A multi-target search method for swarm robots based on triangular pyramid

Abstract

The present invention discloses a multi-target search method for swarm robots based on triangular cones. First, a method of expanding triangular cones is proposed, which can be applied to the prediction and avoidance process of various non-convex dynamic obstacles, which can not only effectively avoid obstacles but also reduce unnecessary obstacle avoidance processes; secondly, according to the two different states of the search process, a roaming direction cone and a target search cone are proposed for the roaming process and the collaborative search process respectively. The roaming direction cone improves the ability to develop global unknown environments, while the target search cone effectively enhances the local search ability. Both methods can effectively improve the search efficiency and can decide whether to change the direction, thereby reducing unnecessary rotation direction processes, that is, effectively reducing the search time consumption, energy loss and mechanical loss caused by the robot rotation direction process. Finally, the three improved methods are combined to construct a swarm robot multi-target search strategy to realize swarm robot multi-target task search.

Description

Technical Field

[0001] The present invention relates to the technical field of robot target search, and in particular to a multi-target search method of a group robot based on a triangular pyramid. Background Art

[0002] Swarm robots inspired by nature are a combination of intelligent swarms and robotics technology. Due to their simple structure, low cost, high robustness and flexibility, they can often be used in many complex and dangerous scenarios, such as post-disaster search and rescue, resource exploration, target search, etc.

[0003] In recent years, the application research of robots in target search has made great progress. Considering the limitations of multi-target search of swarm robots and the characteristics of their workspace, Dadgar et al. proposed a distributed target search algorithm based on particle swarm optimization, which not only considers obstacle avoidance but also avoids falling into local optimality. Tang et al. proposed an improved grouping strategy based on constraint factor particle swarm optimization to solve the problem of collaborative search of multiple targets by swarm robots in unknown environments, and considered obstacle avoidance during the search process; simulation results show that this method has good adaptability, accuracy and efficiency. Li proposed a new search strategy based on probabilistic finite state machine, which has good stability and can achieve a good balance between exploration and development, parallelism and collaboration. DU proposed a distributed particle swarm algorithm, which takes into account the communication limitations and communication energy consumption of robots while performing swarm robot target search, and proposed a dynamic group partitioning method for the situation where multiple targets must be searched simultaneously for limited processing. He proposed a multi-target search algorithm based on 3D simplified virtual forced model, which effectively solved the multi-target search problem and real-time obstacle avoidance problem of drone swarm in complex environment. Tang et al. proposed a system solution based on mechanical particle swarm optimization and artificial potential field. This method not only closely considers the actual application of real robots, but also includes obstacle avoidance capabilities based on module switching strategies, and verifies the effectiveness and robustness of this method through simulation and experiments. Wang et al. proposed a boundary scanning obstacle avoidance strategy and a particle swarm algorithm for target position estimation to address the shortcomings of obstacle avoidance and search efficiency, effectively improving obstacle avoidance and search efficiency.

[0004] However, in most studies, the obstacles faced by swarm robots are mostly simple and static, and few studies have modeled complex dynamic non-convex obstacles. Most obstacle avoidance strategies adopt potential field methods, simplified virtual forces, collision cones, etc. Search algorithms are mainly divided into two categories. One is inspired by swarm intelligence algorithms, such as particle swarm algorithms, ant colonies, wolf packs, etc., and the other is inspired by other methods, such as the method that robots can locate according to odor sources. However, whether it is a search algorithm or an obstacle avoidance strategy, swarm robots immediately respond to trends according to their own algorithms, that is, during the task search process, swarm robots need to turn their direction at all times to search for targets, but excessive rotation not only requires additional energy, but also causes certain damage to the robot itself, and mobile robots do not need to turn their direction at all times during the search process. If the necessity of turning the direction is judged before deciding the next move, the number of turns can be reduced while ensuring search efficiency, thereby reducing energy consumption. Summary of the invention

[0005] The technical problem to be solved by the present invention is: the present invention provides a multi-target search method of a swarm robot based on a triangular cone, which focuses on the necessity of changing direction while improving the various capabilities of the robot, and reduces the time consumption, energy consumption and mechanical wear caused by unnecessary turning processes.

