CN114326694A

CN114326694A - Intelligent agent cluster control method, device, equipment and storage medium

Info

Publication number: CN114326694A
Application number: CN202011015034.XA
Authority: CN
Inventors: 汪建平; 吴巍炜; 黄子尧
Original assignee: City University of Hong Kong CityU
Current assignee: City University of Hong Kong CityU
Priority date: 2020-09-24
Filing date: 2020-09-24
Publication date: 2022-04-12
Anticipated expiration: 2040-09-24

Abstract

An embodiment of the specification provides an intelligent agent cluster control method, an intelligent agent cluster control device, intelligent agent cluster control equipment and a storage medium, wherein the method comprises the following steps: determining a shortest path between any two agents in an agent cluster, and determining a constraint condition according to the shortest path, the communication distance and the perception distance of the agents in the cluster; under the constraint of the constraint condition, calling an RMA algorithm to select an agent from the cluster for sensing a newly added target point; in the communication topology of the cluster, the selected intelligent agent is used as a root node, and a tree structure is established according to the rule that all non-father neighbor nodes are child nodes; and calling a TMP algorithm to control the distributed movement of the agents under the tree structure until the target point is sensed by the root node. The embodiment of the specification can reduce communication overhead among the intelligent agents and simultaneously avoid reducing the perception capability of the intelligent agent cluster.

Description

Intelligent agent cluster control method, device, equipment and storage medium

Technical Field

The present disclosure relates to the field of technologies, and in particular, to an intelligent agent cluster control method, an intelligent agent cluster control apparatus, an intelligent agent cluster control device, and a storage medium.

Background

Agent clusters (e.g. drone clusters, etc.) are capable of providing exceptional perceptual capabilities, and are therefore very important solutions for many applications such as video surveillance, disaster management and road surveillance. In the intelligent agent cluster control, a key problem to be solved urgently is as follows: and sensing all target points on the premise of maintaining the connectivity among all the agents, namely a multi-target point sensing problem. When the communication topology changes rapidly among a cluster of agents, the communication overhead for maintaining connectivity among agents can be very high, which may cause some time-critical tasks to fail. One solution to this problem is to maintain a cluster of agents as a fixed communication topology. The fixed communication topology enables static routing, and thus the communication overhead can be greatly reduced. However, keeping the communication topology unchanged severely limits the perception capabilities of the cluster of agents.

Disclosure of Invention

An object of the embodiments of the present specification is to provide an agent cluster control method, apparatus, device, and storage medium, so as to reduce communication overhead between agents and avoid reducing perception capability of an agent cluster.

In order to achieve the above object, in one aspect, an embodiment of the present specification provides an agent cluster control method, including:

determining a shortest path between any two agents in an agent cluster, and determining a constraint condition according to the shortest path, the communication distance and the perception distance of the agents in the cluster;

under the constraint of the constraint condition, calling an RMA algorithm to select an agent from the cluster for sensing a newly added target point;

in the communication topology of the cluster, the selected intelligent agent is used as a root node, and a tree structure is established according to the rule that all non-father neighbor nodes are child nodes;

and calling a TMP algorithm to control the distributed movement of the agents under the tree structure until the target point is sensed by the root node.

In an embodiment of the present specification, the constraint condition includes: d_t(τ_new,a_c)≤r₁+r₂*l_a,c；

Wherein, tau_newAs a newly added target point, a_cAnchor points c, d for agent a_t(τ_new,a_c) At time t_newAnd a_cDistance between r₁Is the perceived distance of the agent, r₂Communication distance for agent,/_a,cIs the shortest path length between agent a and its anchor point c.

In an embodiment of the present specification, under the constraint of the constraint condition, invoking an RMA algorithm to select an agent from the cluster, so as to sense a new target point, includes:

for each agent in the cluster, determining the number of anchor points of the agent which meet the constraint condition;

selecting the agents with the most anchor points meeting the constraint condition from the cluster for sensing the newly added target points;

wherein, the anchor point refers to: for any two agents a in the cluster_iAnd a_jIf a_iIs responsible for sensing at least one target point, and no other agent on the shortest path between the two senses the target point, then a_iIs a_jAn anchor point of (a).

In an embodiment of this specification, the establishing a tree structure according to a rule that all non-parent neighbor nodes are child nodes includes:

taking all neighbor nodes of the root node as child nodes of the root node;

confirming whether an agent which does not participate in tree building exists in the communication topology of the cluster;

if the leaf nodes exist, taking all the non-father neighbor nodes of each leaf node in the current tree structure as self child nodes;

and sequentially recursing until all the intelligent agents in the communication topology of the cluster participate in tree building.

In an embodiment of this specification, the calling the TMP algorithm to control the distributed movement of the agent under the tree structure includes:

calling a nearest target point NRPT algorithm to find a target position; the target position is the position where the root node keeps the connection between the perceived target point and all the neighbor nodes in the communication topology and is closest to the moving target of the root node;

judging whether the target position satisfies d_t(τ,p_n)<d_t(τ, a); where τ is the moving target of the root node, p_nIs a target location, a is a root node, d_t(τ,p_n) At times t τ and p_nDistance between d_t(τ, a) is the distance between τ and a at time t;

if the target position satisfies d_t(τ,p_n)<d_t(τ, a), moving the root node to the target position, and when the newly added target point is not sensed after the root node finishes moving, enabling the root node to search the target position again until the root node senses the newly added target point.

