable x9, the number of improveable constraint violations n and the number of constraint violations m are the same. Since the number of constraints 1 is smaller in variable x8, agent 200a sets the value of variable x8 as `` It is determined to change to “1”. On the other hand, the agent 200b determines not to change the value of the variable x9.

[0118] At time 2, the variable x8 is prohibited from changing to the value "2". Since variable x8 needs to have a value of “l” or “2”, variable x8 is in a state where its value cannot be changed at all. In addition, the variable change prohibition setting for the variable x9 is cancelled, and the agent 200b determines to change the value of the variable x9. The variable x9 has a constraint violation number m of 1, If the value is changed to “2”, the number of constraint violations m increases to 2, so change it to “4”.

[0119] At time 3, the variable change prohibition setting for variable x8 is cancelled. As with time 1, agent 200a changes the value of variable x8 from “1” to “2”.

[0120] At time 4, since the variable x8 is set to prohibit variable change and the value cannot be changed, the agent 200b changes the value of the variable x9 from "4" to "3" as at time 2.

[0121] At time 5, time 5 is almost the same as time 1, and after this time, the operation of changing the values of variable x8 and variable x9 is continued and the constraint violation state is not resolved. . At this time, the variable x8 and the variable x9 are in a state of constraint violation power. Here, the constraint resolution unit 205 of agent 200a and agent 200b is judged as “not improved” from the improvement trend data, and agent 200a is currently in the variable change prohibition period for variable x8 and agent 200b for variable x9. Increase the set value TP. Agent 200c does not change the current set value TP of the variable change prohibition period regardless of the variable.

[0122] Next, after the current setting value TP of the variable change prohibition period has been adjusted following the first operation example, the local optimal solution escape is forcibly induced by the adjusted new value to be constrained. A second example of operation that resolves the violation will be explained. FIG. 17 is a diagram for explaining an operation example after the TP adjustment of the schedule adjustment apparatus according to the second embodiment of the present invention. In FIG. 17, as in FIG. 16, each row indicates the value of the variable at each time. The second operation example shows the operation example that follows the change of the TP of the variable x8 and variable x9 in the first operation example, but the initial values of each variable are easier to compare. Is the same as the initial value of the operation example shown in Fig. 16. That is, the variable change prohibition period is set for any variable !, and how the state force starts to operate and how it operates is explained. However, for the current setting value TP of the variable, the variable x8 and the variable x9 are set to a period twice the time interval T2, which is the setting value in the previous operation example shown in FIG. 16, that is, 2 XT2 time. .

At time n, similar to the operation at time 0 in FIG. 16, the agent 200a changes the value of the variable x 9 to “3”.

[0124] At time n + 1, agent 200b changes the variable x9 to the value "2". Set prohibited. The constraint violation between variable x8 and variable x9 has not been resolved. Restriction Change the value of variable x8 with a small number of constraints.

[0125] At time n + 2, the agent 200a changes the value of the variable x8 to "1" and prohibits the change to the value "2". Since the value of the variable x8 needs to be “1” or “2”, the agent 200a cannot change the value of the variable x8 at all. Since the constraint violation between the variable x8 and the variable x9 has not yet been resolved, the agent 200b changes the value of the variable x9 from “3” to “4” in the same manner as the operation at time 2 in FIG. . Variable x9 is changed to the value “4” because changing to the value “2” is prohibited.

[0126] At time n + 3, the constraint violation between variable x8 and variable x9 is resolved. Since the change of the value of the variable x8 is still prohibited, the agent 200b determines to change the value of the variable x9. When variable x9 is changed to value “2”, the number of constraint violations increases to 2, but change to value “3” is prohibited, so change to value “2”.

[0127] At time n + 4, constraint violations occur between variable x8 and variable x9 and between variable x9 and variable xlO, respectively. The agent 200b cannot change the value of the variable x9 because the change to the value “3” and the change to the value “4” are both prohibited. However, the agent 200a changes the value to “2” because the constraint violation can be resolved by changing the value of the variable x8. The agent 200b changes the value of the variable xlO from “2” to “3” because the change of the value of the variable x9 that violates the constraint in the vicinity of the variable xlO is prohibited.