[0006] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0007] A multi-target search method for swarm robots based on triangular cones. Before performing a task search, the multi-target search method first initializes the task environment. The initialization parameters at least include swarm robot system parameters, obstacle parameters, and target parameters. The swarm robot system parameters include population size, robot position, and speed information. When the robot fails to detect target information, it is in a roaming search state, and the robot develops a global environment based on the roaming direction cone. After detecting a target signal, the robot switches to a collaborative search state, and uses a target search cone model for local search. The robot is provided with an avoidance strategy, which has the highest priority. The robot senses a threat within the detection range and uses the avoidance strategy for predictive avoidance. When the target signal detected by the robot reaches a preset threshold, the search is deemed to be successful, and when all targets are successfully searched, the search is completed.

[0008] As a further improvement of the above technical solution:

[0009] Preferably, the multi-target search method specifically comprises the following steps:

[0010] Step S1: Modeling

[0011] Construct a module for swarm robots to search for multi-target tasks in a two-dimensional task environment, including a task environment model, a target incentive model, and a task allocation model. The task environment model constructs the task environment required for the swarm robot's multi-target task search; the target incentive model is used to construct a signal model sent by the robot's perception of the target; and the task allocation model is used to construct the resource allocation and sub-alliance formation process of the swarm robot when searching for the target.

[0012] Step S2: Multi-target search strategy of swarm robots based on triangular cones

[0013] An expanded triangular cone model suitable for non-convex obstacles, a roaming direction cone search method and a target search cone search method are designed for searching; the roaming direction cone search method is used for robots in the roaming state to explore unknown global environments, and the target search cone search method is used for robots in the collaborative state to conduct local searches; then, an avoidance strategy is designed based on the combination of the collision cone and the expanded triangular cone for predictive avoidance; the robot searches according to the search strategy and eventually completes all search targets.

[0014] Preferably, in step S1, when the group robots perform multi-target search in a non-convex obstacle environment, the task environment is first initialized, wherein the obstacles include static and dynamic non-convex obstacles; when the robots perform task search in an unknown dynamic environment, the next moving direction is selected according to the searched information, and obstacles and other mobile robots are avoided during the exploration process; when the robots explore in an unknown task environment, target search is performed according to the target incentive model; when the group robots perform task search in a dynamic environment, information is exchanged within the communication range, and a sub-alliance is formed to collaboratively search for the same target.

[0015] Preferably, in step S2, the process of designing an expanded triangular cone model suitable for non-convex obstacles is: processing the obstacle information collected within the detection range of the robot, setting a triangular cone that can surround all obstacles within the detection range, the robot mass point and the two intersection points of the triangular cone and the obstacle form a triangle, and obtaining parameter information by obtaining the information of the inscribed circle of the triangle, and expanding the triangular cone in combination with the size of the robot to obtain an expanded triangular cone.

[0016] Preferably, the design is applicable to an expanded triangular cone model of a non-convex obstacle, comprising the following steps:

[0017] (1) Discretize the threats within the detection range and obtain the corresponding coordinates:

[0018]

[0019] The robot obtains its own position through the sensor. Suppose the position of the robot at a certain moment is calculate To each coordinate vector, get the angle set between each two vectors;:

[0020]

[0021] Find the largest angle and get its corresponding two coordinates: Assume the robot's position at this moment is Then the incircle of the triangle formed by the three points is P O =[x o ,y o ], the radius of the triangular cone threat is r TC ;

[0022] Calculate the semi-apex angle β1 of the triangular cone and the angle between the relative velocity and the center of the inscribed circle

[0023]

[0024]

[0025] (2) According to the size of the robot, the triangular cone is expanded, and the part that needs to be expanded is β2:

[0026]

[0027] Among them, the size radius of the robot is R r , The current robot R i With obstacles O s1 , O s2 The Euclidean distance between them; the half vertex angle of the expanded triangular pyramid β=β1+2*β2;

[0028] (3) Design an avoidance strategy based on collision cone and expanding triangular cone: the collision cone model is used for collision avoidance between robots, and the expanding triangular cone model is used for static and dynamic non-convex obstacles.