In an embodiment of this specification, the calling the TMP algorithm to control the distributed movement of the agent under the tree structure further includes:

if the target position does not satisfy d_t(τ,p_n)<d_t(τ, a), calling a BTNK algorithm to find a bottleneck node which blocks the agent of the root node from moving to the moving target;

when the bottleneck node is the target point which is perceived by the root node, finishing the perception of the bottleneck node and enabling the root node to search for the target position again;

when the bottleneck node is an agent, the root node sends an assistance request to the bottleneck node, and after receiving an assistance response of the agent, the root node searches for a target position again until the root node senses the newly added target point;

the agent receiving the assistance request takes the father node of the father node as the moving target, executes the step corresponding to the root node, and returns the assistance response after moving to a position closer to the moving target. In an embodiment of the present specification, the invoking NRPT algorithm to find the target location includes:

generating a circle by taking the position of each neighbor node of the root node as a circle center and communication as a radius respectively;

and finding the target position in the intersection area of all circles.

In an embodiment of this specification, said invoking the BTNK algorithm to find a bottleneck node that blocks the movement of the root node to its moving target includes: determining all agents having a distance from the root node equal to a communication distance as a first set; determining all target points with the distance equal to the perception distance from the root node as a second set;

finding an element in the union of the first set and the second set so that the element is located at p_t(a) And p_t(τ) the angle of intersection is largest compared to other elements and returns the element as the bottleneck node of the root node;

wherein p is_t(a) Is the position of the root node a at the current time t, p_t(τ) is the position of the moving target τ at the current time t.

On the other hand, an embodiment of the present specification further provides an intelligent agent cluster control apparatus, including:

the determining module is used for determining the shortest path between any two agents in the agent cluster and determining constraint conditions according to the shortest path, the communication distance and the perception distance of the agents in the cluster;

the selecting module is used for calling an RMA algorithm to select an agent from the cluster under the constraint of the constraint condition so as to sense a newly added target point;

the establishing module is used for establishing a tree structure in the communication topology of the cluster by taking the selected intelligent agent as a root node and according to the rule that all non-father neighbor nodes are child nodes;

and the moving module is used for calling a TMP algorithm to control the distributed movement of the agents under the tree structure until the target point is sensed by the root node.

In another aspect, the embodiments of the present specification further provide a computer device, which includes a memory, a processor, and a computer program stored on the memory, and when the computer program is executed by the processor, the computer program executes the instructions of the above method.

In another aspect, the present specification further provides a computer storage medium, on which a computer program is stored, and the computer program is executed by a processor of a computer device to execute the instructions of the method.

As can be seen from the above technical solutions provided in the embodiments of the present specification, since the distributed movement of the agents is controlled and implemented by a topology-invariant motion planning algorithm (i.e., a TMP algorithm) in a tree structure, the physical topology of the agent cluster is allowed to be partially changed while the communication topology of the agent cluster is maintained fixed, so that the communication overhead generated by dynamic routing can be reduced as much as possible. Moreover, the agent for sensing the newly added target point is selected from the cluster based on the maximum anchor point retention algorithm (namely, the RMA algorithm), which can be beneficial to further reduce the loss of the covered old target point when the agent cluster covers the newly added target point, thereby being beneficial to avoiding reducing the sensing capability of the agent cluster.

Drawings

In order to more clearly illustrate the embodiments of the present specification or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, it is obvious that the drawings in the following description are only some embodiments described in the present specification, and for those skilled in the art, other drawings can be obtained according to the drawings without any creative effort. In the drawings:

FIG. 1 is a flow chart of a method for agent cluster control in some embodiments provided herein;

FIG. 2 is a schematic diagram of an agent seeking a target location in one embodiment provided herein;

fig. 3 to 8 are schematic diagrams illustrating a process of sensing a newly added target point by an agent according to an embodiment of the present disclosure;

fig. 9 is a schematic diagram illustrating a variation of the number of average sensing target points with the size of an agent according to different algorithms in an embodiment provided in the present specification;

fig. 10 is a schematic diagram illustrating a variation of the number of average sensing target points of an agent cluster in the last round of simulation with the size of the agent under different algorithms in an embodiment provided in the present specification;

FIG. 11 is a graph illustrating end-to-end latency and packet transmission rate of a cluster of agents as a function of agent size under various algorithms in one embodiment provided herein;

fig. 12 is a schematic diagram illustrating comparison between end-to-end delay, average packet forwarding times, and route traffic averaged to each agent and transmission data of an agent cluster under different routing algorithms in an embodiment provided in the present specification;

FIG. 13 is a block diagram of an intelligent agent cluster control device in some embodiments provided herein;

FIG. 14 is a schematic diagram of the deployment of an agent to a multi-agent mobile aware platform system in some embodiments provided herein;

FIG. 15 is a block diagram of a computer device in some embodiments provided herein.

[ description of reference ]

131. Determining a fetching module;

132. a selection module;

133. establishing a module;

134. a moving module;

1502. a computer device;

1504. a processor;

1506. a memory;

1508. a drive mechanism;

1510. an input/output module;

1512. an input device;

1514. an output device;

1516. a presentation device;

1518. a graphical user interface;

1520. a network interface;

1522. a communication link;

1524. a communication bus.

Detailed Description

In order to make those skilled in the art better understand the technical solutions in the present specification, the technical solutions in the embodiments of the present specification will be clearly and completely described below with reference to the drawings in the embodiments of the present specification, and it is obvious that the described embodiments are only a part of the embodiments of the present specification, and not all of the embodiments. All other embodiments obtained by a person skilled in the art based on the embodiments in the present specification without any inventive step should fall within the scope of protection of the present specification.