[0128] At time n + 5, the constraint violations between variable x8 and variable x9 and between variable x9 and variable xlO are resolved. A constraint violation has occurred between the variable xlO and the variable xl l, but if the variable xl l changes its value, the constraint violation can be resolved, so Agent 200c changes the value of variable xl 1 from “2” to “3” To "".

[0129] At time n + 6, all the constraints between the variables are satisfied, and the schedule adjustment device ends its operation.

[0130] The above two operation examples have been described. In the first operation example shown in Fig. 16, at time 5, the state determined to change the younger one of the variable names stochastically continued, and temporarily entered a local optimal solution loop. However, by adjusting the value of the current set value TP during the variable change prohibition period in the processing after time 5, the process moved to the second operation example. Fig. 17 In the second operation example shown in Fig. 1, the values of the variables xlO and xl l should be changed. By trying to solve this by changing only the variables X 8 and x9, the local optimal solution is temporarily I fell into. However, since the variable change prohibition period was adjusted to be longer than in the first operation example, the variable xlO changed its value and the local optimal solution escaped, completing the operation at time n + 6.

[0131] With this configuration, in this embodiment, the agent is prohibited from changing the variable continuously during the period according to the resolution state of the constraint violation, and may fall into a local optimal solution. In addition, it is possible to provide a distributed constraint satisfaction device that further reduces the possibility that some agents fall into an infinite loop, and as a result, the entire set of agents can reach the solution more quickly.

Note that, more specifically, the agent according to the second embodiment is configured by computer hardware and software. The agent according to the second embodiment has the configuration shown in FIG. 29, for example, in the same manner as the conventional agent. The variable storage unit 101, the constraint storage unit 102, the neighborhood status storage unit 103, the variable change prohibition period storage unit 104, and the improvement trend storage unit 208 are realized by the memory 902, the secondary storage unit 905, and software that manages them. The constraint solving unit 205 and the improvement ease generating unit 107 are configured by a software module stored in the CPU 901, the memory 902, and the secondary storage unit 905. The communication unit 106 includes a network interface 906 and software for controlling the network interface 906. In addition, a user who uses the schedule adjustment device can interactively set his / her schedule by using the display unit 903 such as a display and the input unit 904 such as a mouse, a keyboard, and a voice input device. Further, the schedule adjustment result can be confirmed by the display unit 903.

[Embodiment 3]

Next, a movement planning device for a cooperative work robot (hereinafter referred to as a robot) in which a plurality of agents according to the third embodiment of the present invention are connected will be described. Many of the agents that can be used in this embodiment do not need to continuously change the variables for a certain period of time under certain conditions. The goal is to eliminate constraint violations, and the goal is for the entire set of agents to reach the solution faster as a result of falling into the local optimal solution.

[0134] First, the configuration and operation of an agent that works with this embodiment will be described. Figure 18 is a configuration diagram of an agent according to the third embodiment of the present invention. In FIG. 18, the agent 700 has almost the same configuration as that of the agent 100 shown in the first embodiment. However, the agent 700 has a planned coordinate storage unit 701 that stores a planned coordinate series instead of the variable storage unit 101, and The difference is that a neighboring planned coordinate storage unit 703 that stores the planned coordinate series of neighboring agents in place of the storage unit 103 is provided. The data structure of the planned coordinate series stored in the planned coordinate storage unit 701 is an arrangement structure of position coordinates. Further, the data structure of the planned coordinate series stored in the neighboring planned coordinate storage unit 703 has a table structure in which data for identifying neighboring agents and array structure data of position coordinates are used as records.

[0135] Next, the operation of the agent will be described. The basic operation of the agent 700 is the same as that of the agent 100 shown in the first embodiment.

[0136] Note that an agent that works in the present embodiment is mounted on a robot and is configured to communicate with an agent mounted on another robot. FIG. 19 is a diagram showing a configuration example of a collaborative work robot equipped with an agent that can perform the third embodiment of the present invention. In FIG. 19, the robot 800 is an external environment detection unit 801 including sensors for detecting its own position, the position of another robot, an obstacle, etc., a moving unit 802 for a motor or leg force for movement, unplanned or predicted A collision avoidance unit 803 for avoiding a collision with an outside obstacle, and 700 agents who make their own movement plan according to the constraints between various inputs from sensors and the movement plan with other robots. With such a configuration, after the robot 800 starts up, the agent 700 determines other data according to the variable data and constraint data that are preliminarily set and the data acquired by the external detection unit 801 and the collision avoidance unit 803. It communicates with the agent of the robot, adjusts the movement plan, makes a movement plan, and instructs the movement unit 802 to move the movement plan.