[0029] Preferably, the collision cone model is constructed as follows: the robot is regarded as a particle, and the obstacle is expanded into a threat circle with a radius r according to the threat radius of the obstacle and the size of the robot. A ray l is drawn from the robot position, and the intersection of l and the threat circle is M. The set of all rays drawn from the robot position that can intersect with the threat circle is collectively referred to as a collision cone. The relative velocity vector between the robot and the obstacle falls outside the collision cone area, and the robot avoids collision with the obstacle.

[0030] Preferably, the roaming direction cone search method adopts a roaming search strategy based on the roaming direction cone, and the roaming search strategy is to increase the step length of each step of the robot and avoid the previous position direction of the robot; the roaming direction cone takes the approach of avoiding all robots within the maximum distance range that the robot can detect the target; before deciding on the next moving direction, the robot determines the necessity of turning direction through the roaming direction cone, and if the speed direction at the previous moment falls within the feasible range of the roaming direction cone, there is no need to turn the direction.

[0031] Preferably, the target search cone search method adopts a collaborative search strategy based on the target search cone, including the following three cases:

[0032] (1) When the current position of the robot is the individual optimal or the population optimal, the robot first predicts the robot position at the next moment based on the current direction. The robot position at the next moment is used as the center of the target search cone, and the robot radius is used as the radius of the target search cone.

[0033] (2) When the robot's current position is neither the individual optimal nor the group optimal and the individual optimal position is different from the group optimal position, the robot uses the angle between the individual optimal and the group optimal as the next search direction range;

[0034] (3) When the current position of the robot is neither the individual optimal nor the population optimal, but the individual optimal position is equal to the population optimal position, the robot uses the individual optimal position as the center and the maximum speed of the robot as the target search cone radius to expand the search range.

[0035] The multi-target search method of swarm robots based on triangular pyramids provided by the present invention has the following advantages compared with the prior art:

[0036] (1) The triangular pyramid-based swarm robot multi-target search method of the present invention aims at the obstacle avoidance ability problem, global development ability problem and local search ability problem in an unknown dynamic non-convex obstacle environment. While improving various abilities, it focuses on the necessity of changing direction, and reduces the time consumption, energy consumption and mechanical wear caused by unnecessary corner turning processes. The MSTC strategy is proposed in three aspects. For the obstacle avoidance problem, the ETC obstacle avoidance model can better solve the obstacle avoidance problem in a complex dynamic non-convex obstacle environment; the RDC strategy effectively improves the robot's ability to develop an unknown environment, thereby improving the efficiency of completing task searches; for the local search ability problem, the TSC strategy can prevent the robot from falling into the local optimum while effectively improving the local search ability, and ultimately improving the efficiency of completing target searches. In addition, the three methods can first determine the necessity of the direction of rotation before moving at the next moment, thereby reducing the number of corner turns.

[0037] (2) The present invention's swarm robot multi-target search method based on triangular cone focuses on the necessity of changing direction for collision avoidance and multi-target search problems in unknown dynamic non-convex obstacle environments, thereby reducing unnecessary corner turning processes of swarm robots during task search, that is, reducing the time consumption, energy consumption and mechanical wear caused by the corner turning process. For the obstacle avoidance problem, the present invention improves the collision cone model and proposes a collision avoidance model suitable for complex dynamic non-convex obstacles, namely, the expanded triangular cone (ETC). This method can not only effectively avoid obstacles, but also predict collision conflicts to reduce unnecessary collision avoidance processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a task environment diagram of the swarm robot multi-target search of the present invention.

[0039] Figure 2 Schematic diagram of the three state transition relationships of the robot in an embodiment of the present invention.

[0040] Figure 3 Schematic diagram of a collision cone model in an embodiment of the present invention.

[0041] Figure 4 Schematic diagram of an expanded triangular pyramid model in an embodiment of the present invention.

[0042] Figure 5 Schematic diagram of a robot collision avoidance model in an embodiment of the present invention.

[0043] Figure 6 Schematic diagram of the roaming direction cone model in an embodiment of the present invention.

[0044] FIG. 7( a ) is a schematic diagram of a target search cone model in which the current position of the robot is the individual optimal / population optimal position in an embodiment of the present invention.