Embodiments of the present description relate to a mobility (or migration) control technique for a cluster of agents. The agent cluster refers to a movable agent cluster, and may include, but is not limited to, a mobile robot cluster, a drone cluster, and the like.

Due to their mobility and agility, agent clusters can be used in different scenarios and applications (e.g. mapping, area search and rescue, grid monitoring, oil and gas pipe monitoring, road monitoring, disaster monitoring, border patrol, fire fighting, etc.). A key problem in controlling an agent cluster is how to sense all target points (or referred to as interest points, which are location points that need to be sensed by an agent) under the condition of maintaining global connectivity of the agent cluster, for example, when the agent cluster is tasked with fire monitoring of an oil storage, the agent cluster needs to sense (i.e., collect) data such as images and environments of each oil storage tank in the oil storage, and the location of each oil storage tank is a target point. However, due to changes in communication topology caused by the movement of a cluster of agents, communication links between agents need to be updated and maintained frequently. The overhead generated in this process may significantly affect the network performance of the agent communication (e.g., cause the end-to-end delay to increase, and the packet transmission rate to decrease, etc.), so that some tasks with strict time requirements (e.g., real-time monitoring task) may fail.

In view of this, embodiments of the present specification provide an agent cluster control method, which may be applied to a cluster control system side of an agent. Referring to fig. 1, in some embodiments of the present description, the method for controlling a cluster of agents may include the following steps:

s101, determining a shortest path between any two agents in an agent cluster, and determining a constraint condition according to the shortest path, the communication distance of the agents in the cluster and a perception distance.

And S102, under the constraint of the constraint condition, selecting an intelligent agent from the cluster according to a maximum anchor point preservation algorithm (RMA) for sensing a newly added target point.

S103, in the communication topology of the cluster, the selected intelligent agent is used as a root node, and a tree structure is established according to the rule that all non-father neighbor nodes are child nodes.

S104, controlling distributed movement of the intelligent bodies in the tree structure according to a Topology-fixed Motion Planning (TMP) algorithm until the target point is sensed by the root node.

In the embodiment of the present specification, since the distributed movement of the agents is realized by the TMP algorithm control under the tree structure, it is possible to allow the physical topology of the agent cluster to be partially changed while maintaining the communication topology of the agent cluster fixed, so as to allow the coverage of the selected agent to the new target point, and simultaneously reduce the loss of the old target point that has been covered due to the coverage of the new target point as much as possible. Moreover, the intelligent agent for sensing the newly added target point is selected from the cluster based on the RMA algorithm, so that when the intelligent agent cluster covers the newly added target point, the loss of the covered old target point can be further reduced. Coverage in embodiments of the present description means that the target point falls within the perceived distance range of the agent.

It will be understood by those skilled in the art that in the embodiments of the present disclosure, maintaining (or fixing) the communication topology of an agent means that, initially, a connection exists between an agent and which agents, and then, during a subsequent movement, the distance between the agent and the agents is kept smaller than the communication distance, so as to ensure that the connection between the agent and the agents is not disconnected. While allowing partial changes in the physical topology of a cluster of agents means that an agent moves while considering only some of the agents that are initially connected to remain connected and not disconnected, rather than remaining connected to all of the initially connected agents.

In some embodiments of the present description, each agent in the agent cluster may be at a perceived distance r₁Sensing the target point within the range and when the target point is at the communication distance r₂When within range, any two agents in the agent cluster may communicate with each other. Note that a is a node set of the agent cluster, T is a communication topology (i.e., an edge set) of the agent cluster, G (a, T) is a connected acyclic graph in which the node set is a, and the edge set is T.

Thus, for any two agents a_i,a_j(a_i,a_jE.g. A), the shortest path between the two in the acyclic graph G (A, T) can be calculated and stored, and l is recorded_i,jIs the path length. Computing any two agents a_i,a_jThe shortest path in the acyclic graph G (a, T) may be used for subsequent generation of constraints. In some exemplary embodiments, any two agents a in the acyclic graph G (a, T) may be computed based on, for example, a depth or breadth first search algorithm, a froude algorithm, a Dijkstra algorithm, or a Bellman-Ford algorithm, among others_i，a_jThe shortest path between them.

Research shows that the shortest path, the sensing distance and the communication distance between any two agents in an agent cluster are important factors influencing the communication overhead and the sensing capability of the agent cluster. Therefore, in order to take account of the communication overhead and the sensing capability of the agent cluster, constraint conditions can be set according to the shortest path, the communication distance and the sensing distance.

In some embodiments of the present description, the constraint may be d_t(τ_new，a_c)≤r₁+r₂*l_a，c. Wherein, tau_newAs a newly added target point, a_cAnchor points c, d for agent a_t(τ_new，a_c) At time t_newAnd a_cDistance between r₁Is the perceived distance of the agent, r₂Communication distance for agent,/_a，cIs the shortest path length between agent a and its anchor point c. It can be seen that, in the above constraint, the left side of the inequality represents the distance between the anchor point and the newly added target point, and the right side of the inequality represents the theoretical maximum distance (i.e., communication distance) between the agent and the anchor point plus the perceived distance; for an agent, an anchor point, and a newly added target point that satisfy the inequality, the agent may be considered approximately able to overlay the newly added target point while the anchor point may remain immobile (i.e., the edge between the agent and the anchor point in the communication topology may be maintained unbroken).