Next, a specific operation of the movement plan planning apparatus in which a plurality of agents according to the third embodiment is connected will be described. Here, the agents mounted on the two robots Ra and Rb constitute a virtual movement planning device, and plan to move to the destination in cooperation. FIG. 20 is a diagram for explaining a movement problem in which the movement planning apparatus according to the third embodiment of the present invention creates a plan. In FIG. 20, two robots Ra811 and Rb812 work in the same room. Robot Ra811 and Robot Rb812 have their own Sensors can detect and confirm your position in the room, the position of the opponent, and the position of obstacles. However, depending on conditions such as the accuracy of the sensor and the size of the robot body, the movement plan is planned with quantized tile coordinate values. The space in which the cooperative robots Ra811 and Rb81 2 can move has a width of 5 in the X direction and a width of 2 in the Y direction. Robot Ra811 and robot Rb812 exist in this space.

The cooperative work robot Ra811 is located at the initial coordinates (1, 1), and the robot Rb812 is located at the initial coordinates (0, 0). In addition, there is an obstacle K813 at coordinates (3, 1), and the robot Ra811 and the robot Rb812 cannot move here. It is assumed that the presence of the obstacle K813 has already been recognized by the sensors of the robot Ra811 and the robot Rb812. Also, only one robot can move in the X or Y direction in one step. The robot Ra811 and the robot Rb81 2 move to the goal Ga814 and the goal Gb815 on the shortest path, respectively, and the robot Rb812 stays when it reaches the goal Gb815. Under these conditions, the mouth bot Ra811 reaches the final coordinate, that is, the goal Ga814 (coordinate (4, 1)), and the robot Rb812 reaches the goal Gb815 (coordinate (4, 0)) within 5 steps. Consider setting up this mobility plan.

[0139] Here, the planned coordinate series for the robot Ra811 is from the planned coordinates (xl l, y 11) to the planned coordinates (xl5, yl5), and for the robot Rb812, the planned coordinate (x21, y21) force is also the planned coordinates (x25 , y25). The planned coordinates (xl l, yl l) and the planned coordinates (xl5, yl5) are the initial position of the mouth bot Ra811 and the position of the goal Ga814, respectively. The planned coordinates (x21, y21) and the planned coordinates (x25, y25) Is the initial position of robot Rb812 and the position of goal Gb815. These values are already determined, and the movement planning problem is to fill in the coordinates between them

[0140] Next, the constraint conditions of the planning problem will be described. The constraint plan coordinate series is a variable, and the restriction is that the robot Ra811 and the robot Rb812 do not occupy the same coordinates, and cannot take the same coordinates as the obstacle K813. In addition, the movement capabilities of robot Ra811 and robot Rb812 are also a limitation. FIG. 21 is a restriction network diagram of a movement problem created by the movement planning apparatus according to the third embodiment of the present invention. In Figure 21, the X coordinate value of the planned coordinate is “0”, “1”, “2”, “3”, “4”, and the Y coordinate value is “0”, “1”. The value of the However, the robot Ra811 and the robot Rb812 cannot have the same coordinates at the same time. Also, the coordinate (3, 1) has an obstacle K813 and cannot have this planned coordinate value.

[0141] Further, the movement operator "op" indicates restrictions on the movement command. There are a total of five movements: one movement operator that moves the coordinates one step up, down, left, or right in one step, and one movement operator that does not move at all.