[0045] 7( b ) is a schematic diagram of a target search cone model in which the current position of the robot is neither the individual optimal nor the population optimal and the individual optimal position is different from the population optimal position in an embodiment of the present invention.

[0046] 7( c ) is a schematic diagram of a target search cone model in which the current position of the robot is neither the individual optimal position nor the population optimal position, but the individual optimal position is equal to the population optimal position in an embodiment of the present invention.

[0047] Figure 8 It is a flow chart of multi-target search of group robots in an unknown dynamic complex non-convex obstacle environment of the present invention.

[0048] FIG. 9( a ) is an environmental diagram at the initial moment in the RDC simulation experiment when the method of the present invention is verified.

[0049] FIG. 9( b ) is an environment diagram of T=30-40 in the RDC simulation experiment when verifying the method of the present invention.

[0050] FIG9( c ) is an environment diagram of T=60-70 in the RDC and ETC simulation experiments when the method of the present invention is verified.

[0051] FIG9( d ) is an environment diagram of T=70-89 in the TSC and ETC simulation experiments when verifying the method of the present invention.

[0052] FIG. 9( e ) is an environment diagram of T=140-167 in a TSC simulation experiment when verifying the method of the present invention.

[0053] FIG9( f ) is an environment diagram of T=229 in a TSC simulation experiment when verifying the method of the present invention.

[0054] FIG9( g ) is an environment diagram of T=229 in a TSC simulation experiment when verifying the method of the present invention.

[0055] FIG9(h) is a schematic diagram of the entire path of the robot that finds the target position during the entire search process when verifying the method of the present invention.

[0056] FIG. 10( a ) is a statistical diagram comparing the relationship between the number of swarm robot populations and the number of iterations under six modes when the method of the present invention is verified.

[0057] FIG10( b ) is a comparative statistical diagram of the relationship between the number of robot swarms and the number of corner turns under six modes when the method of the present invention is verified. DETAILED DESCRIPTION

[0058] The specific embodiments of the present invention are described in detail below. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0059] like Figures 1 to 4 As shown, the multi-target search method of a swarm robot based on a triangular pyramid of the present invention comprises the following steps:

[0060] Step S1: Modeling

[0061] The multi-target task search of swarm robots in a two-dimensional task environment requires the construction of multiple modules, which are mainly divided into: task environment model, target incentive model and task allocation model. Among them, the task environment model mainly constructs the task environment required for the multi-target task search of swarm robots, including robots, targets, obstacles, etc.; the target incentive model is mainly used to build the signal model sent by the robot to perceive the target; the task allocation model is used for the resource allocation and sub-alliance formation process of the swarm robot when searching for the target.

[0062] Step S1-1, task environment model

[0063] When a group of robots conducts multi-target search in a two-dimensional dynamic non-convex obstacle environment, its task environment mainly consists of three parts: robot R i {i=1,2,…,N R}、Target T j {j=1,2,…,N T} and obstacles O s {s=1,2,…,N O Among them, obstacles include static and dynamic non-convex obstacles, such as Figure 1 The figure shows a task environment diagram of a swarm robot multi-target search. The task environment is 1000×1000unit, the small dots '●' represent robots, the five-pointed stars '☆' represent targets to be searched, and the gray irregular objects are obstacles that the robot needs to avoid. When the robot performs task search in an unknown dynamic environment, it autonomously selects the next moving direction based on the searched information, and avoids static and dynamic non-convex obstacles and other mobile robots during its exploration. The target is unknown, and the robot searches for the target according to the target incentive model when exploring in the task environment. The position information of the dynamic obstacle at a certain moment follows the following movement model:

[0064]

[0065] in, is the obstacle O at time t s The position vector, is the coordinate position of the obstacle in the two-dimensional Cartesian coordinate system, is the obstacle O at time t s The velocity vector, is the obstacle O at time t s The angle with the positive x-axis, V om The maximum speed at which obstacles move on the map.