In some embodiments of the present description, when selecting an agent from the cluster according to the RMA algorithm, d may be noted_t(τ_new,a_c) At time t_newAnd a_cEuclidean distance between, first, the RMA algorithm for any two agents a_i，a_jε A is defined as follows: if a_iIs responsible for sensing at least one target point, and no other agent on the shortest path between the two senses the target point, so it can be called a_iIs a_jAn anchor point of (a). Secondly, for each agent a belonging to A, the number of agents which have positive responsibility perception of at least one target point on the shortest path from a to other agents can be detected, namely the anchor point set of a at the current moment t

Can be calculated; obviously, a at the present time tThe anchor point set can maintain the communication topology unchanged so as not to influence the perception of the anchor point set. Then, for each agent a, the RMA algorithm calculates that the condition d is satisfied_t(τ_new，a_c)≤r₁+r₂*l_a，cThe number of anchor points of (c); wherein, can be written k_aThe number of anchor points for the agent. Finally, the RMA algorithm may select agent a_root＝argmax_a∈ _Ak_aTo be responsible for sensing tau_new(i.e., newly added target points).

Those skilled in the art will appreciate that the selection of agents from the cluster according to the RMA algorithm is merely an exemplary illustration, and the description is not limited thereto; in other embodiments of the present specification, an algorithm such as random sampling, nearest selection, etc. may be selected instead according to actual needs.

In some embodiments of the present description, in order to facilitate subsequent control of the movement of the agent, a tree structure may be established in a communication topology of a cluster by using the selected agent as a root node and according to a rule that all non-parent neighbor nodes are child nodes.

In some embodiments of the present specification, the establishing a tree structure according to a rule that all non-parent neighbor nodes are child nodes may include:

(1) and taking all neighbor nodes of the root node as child nodes of the root node. In the embodiments of the present specification, the neighbor node of one node refers to: and in the communication topology of the cluster, the distance between the intelligent agent and the node is one hop. Thus, all neighbor nodes of the root node refer to: and in the communication topology of the cluster, all the agents with the distance of one hop from the root node.

For example, taking the embodiment shown in FIG. 3 as an example, the cluster includes three agents u₁，u₂，u₃If u is₁Is selected as the root node due to the sum of u and₁an agent with a distance of one hop of u only₂Therefore u is₂Can be used as u₁Of (d) (corresponding, u)₁I.e. u₂Parent node of).

(2) And confirming whether an agent which does not participate in tree building exists in the communication topology of the cluster. If not, completing the establishment of the tree structure; otherwise, executing step (3). In the embodiments of the present specification, the fact that one node does not participate in building a tree means that: the node does not belong to a data element in the tree, i.e. the node does not form part of the tree structure.

(3) And if the node exists, regarding each leaf node in the current tree structure (the leaf node refers to a node without a child node in a tree), taking all non-parent neighbor nodes as the child nodes of the node. In embodiments of the present specification, all non-parent neighbor nodes of a leaf node refer to: and in the communication topology of the cluster, all the neighbor nodes except the parent node are arranged.

For example, taking the exemplary embodiment shown in FIG. 3 as an example, assume that the current tree structure is represented by u₁And u₂Form (i.e. u)₃Not currently participating in building a tree), where u₂Is u₁The child node of (c), then it is obvious that u is at this time₂Is the leaf node of the current tree structure. To be in a cluster (u)₁，u₂,u₃) In communication topology of u₂All neighbor nodes of (1) include u₃And u₁But due to u₁Is u₂Of a parent node of, thus, u₂All non-parent neighbor nodes of u₃。

(4) And repeating the steps (2) to (3) until all the intelligent agents in the communication topology of the cluster participate in tree building.

In the embodiments of the present specification, distributed movement refers to: during the moving process, the intelligent agents can not directly communicate with all the intelligent agents in the cluster, namely, the intelligent agents can only directly communicate with the intelligent agents in the cluster within the communication range of the intelligent agents, and can forward messages through the intelligent agents so as to realize the communication with the intelligent agents in the cluster outside the communication range of the intelligent agents.

In some embodiments of the present description, the new target point is made as a parent node of the root node before the TMP algorithm is invoked to control the distributed movement of the agent in the tree structure. I.e., TMP algorithm can make τ_newIs formed as a_rootThe moving object of (2) is,and define τ_newIs a_rootThe parent node of (2). Furthermore, for each agent that is not a root node, it is possible to have the parent node of its parent node as its own moving target (even if its grandparent node is its own moving target). Therefore, the non-root node can help the movement of the parent node through self movement, namely, the limitation of self on the movement of the parent node is reduced, and finally, the reduction of the movement limitation on the root node is realized.

On the basis, calling a TMP algorithm to control the distributed movement of the agents under the tree structure can comprise the following steps:

(1) calling a NeaRest target PoinT (Nearest degraded PoinT, NRPT for short) algorithm to find a target position; the target position is the position where the root node keeps the connection between the perceived target point and all the neighbor nodes of the root node and is closest to the moving target of the root node;

(2) judging whether the target position meets d_t(τ，p_n)<d_t(τ, a); where τ is the moving target of the root node, p_nIs a target location, a is a root node, d_t(τ，p_n) At times t τ and p_nDistance between d_t(τ, a) is the distance between τ and a at time t.

Wherein the calling the NRPT algorithm to find the target location may include: generating a circle by taking the position of each neighbor node of the root node as a circle center and communication as a radius respectively; and finding the target position in the intersection area of all circles.