[0142] FIG. 22 is an explanatory diagram of the mobile operator of the cooperative work robot according to the third embodiment of the present invention. In Figure 22, the “left” operator decrements the X coordinate value by 1 and does not change the Y coordinate value. That is, “X coordinate value change” is “1”, and “Y coordinate value change” is “not changed”. The “right” operator increases the X coordinate value by 1, and does not change the Υ coordinate value. That is, “X coordinate value change” is “+1”, and “Y coordinate value change” is “Do not change!”. The “up” operator increases the Y coordinate value by 1, and does not change the X coordinate value. That is, “Y coordinate value change” is “+1”, and “X coordinate value change” is “not changed”. The “down” operator decrements the Υ coordinate value by 1 and does not change the X coordinate value. That is, “Y coordinate value change” is “−1”, and “X coordinate value change” is “not changed”. Also, the “stay” operator does not change the coordinate value. In other words, “X coordinate value change” and “Y coordinate value change” are “do not change!”.

[0143] Robot Ra811 and Robot Rb812 select one of these five operators at a time. At this time, as a restriction between these operators, the coordinate series is determined between the coordinates before and after the movement for each movement.

[0144] Next, an operation example for eliminating restrictions between variables of each agent in a movement plan planning apparatus in which a plurality of agents that are useful in the present embodiment are connected will be described. FIG. 23 is a diagram for explaining an initial state of the movement planning apparatus according to the third embodiment of the present invention. In Fig. 23, the robot Ra811 and the robot Rb812 first have their own planned coordinate series (xl l, y 11) and planned coordinate series (xl 5, yl5) and planned coordinates stored in the planned coordinate storage unit 701, respectively. The plan coordinate series (x25, y25) from the series (x21, y21) is shown as planned by the constraint resolution unit 105 based on the restrictions of the moving operator. After making the initial plan, the robot Ra811 and the robot Rb812 send and receive the planned coordinate series via the communication unit 106, and from the planned coordinates (xl2, y12) to the planned coordinates (xl5, yl5) and the planned coordinates (x22, y22) ) Plan Detects a violation of the constraint because the mark (x25, y25) exists at the same coordinates

[0145] FIG. 24 is a diagram for explaining an intermediate state of the movement planning apparatus according to the third embodiment of the present invention. In Fig. 24, the robot Ra811 with priority to eliminate the constraint condition changes the value of the planned coordinate (xl2, yl2) from coordinate (1, 0) to coordinate (2, 1) for the state force in Fig. 23. This shows the state where the variable is set to prohibit modification. Robot Ra811 cannot change the planned coordinate series after the planned coordinates (xl2, yl2) in order to reach the goal Ga814 within the target number of movements. Then, the robot Ra811 transmits to the robot Rb812 improvement ease data including information on the previous change and variable change prohibition period generated by the improvement degree generation unit 107. The robot Rb812 that received the improvement ease data of the robot Ra811 changes the planned coordinates (x23, y23) and changes the subsequent planned coordinates.

FIG. 25 is a diagram for explaining the final state of the movement plan planning apparatus according to the third embodiment of the present invention. In Fig. 25, the planned coordinates of robot Ra811 and robot Rb812 are shown in a state where all the constraints are satisfied. As shown in FIG. 25, the robot Ra811 can reach the goal Ga814, good!] Coordinates (4, 1). Robot Rb812 can reach Gonore Gb815, ie, coordinates (4, 0). Thereby, the movement plan of the robot Ra811 and the robot Rb812 is completed.

[0147] With this configuration, in Embodiment 3, the agent uses the coordinate series as a variable, and the sensor device acquires the initial value of the variable data and constraint data and the value at the time of change from the outside. And create a plan. In addition, by controlling the operating part according to the created plan, the possibility of falling into a local optimal solution and the possibility of some agents falling into an infinite loop are further reduced. As a result, the entire set of agents installed on the robot can make a movement plan quickly.

[0148] In the third embodiment, the coordinate series is used as a variable. However, the mobile operator may be used as a variable.

[0149] (Example)

An effect measurement experiment was conducted using the first embodiment described above. Distributed algorithm In this experiment, the speed of the network is generally evaluated by the number of communications.

The problem of the task assignment device in 1 is solved using a simulator and evaluated by the number of message exchanges.