[0066] When searching for tasks, the group robots can sense the surrounding environment based on the sensors they carry and make reaction trends accordingly. When the sensed distance is less than a certain threshold, the robot selects the obstacle avoidance mechanism and determines the next action based on the obstacle avoidance mechanism. In addition, during the movement of the group robots, they can com They communicate with each other to exchange information and collaborate in searching for targets.

[0067] When a group of robots performs a task search in a dynamic environment, the task target is unknown, but the robot can sense the signal sent by the target based on the sensor it carries. Assuming that the strength of the target signal sensed by the robot is positively correlated with the distance from the robot to the target, when the target signal strength sensed by the robot reaches a certain threshold I lWhen , the target is considered to be successfully searched; when all targets in the task environment are successfully searched, the multi-target search task of the swarm robot is completed and the robot stops the task search.

[0068] Step S1-2, target incentive model

[0069] When a group of robots performs a task search in a dynamic environment, the target is unknown in advance, but they can sense the signal emitted by the target based on the sensors they carry. The detection incentive model of the signal emitted by the target is as follows:

[0070]

[0071] Among them, I ij Indicates robot R i To target T j The response function is related to the distance d from the robot to the target and is inversely proportional to the square of the distance; m represents the attenuation coefficient of the target signal when it propagates in a dynamic environment; Q is the constant power emitted by the sensor carried by the target; rand() represents the random disturbance of the target signal when it propagates in a dynamic environment; d m Indicates the maximum sensing distance at which the robot can detect the target signal.

[0072] Step S1-3, task allocation model

[0073] When swarm robots perform task search in a dynamic environment, they can exchange information within their communication range and work together to quickly search for the target. Therefore, when swarm robots perform task search, their states can be divided into three types: roaming search status (RSS), collaborative search status (CSS) and declaration status (DC), such as Figure 2 As shown in the figure, the migration relationship between the three states of the robot during task search. In the collaborative search state, the robot can form a sub-alliance to search for the same target together. If the robot has multiple candidate targets at the same time, it will give priority to the same target as the previous step. If there is no same target, it will select the current only target according to the roulette method. The roulette formula is as follows:

[0074]

[0075]

[0076] Among them, k i Indicates robot R i As the number of candidate targets, p ijIndicates the probability that the alternative target j is selected. P(i,j) refers to the cumulative probability of the alternative target j. When P(i,j - 1) < r a <P(i,j), the robot R i selects the target T j as the target for collaborative search, where r a is a uniform random number within the interval (0,1).

[0077] Table 1 is the sorting mechanism of subgroup alliance sub9 when T = 60. Among them, type Ⅰ refers to the robots that can directly detect target information, and type Ⅱ refers to the robots that can obtain target information only through communication with type Ⅰ. According to the priority principle of selecting robots in the sub-alliance mechanism, its priority sorting is: R 19 >R 33 >R6>R 28 >R 18 >R 39 >R 34 >R 11 >R8.

[0078] Table 1 Subgroup sub9 alliance ranking mechanism (T = 60)

[0079]

[0080] It is set that when the subgroup alliance searches for the same target, if the number of subgroup alliance members is too large, it will lead to overcrowding in the search area. At this time, the robots give priority to collision avoidance and then select to approach the target. It can be seen that too many members may consume too much search time. Set the number of sub-alliance members N m to follow formula (5):

[0081]

[0082] where, N Ⅰ represents the number of type Ⅰ robots. When the number of type Ⅰ robots is too small, it is necessary to urgently recruit alliance members to ensure the diversity of the alliance. When the number of type Ⅰ robots reaches a certain value n m at this time, the number of alliance members does not need to be too large to avoid congestion.

[0083] Step S2, Multi-target search strategy of swarm robots based on triangular pyramid

[0084] When a group of robots performs task search in an unknown non-convex complex environment, in order to find all targets as quickly as possible, a common method is to reduce the number of iterations, thus ignoring the time, energy, and mechanical wear consumed by each robot rotation direction. In this embodiment, an ETC model suitable for dynamic non-convex obstacles, a roaming direction cone and a target search cone method for search are proposed, which can not only safely and effectively avoid static and dynamic non-convex obstacles, but also avoid unnecessary collision avoidance processes in advance; the roaming direction cone and the target search cone respectively enhance the robot's ability to develop an unknown global environment and explore a local environment. Therefore, these three methods can effectively reduce the robot's unnecessary corner turning process during movement while improving the search efficiency, thereby further reducing the time consumption, energy consumption, and mechanical wear caused by the corner turning process.