For example, in the exemplary embodiment shown in FIG. 2, for agent u_rootFinding a target position p_nWhen the set C includes two circles, u respectively₁,u₂A circle having the communication distance as a radius and serving as a circle center; then, the distance u is searched in the comparison area of the two circles_rootThe position where the target point tau is closest, and the found position p is judged_nCompared with u_rootWhether the current location is closer to τ or not, due to p found in FIG. 2_nThe points are closer together, hence for u_rootSuch a location is successfully found.

(3) If the target position satisfies d_t(τ,p_n)<d_t(τ, a), the root node may be moved to the target location, and when the new target point is not sensed after the root node finishes moving, the root node is enabled to search for the target location again until the root node senses the new target point.

(4) If the target position does not satisfy d_t(τ,p_n)<d_tAnd (tau, a), calling a BoTtleNecK (BoTtleNecK, BTNK for short) algorithm to find a BoTtleNecK node which prevents the root node from moving to the moving target of the root node. In embodiments of the present description, a bottleneck node may be a target point or an agent. And when the bottleneck node is the target point which is perceived by the root node, finishing the perception of the bottleneck node and enabling the root node to find the target position again. And when the bottleneck node is an agent, enabling the root node to send an assistance request to the bottleneck node, and enabling the root node to search for the target position again after receiving an assistance response of the agent until the root node senses the newly added target point.

Obviously, when the bottleneck of the root node is an agent and the agent cannot further reduce the limitation on the movement of the root node only by means of the movement of the agent, it indicates that the agent also has the bottleneck of the agent, and at this time, the agent can search the bottleneck node of the agent and perform corresponding processing as the root node.

Therefore, the NRPT algorithm may also target a non-root node agent a to find a location p under the conditions of ensuring that its target can be continuously sensed and ensuring that its neighbors can still communicate with it if they do not move in T_nSo that d is_t(τ,p_n)<d_t(τ, a), where τ is the moving target of a. Note p_t(. cndot.) is the location of an agent or target point at the current time t, N (-) is the set of neighbors that are only one hop away from an agent,

for a set of target points that an agent is currently responsible for sensing, the steps of the NRPT algorithm may beThe summary is as follows:

a) initializing a set of circles

b) For any a_ne.N (a) addition of p_t(a_n) Centered at r_cA circle of radius into set C.

c) Solving the problem of finding a point p in the intersection region of all circles in C_nSo that no other point distance p is present in the intersection region_t(τ) is closer to this geometry problem.

d) If d is_t(τ,p_n)<d_t(τ, a) then returns p_nOtherwise, the NRPT fails to find the location.

That is, the agent that has received the assistance request executes the step corresponding to the root node with the parent node of its parent node as its moving target, and returns the assistance response after moving to a position closer to its moving target. Specifically, for any agent that receives the assistance request:

(1) calling an NRPT algorithm to find a target position; the target position is the position where the intelligent agent keeps the connection between the perceived target point and all the neighbor nodes of the intelligent agent and is closest to the moving target of the intelligent agent;

(2) judging whether the target position meets d_t(τ,p_n)<d_t(τ, a); where τ is the moving target of the agent, p_nIs the target location, a is the agent, d_t(τ,p_n) At times t τ and p_nDistance between d_t(τ, a) is the distance between τ and a at time t.

(3) If the target position satisfies d_t(τ,p_n)<d_t(τ, a) may move the agent to the target location and return the assistance response to the root node after the agent has finished moving.

(4) If the target position does not satisfy d_t(τ,p_n)<d_t(τ, a), calling BTNKThe algorithm looks for bottleneck nodes that prevent the agent from moving to its moving target. And when the bottleneck node is the target point which is sensed by the intelligent agent, finishing sensing the bottleneck node and enabling the intelligent agent to find the target position again. And when the bottleneck node is another agent, the agent sends an assistance request to the other agent, and the target position is searched again after the assistance response of the other agent is received, so that the assistance response is returned to the root node. For the other agent that receives the assistance request, steps similar to those of the agent may be executed again for recursion.

The invoking the BTNK algorithm to find a bottleneck node that blocks the root node from moving to its moving target may include the following steps:

determining all agents having a distance from the root node equal to a communication distance as a first set; determining all target points with the distance equal to the perception distance from the root node as a second set;

finding an element in the union of the first set and the second set so that the element is located at p_t(a) And p_tThe angle at which (τ) intersects is the largest compared to the other elements and returns that element as the bottleneck node of the root node. Wherein p is_t(a) Is the position of the root node a at the current time t, p_t(τ) is the position of the moving target τ at the current time t.

Wherein the element is located at a position corresponding to p_t(a) And p_tThe angle at which (τ) intersects is greatest compared to the other elements means: if the current position of the element is point A, position p_t(a) Is point B, position p_t(τ) is point C, then the angle crossed here is ≈ ABC, that is, the angle between ray BA and ray BC. Studies have shown that the greater the angle, the greater the obstruction of the element to its parent. Thus, the element may be returned as a bottleneck node.

Similarly, the target of the BTNK algorithm may be a non-root agent a for which p cannot be found by an NRPT algorithm_nFor which finding a hindrance to its directionAnd moving the bottleneck node close to the target. The steps of the BTNK algorithm can be summarized as follows:

a) set of agents N computing bottleneck nodes that may be a_s＝{a_s|a_s∈N(a)∧d_t(a,a_s)＝r_cAnd a set of target points that may be bottleneck nodes of a

b) In the collection

Find the element so that the element's current position is associated with p_t(a) And p_tAngle of intersection (defined in the interval 0, π)]Above) is the largest compared to other elements and then returns the element as the bottleneck node for a. Wherein the content of the first and second substances,

is the set of target points that agent a is covering at the current time t.