[0150] Hereafter, the method according to the present invention and the distributed breakout algorithm that is the fastest among existing algorithms (Makoto Yokoo, 1 other, "distributed breakout: iterative improved distributed constraint satisfaction algorithm", IPSJ Journal, 1998, pp. 3989, 6, p. 1889-1897). The operation of the distributed algorithm is inherently asynchronous among the distributed agents. In this experiment, each agent exchanges messages and processes synchronously on the simulator, and all the constraints are satisfied. A comparison is made based on the total number of message exchanges (hereinafter referred to as the number of cycles) up to the point of detection.

[0151] In the task assignment device in the first embodiment, the constraint network in Fig. 7 in which each agent can select only two types of tasks is used as an example. In this experiment, each agent performed experiments using a more complicated constraint network problem in which three or four types of tasks could be selected.

[0152] Figure 26 shows an example of the constraint network. Explain the problem when three tasks can be selected. As shown in Fig. 26, the problem is created by first dividing the agents into three groups, and connecting the agents in different groups randomly with constraints. The constraint is an X ≠ Y type constraint. When you create a problem like this, you can always create a problem that has a solution.

[0153] Figure 26 shows 10 types of problems with a total of 120 agents and a total of 324 constraints in the force experiment with 12 agents. This is Issue I.

[0154] Similarly, when four tasks can be selected, we divided the agents into four groups and created a problem with a solution by randomly connecting the agents in different groups. Ten types of problems with a total of 120 agents and 564 constraints were prepared. This is Issue II.

[0155] In this experiment, for each of the 10 types of tasks I and II, each was initialized with a random value at the time of execution, and the number of cycles to reach the solution was measured. Executed 10 times for each of the 10 types of Task I and 10 times for each of the 10 types of Task II, and reached the solution The average number of cycles until then was tabulated as an evaluation value. However, if the solution is not reached even after exchanging messages of 10,000 cycles (times), it is treated as if it did not reach the solution, and 10000 cycles are included in the average calculation as an evaluation value.

[0156] Figures 27 and 28 show the results. Figure 27 shows the average number of cycles to reach the solution. Figure 28 shows the solution arrival rate. As shown in Fig. 27, in both Problem I and Problem 解, the average number of cycles to reach the solution by the method of the present invention is much smaller than the average number of cycles to reach the solution by the existing method. Further, as shown in FIG. 28, the solution arrival rate according to the method of the present invention is 100% in both the problem I and the problem II. On the other hand, the achievement rate by the existing method reached 100% for both Problem I and Problem 、.

[0157] It can be seen that the method of the present invention reliably reaches the solution in a shorter time than the existing method. In particular, in Problem II, the existing method has 5277.30 cycles (solution attainment rate of 74%), while the method of the present invention has 407.82 cycles (solution attainment rate of 100%). Appears.

[0158] In Embodiments 1 to 3, each problem can be solved by exchanging information several times between agents. But the actual problem is very complex. For example, in the third embodiment, the number of coordinates that the robot can take is very small, the force of the form The constraints in the real world are complicated, and the number of times of information exchange is very large. However, solving such a distributed constraint satisfaction problem with such a complex problem requires a very long time with the existing algorithm, but according to the present invention, it can be solved at high speed. Embodiments 1 to 3 simplify the problem to briefly explain it, and the scope of application of the agent according to the present invention is not limited to a simple problem. .

Note that the present invention is not limited to the field of the above-described embodiment. For example, when multiple autonomous robots work together, it can also be applied to task distribution to robots, robot position identification, and map creation by multiple robots. In such a problem, variables and constraints are distributed for each robot, and it is necessary to solve them by using a plurality of robots. Therefore, the present invention can be applied.

[0160] Even when a plurality of sensors are connected to the network, the present invention provides a It can be applied to integration of sensing information. In order to remove the sensing noise and error of each sensor and obtain accurate information as a whole, it is necessary to solve the variables and restrictions of each sensor and the restrictions on the sensing information between sensors. it can. Alternatively, the present invention can be applied to tasks such as task assignment to each sensor, formation of a network between sensors, assignment of communication frequency between sensors, and a device for sending sensing information to a target node.

[0161] Further, the present invention can also be applied to an apparatus that assigns a tracking target to an observation device in tracking a plurality of observation targets by a plurality of observation devices. When assigning a monitoring target to a surveillance camera or assigning a tracking target to a radar, variables and constraints such as the performance of each observation device, the observation range, and movement of the tracking target are distributed among the observation devices. The present invention can be applied when operating in a distributed environment.