It should be noted that the above-mentioned TMP algorithm, NRPT algorithm and BTNK algorithm are only exemplary, and the description does not limit the same; in other embodiments of the present description, any other suitable algorithm may be selected instead according to actual needs.

For ease of understanding, the migration change process in which three agents form a cluster to cover two target points is described below with reference to fig. 3 to 8. Referring to FIG. 3, a cluster includes three agents u₁,u₂，u₃(ii) a The communication topology includes links (u)₁，u₂) And (u)₂，u₃) (ii) a Initially the old emerging target point τ has been picked up by agent u₃Coverage, τ_newIs a newly added target point. According to RMA algorithm u₁Becomes responsible for sensing the newly emerging target point τ_newRoot node agent of (1), but since u₂Limitation of u₁Finding that oneself can not move to the distance tau after running NRPT algorithm_newMore closely, therefore, itRun the BTNK algorithm, consider u₂Is a bottleneck node, and is directed to u₂Please help. U is shown in connection with FIG. 4₂Will be tau after receiving the help request_newSet as its own target point, and u₁But also to move closer to its target. When u is shown in connection with FIG. 5₂Successfully move to a position closer to the target point and go to the parent node u₁And carrying out assistance response. When u is shown in connection with FIG. 6₁After receiving the assistance response, the bottleneck node which limits the self-movement is solved, so that the movement attempt can be carried out again. U is shown in connection with FIG. 7₁After the re-movement attempt, the successful movement to the distance tau is found_newCloser, and can be given by_newCovering (i.e. the target point τ can be perceived)_new). Finally adding tau_newAnd old τ are covered at the same time, i.e. u₁Perception tau_new，u₃Sense τ (e.g., as shown in fig. 8).

To evaluate the feasibility of the agent cluster control method of the embodiments of the present specification, a simulation test may be performed. In a simulation test, the communication distance of an intelligent agent cluster is set to be 70m, the induction distance is set to be 50m, the moving speed of an intelligent agent is set to be 8m/s, the maximum number of neighbor nodes in T is set to be 3, and the positions of target points are obtained by uniformly sampling in a circle which takes the geometric centers of all the positions of the intelligent agent as the center and takes 1118m as the radius. As shown in connection with fig. 9, clusters containing 15, 20, 50, 80, 100 agents were then evaluated separately. In fig. 9, the ordinate represents the number of the cluster average sensing target points, and the abscissa represents the scale of the agent cluster; the curve corresponding to RMA + TMP is the test result of the intelligent agent cluster control method adopting the embodiment of the specification; the curve corresponding to RAND + TMP is a test result obtained after the RMA algorithm is replaced by a random sampling method; the curve corresponding to EXP represents the desired perceptual performance in the case when all agents are stationary and connectivity is achieved with minimal perceptual area overlap cost. The experimental results presented in fig. 9 show that the agent cluster control method using the embodiments of the present specification significantly surpasses other algorithms. For all simulation runs, the mean ratio of RMA + TMP to RAND + TMP sensed 5.44 more target points and the mean ratio of EXP sensed 5.60 more target points. Thereby verifying the feasibility of the agent cluster control method of the embodiments of the present specification.

Further, fig. 10 shows the average perceived target point number for each algorithm in the last simulation run when all target points have been initially present and need to be perceived as much as possible while maintaining the communication topology constraints. While such a motion-aware optimization problem has proven to be an NP-hard problem, it can still be translated into a mixed integer linear programming problem by relaxing the constraints. Therefore, the problem that when the initial positions of all target points are given, under the condition that the feasible limiting conditions of the communication topology are ensured to be met, a specific number of intelligent agents are deployed to sense more target points can be solved. The result of this mixed integer linear programming can be considered a better (or optimal) solution to the motion perception problem. In fig. 10, the ordinate represents the average number of sensing target points of the cluster in the last round of simulation, and the abscissa represents the scale of the intelligent agent cluster; the curve corresponding to the OPT (optimal) represents the optimal solution of the mobile sensing problem; the curve corresponding to RMA + TMP is the test result of the intelligent agent cluster control method adopting the embodiment of the specification; the curve corresponding to RAND + TMP is a test result obtained after the RMA algorithm is replaced by a random sampling method; the curve corresponding to EXP represents the desired perceptual performance in the case when all agents are stationary and connectivity is achieved with minimal perceptual area overlap cost. The results in fig. 10 show that the agent cluster control method using embodiments of the present specification outperforms the EXP and RAND + TMP algorithms significantly and as the number of agents in the cluster increases, the algorithms behave very close to OPT.

Further, the behavior of the communication between agents when undergoing the same mobility procedures and traffic generation but using different routing protocols is shown in FIG. 11. Mobility planning for agent applications results from the RMA + TMP algorithm (hereinafter shortly referred to as TMP), dynamic routing protocols (OLSR and DSDV) and static unicast routing tables (P2P) based on the communication topology, respectively, being used for routing. Test results show that the intelligent agent cluster control method of the embodiment of the present specification achieves an extremely low End-to-End Delay (End to End Delay), and the average is about 1/32 times of that of TMP-OLSR in terms of synthesizing all simulation rounds. In the cluster scale 80 experiments, the end-to-end delay of TMP-P2P averaged approximately 1/53 times that of TMP-OLSR and 1/63 times that of TMP-DSDV. In addition, TMP-P2P is the only method to achieve 100% Packet transmission rate (Packet Delivery Ratio), i.e. there is no Packet loss.

Further, fig. 12 shows an End-to-End Delay (End to End Delay), an average number of packet forwarding times (Hop Count), an average Routing Traffic to each Agent (Routing Traffic Per Agent) and a Ratio of Routing Traffic to transmission data (Routing Load Ratio) in the case of using different Routing protocols when the same mobility procedure and Traffic occur are experienced, respectively, which reveals the reasons behind the result shown in fig. 11. The results shown in fig. 12 indicate that while the average number of packet forwarding is less when OLSR is used than when static routing (P2P) is used, OLSR generates about 2.93 to 12.25 times more traffic than the latter. Furthermore, the OLSR generates traffic of about 1.86 times the size of the transmission data for all the test rounds. Therefore, the test result shows that the intelligent agent cluster control method in the embodiment of the present specification significantly improves the communication performance of the intelligent agent network by greatly reducing the overhead caused by the dynamic routing protocol.

While the process flows described above include operations that occur in a particular order, it should be appreciated that the processes may include more or less operations that are performed sequentially or in parallel (e.g., using parallel processors or a multi-threaded environment).

Referring to fig. 13, in correspondence to the above-described agent cluster control method, an agent cluster control apparatus according to some embodiments of the present specification may include:

the determining module 131 may be configured to determine a shortest path between any two agents in a cluster of agents, and determine a constraint condition according to the shortest path, a communication distance of the agents in the cluster, and a sensing distance.

The selecting module 132 may be configured to invoke an RMA algorithm to select an agent from the cluster under the constraint of the constraint condition, so as to sense a newly added target point.

The establishing module 133 may be configured to establish a tree structure in the communication topology of the cluster by using the selected agent as a root node and according to a rule that all non-parent neighbor nodes are child nodes.

The moving module 134 may be configured to invoke a TMP algorithm to control distributed movement of the agents in the tree structure until the target point is sensed by the root node.

In the intelligent agent cluster control apparatus according to some embodiments of the present specification, the constraint condition may include: d_t(τ_new，a_c)≤r₁+r₂*l_a，c；

Wherein, tau_newAs a newly added target point, a_cAnchor points c, d for agent a_t(τ_new，a_c) At time t_newAnd a_cDistance between r₁Is the perceived distance of the agent, r₂Communication distance for agent,/_a，cIs the shortest path length between agent a and its anchor point c.

taking all neighbor nodes of the root node as child nodes of the root node;

In the agent cluster control device according to some embodiments of the present specification, the calling TMP algorithm controls agent distributed movement under the tree structure, including:

calling a nearest target point NRPT algorithm to find a target position; the target position is the position where the root node keeps the connection between the perceived target point and all the neighbor nodes of the root node and is closest to the moving target of the root node;

judging whether the target position satisfies d_t(τ，p_n)<d_t(τ, a); where τ is the moving target of the root node, p_nIs a target location, a is a root node, d_t(τ，p_n) At times t τ and p_nDistance between d_t(τ, a) is the distance between τ and a at time t;

In the agent cluster control device according to some embodiments of the present specification, the invoking the TMP algorithm to control the agent distributed movement under the tree structure further includes:

if the target position does not satisfy d_t(τ,p_n)<d_t(τ, a), invoking the BTNK algorithm to find the agent blocking the root node moves towards itA bottleneck node for target movement;

the agent receiving the assistance request takes the father node of the father node as the moving target, executes the step corresponding to the root node, and returns the assistance response after moving to a position closer to the moving target.

In the intelligent agent cluster control apparatus according to some embodiments of the present specification, the invoking NRPT algorithm to find the target location may include:

and finding the target position in the intersection area of all circles.

In the intelligent agent cluster control apparatus according to some embodiments of the present specification, the invoking the BTNK algorithm to find a bottleneck node that blocks the root node from moving to its moving target may include:

For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functions of the various elements may be implemented in the same one or more software and/or hardware implementations of the present description.

In some embodiments of the present description, fig. 14 shows a system configuration diagram when deploying an agent cluster control device to an agent (mobile robot or drone) cluster. Specifically, each agent may be configured with a Linux kernel computing device (e.g., Raspberry Pi 3b) on which an agent program (i.e., the agent cluster control apparatus described above) may be run. The agent program of the agent is responsible for communicating with the mobile controller on the agent at first, send the control command to it, collect and deal with the data that the agent of the agent perceives tentatively, the agent program of the agent will carry on the task decision at the same time, for example carry out according to the order of the control method of the cluster of the aforesaid agent of the agent, because may need to communicate with other agents in this process, the agent program of the agent can communicate in order to cooperate and finish the perception task. The agent program of the agent is also responsible for transmitting the primarily processed data back to the cluster monitoring center directly or by other agent programs of the agent, and receiving the task instruction of the agent program, such as the task allocation scheme output by the RMA algorithm.

In some embodiments of the present description, a cluster monitoring center program may be run on an electronic device (e.g., a personal computer), which has a graphical user interface on an upper layer to visually present a current cluster state and task execution to a user, and a network and communication component on a lower layer to communicate with the agent program to collect its perception information and send task instructions to it, and communicate with each component in the monitoring center program to input perception information, cluster state and accept their input, such as task related instructions. The core part of the cluster monitoring center program is functional components, such as a component responsible for safety monitoring, an information fusion processing component and a task allocation component related to a perception task, and the RMA algorithm is operated on the functional components to select an agent responsible for perceiving a new point of interest. In addition, other components related to specific perception tasks can be accommodated under the framework.