[0162] Further, the present invention can be applied to production planning, inventory planning, delivery planning and the like in supply chain management. Alternatively, the present invention can be applied to solving various logistic problems and planning. Of course, the problem can be solved by distributing it, but the present invention can also be applied to concealing some information.

[0163] The present invention can also be applied to a power system facility work stoppage plan. The present invention can also be applied to energy delivery plans based on energy demand forecasts. For example, in power distribution plans for power plants, the present invention can be applied even when various constraints and variables such as power plant capabilities and maintenance plans, demand forecasts, and distribution networks are distributed and operated in a distributed environment. it can.

[0164] The present invention can also be applied to an air conditioning plan using a plurality of air conditioning devices. Even when distributed air conditioning equipment adjusts the temperature independently, variables and constraints are dispersed! The present invention can also be applied to the case where each device solves the stale state. Further, the present invention can be applied to failure diagnosis of a system consisting of a plurality of devices, communication route routing, communication frequency allocation of a wireless network, and the like.

[0165] The present invention can also be applied to schedule creation and work assignment for railways, buses, and the like. For example, the creation of a diagram at a railway company that has multiple trains entering each other can be performed even when the variables of the diagram and restrictions on entry are distributed among the companies and the control of the variables is solved by each device. The invention can be applied.

Industrial applicability

The agent that is effective in the present invention is suitable for a control device or the like that is mounted on a network-connected task allocation device, schedule device, robot, etc., autonomously communicates asynchronously with neighboring agents and cooperates to eliminate restrictions. .

Claims

The scope of the claims

[1] In an agent where multiple agents work together asynchronously to find a solution,

A variable storage unit for storing variable data indicating a current value of the solution to be obtained;

A constraint storage unit that stores constraint data indicating a combination of values of the variable data and variable data stored by neighboring agents;

A variable change prohibition period storage unit for storing variable change prohibition period data indicating a period during which the change of the variable data is prohibited;

An easy-to-improvement generation unit that generates easy-to-improvement degree data indicating the degree of ease with which the variable data of the own agent satisfies the constraint data;

A communication unit that transmits / receives the variable data and the improvement ease data generated by the improvement degree generation unit to / from a neighboring agent;

A neighborhood situation storage unit for storing the improvement ease data and variable data obtained from the neighboring agents;

The improvement ease data generated by the improvement degree generation unit is compared with the improvement degree data of the neighboring agent stored in the neighborhood status storage unit, and according to the variable change prohibition period data, The variable data is changed to a value satisfying the combination of the constraint data so as to eliminate the constraint violation with the neighboring agent variable data stored in the neighborhood status storage unit, and A constraint resolution unit that sets variable change prohibition period data in a predetermined period;

Agent with.

[2] The improvement ease generator generates a constraint number obtained by calculating a total number of the constraint data for the variable data, and the variable data and the variable data of the neighboring agent satisfy a combination of the constraint data values. Of the constraint data for which the total number of constraint data is obtained and the constraint data whose variable data and the neighboring agent variable data do not satisfy the combination of the constraint data values, the variable data is The variable change prohibition period The improvement possible number obtained by calculating the total number of combinations that are not included in the data and that can be changed to satisfy the combination of the values of the constraint data by changing the values of the variable data Generate ease of improvement data consisting of at least one of The agent of claim 1.

[3] The degree to which the variable data satisfies the constraint data! / The degree to speak! Further includes an improvement trend storage unit that stores improvement trend data indicating past transitions of

The constraint resolution unit additionally updates the improvement trend data stored in the improvement trend storage unit with the number of constraint violations included in the ease of improvement data before changing the variable data, and improves the improvement. Determine the length of the variable change prohibition period according to the trend data, and set the variable change prohibition period data after changing the variable data

The agent according to claim 2.

[4] The improvement ease generating unit generates the improvement ease data every first fixed time interval, and the communication unit transmits the improvement ease data to the neighboring agents. The agent according to any one of items 3.