As shown in fig. 15, a computer device is also provided in some embodiments of the present description, the computer device 1502 may include one or more processors 1504, such as one or more Central Processing Units (CPUs) or Graphics Processors (GPUs), each of which may implement one or more hardware threads. The computer device 1502 may also include any memory 1506 for storing any kind of information such as code, settings, data, etc., and in a particular embodiment a computer program that runs on the memory 1506 and on the processor 1504 and that when executed by the processor 1504 may perform instructions according to the above described methods. For example, and without limitation, the memory 1506 may include any one or more of the following in combination: any type of RAM, any type of ROM, flash memory devices, hard disks, optical disks, etc. More generally, any memory may use any technology to store information. Further, any memory may provide volatile or non-volatile retention of information. Further, any memory may represent fixed or removable components of the computer device 1502. In one case, when the processor 1504 executes associated instructions stored in any memory or combination of memories, the computer device 1502 can perform any of the operations of the associated instructions. The computer device 1502 also includes one or more drive mechanisms 1508, such as a hard disk drive mechanism, an optical drive mechanism, and the like, for interacting with any of the memories.

The computer device 1502 may also include an input/output module 1510(I/O) for receiving various inputs (via input device 1512) and for providing various outputs (via output device 1514)). One particular output mechanism may include a presentation device 1516 and an associated graphical user interface 1518 (GUI). In other embodiments, input/output module 1510(I/O), input device 1512, and output device 1514 may also be excluded, as just one computer device in a network. The computer device 1502 may also include one or more network interfaces 1520 for exchanging data with other devices via one or more communication links 1522. One or more communication buses 1524 couple the above-described components together.

Communication link 1522 may be implemented in any manner, such as over a local area network, a wide area network (e.g., the Internet), a point-to-point connection, and the like, or any combination thereof. The communication link 1522 may comprise any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc., governed by any protocol or combination of protocols.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the specification. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processor to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processor, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processor to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processor to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

In a typical configuration, a computer device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.

Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium which can be used to store information that can be accessed by a computer device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.

As will be appreciated by one skilled in the art, embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, embodiments of the present description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and so forth) having computer-usable program code embodied therein.

The embodiments of this specification may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The described embodiments may also be practiced in distributed computing environments where tasks are performed by remote processors that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment. In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of an embodiment of the specification. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.

The above description is only an example of the present application and is not intended to limit the present application. Various modifications and changes may occur to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. An agent cluster control method, comprising:

under the constraint of the constraint condition, calling a maximum anchor point Retention (RMA) algorithm to select an agent from the cluster for sensing a newly added target point;

and calling a topology-invariant motion planning (TMP) algorithm to control the distributed movement of the agents under the tree structure until the target point is sensed by the root node.

2. The agent cluster control method of claim 1, wherein the constraints comprise: d_t(τ_new,a_c)≤r₁+r₂*l_a，c；

Wherein, tau_newAs a newly added target point, a_cAnchor points c, d for agent a_t(τ_new,a_c) At time t_newAnd a_cDistance between r₁Is the perceived distance of the agent, r₂Communication distance for agent,/_a，cIs the shortest path length between agent a and its anchor point c.

3. The agent cluster control method of claim 1, wherein invoking an RMA algorithm to select an agent from the cluster for sensing a newly added target point under the constraint of the constraint comprises:

wherein, the anchor point refers to: for any two agents a in the cluster_iAnd a_jIf a_iPositive and negativeBlame perceives at least one target point, and no other intelligent agent on the shortest path between the two perceives the target point, then a_iIs a_jAn anchor point of (a).

4. The method of claim 1, wherein the establishing a tree structure according to the rule that all non-parent neighbor nodes are child nodes comprises:

taking all neighbor nodes of the root node as child nodes of the root node;

5. The agent cluster control method of claim 1, wherein the calling TMP algorithm controls agent distributed movement under the tree structure, comprising:

judging whether the target position satisfies d_t(τ，p_n)<d_t(τ, a); where τ is the moving target of the root node, p_nIs a target location, a is a root node, d_t(τ,p_n) At times t τ and p_nDistance between d_t(τ, a) is the distance between τ and a at time t;

if the target position satisfies d_t(τ，p_n)<d_t(τ, a), moving the root node to the target position, and when the root node does not sense the newly added target point after the root node finishes moving, enabling the root node to search againAnd the target position is reached until the root node senses the newly added target point.

6. The agent cluster control method of claim 5, wherein the calling the TMP algorithm controls agent distributed mobility under the tree structure, further comprising:

if the target position does not satisfy d_t(τ,p_n)<d_t(τ, a), calling a bottleneck BTNK algorithm to find a bottleneck node which blocks the root node from moving to the moving target of the root node;

7. The agent cluster control method of claim 5, wherein said invoking the NRPT algorithm to find a target location comprises:

and finding the target position in the intersection area of all circles.

8. The agent cluster control method of claim 6, wherein the invoking the BTNK algorithm to find a bottleneck node that blocks the root node from moving to its moving target comprises:

9. An agent cluster control apparatus, comprising:

10. A computer device comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program, when executed by the processor, performs the instructions of the method of any one of claims 1-8.

11. A computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor of a computer device, executes instructions of a method according to any one of claims 1-8.