[5] The improvement ease generation unit generates the improvement ease data every second fixed time interval, and the constraint resolution unit stores the neighboring agent according to the variable change prohibition period data. Change the variable data and set the variable change prohibition period data so as to eliminate the constraint violation with the variable data

The agent according to claim 1 or claim 3.

6. The agent according to claim 5, wherein the second constant time interval is greater than the first constant time interval.

7. The agent according to claim 6, wherein the variable change prohibition period indicating a period during which change of the variable data is prohibited is k times the second constant time interval, and k is an integer.

[8] When the constraint resolution unit determines whether to change the variable data, the improvement ease data of the own agent and the improvement degree of the neighboring agent stored in the neighborhood state storage unit are stored. Compared with data, at least when the number of improvement possible is the largest, when there is no agent force S that can change the variable data by violating the constraint other than the own agent, and when the number of constraint violation is the largest, It is determined that the variable data is to be changed when the number of constraints is the smallest or any one of

The agent according to claim 1 or claim 3.

[9] When changing the variable data, the constraint resolution unit satisfies the constraints from the constraint data. Select the value of the variable to be added, and update the variable data in the variable storage unit

The agent according to claim 1 or claim 3.

[10] The constraint resolution unit compares the total ml of constraint violations in the latest third fixed time interval of the improvement trend data with the total mO of constraint violations in the previous third fixed time interval. Therefore, if ml <mO, it is determined that there is an improvement trend, and the current set value for the variable change prohibition period is shortened, and if ml≥mO, it is determined that there is no improvement trend and the variable change prohibition period 4. The agent according to claim 3, wherein the currently set value is lengthened and the variable change prohibition period data is set.

[11] Variable data that indicates the current value of the solution that each agent seeks, constraint data that indicates a combination of the variable data and variable data of neighboring agents, and change of the value of the variable data In a distributed constraint satisfaction method in which a plurality of agents cooperate to obtain a solution of the variable data, each of which has variable change prohibition period data indicating a period during which the variable data is prohibited, and all constraint relationships between the variable data are established.

Each agent

A generation step of generating the improvement ease data indicating the ease with which the variable data of the self agent satisfies the constraint data;

Asynchronously transmitting and receiving the variable data and the improvement ease data generated in the generation step to each neighboring agent;

A step of determining whether to change the variable data of the self by comparing the variable data and the ease of improvement data with the self-improvement data of the self and the improvement ease data of each neighboring agent; ,

A change step in which, when it is determined in the determination step that the variable data of the self is changed, the variable data of the self is changed to a value satisfying the combination of the constraint data and notified to the neighboring agents;

A change prohibition step for setting variable change prohibition period data in a predetermined period when the variable data is changed in the change step;

A dispersion constraint satisfaction method comprising:

[12] In the generating step, a control for obtaining a total number of the constraint data for the variable data. The divisor, the variable data and the variable data of the neighboring agent satisfy the combination of the values of the constraint data, the number of constraint violations for which the total number of constraint data is obtained, the variable data and the neighborhood Among the constraint data whose agent variable data does not satisfy the combination of the constraint data values, the variable data is not included in the variable change prohibition period data, and the variable data value is changed. To generate improvement degree ease data consisting of at least one of the improvement possible numbers, which is obtained as a total number of combinations that can be changed to satisfy the combination of values of the constraint data.

The dispersion constraint satisfaction method according to claim 11.

[13] When it is determined that the variable data is to be changed, a step of additionally storing the current value of the number of violations of the constraint as improvement trend data in the home agent;

A step of adjusting a period during which the change of the variable data is prohibited based on the improvement trend data;

The dispersion constraint satisfaction method according to claim 12, further comprising:

[14] In the determination step, when comparing the self-improvement data of self with the ease-of-improvement data of each neighboring agent, if there is at least the largest possible number of improvements and a constraint violation other than the self agent If the variable data is changed when there is no agent that can change the variable data, when the number of violations of the constraint is the largest, or when the number of the constraints is the smallest. decide

The dispersion constraint satisfaction method according to claim 11.

[15] In the changing step,

When changing the variable data of its own, the variable data satisfying the constraint is selected and determined from the constraint data, and the variable data is updated.

The dispersion constraint satisfaction method according to claim 11.