[0085] Step S2-1: Constructing a swarm robot system motion control model

[0086] The movement of swarm robots in a dynamic non-convex obstacle environment follows the following motion model:

[0087]

[0088] in, is the robot R at time t i The position vector, is the coordinate position of the robot in the two-dimensional Cartesian coordinate system, is the robot R at time t i The velocity vector, is the robot R at time t i The angle with the positive x-axis, V rm The maximum speed of the robot moving on the map.

[0089] Step S2-2, avoidance strategy of group robots in the task environment

[0090] In the existing research on multi-target search methods of swarm robots, in order to facilitate the analysis of results, the obstacles set are mostly simple in structure, single in shape, and mostly static. However, in the actual search environment, obstacles are mostly uncertain factors, that is, complex, non-convex and dynamic environments. Since the collision cone can predict the movement direction of the obstacle and can be judged by simple geometric relationships, it is widely used in dynamic obstacle avoidance. However, the obstacles avoided by the collision cone are mostly simple, single obstacles such as circles or balls.

[0091] In this embodiment, a model is built for complex dynamic non-convex obstacles, and a triangular cone obstacle avoidance model is applied to predict the movement trajectory of the robot and the obstacle in real time to determine whether avoidance is required, and then the next movement is decided based on the judgment result. The triangular cone model for non-convex obstacle avoidance is combined with the collision cone model for robots to establish an obstacle avoidance strategy for the group robot system in the task environment. In this embodiment, in order to explain that the non-convex static and dynamic obstacles and other robots detected within the robot's obstacle avoidance range are collectively referred to as threats, the design of the avoidance strategy includes the following steps:

[0092] Step S2-2-1, construct collision cone model

[0093] The robot is regarded as a particle, and the obstacle is collided into a threat circle with a radius of r according to the threat radius of the obstacle and the size of the robot. Suppose a ray l is drawn from the position of the robot, and the intersection of l and the threat circle is M, that is, the set of all rays drawn from the position of the robot that can intersect with the threat circle is collectively called the collision cone. The relative velocity vector of the robot and the obstacle falls outside the collision cone area, and the robot can avoid collision with the obstacle.

[0094] like Figure 3 Shown is a collision cone model, where is the relative velocity vector V RE With the position vector R i E s The angle, R i E s Indicates robot R i To obstacle E s The vector of Respectively represent R i E s The values ​​projected on the x and y axes; is the relative velocity vector V RE The angle with the x-axis, where the relative velocity vector refers to the robot R i Velocity Vector With obstacle E s Velocity Vector The relative velocity vector; β is the semi-vertex angle of CC, r is the obstacle E s The threat radius, YesR i E s The angle between the x-axis and the When the obstacle E s Threatening, Robot R i Need to E s To avoid collision; When E s Non-threatening, robot R iNo collision avoidance is required. Solving the relevant parameters yields:

[0095]

[0096]

[0097]

[0098]

[0099] Step S2-2-2, dynamic non-convex obstacle avoidance model based on ETC

[0100] When the robot moves in a complex dynamic non-convex obstacle environment, it does not know the specific position and shape of the obstacle in advance. It can only determine whether there is an obstacle within its detection range through the sensors it carries, and return the detected obstacle information to the robot system. The robot system then determines whether it needs to avoid the obstacle based on its own algorithm (the algorithm that the robot has set during operation is a conventional setting method for the robot and will not be described in detail here) and responds to the trend based on the judgment result.

[0101] Traditional collision cone obstacle avoidance methods mostly transform threats into circles or ellipses with a certain threat radius. However, in actual search environments, obstacles are mostly irregular patterns. If all obstacles are transformed into special shapes such as circles or ellipses, too much search resource environment will be wasted, resulting in unnecessary collision avoidance processes. This embodiment adopts the method of expanding triangular cones, which can effectively reduce unnecessary avoidance processes while solving the problem of avoiding complex non-convex obstacles. Figure 4 The figure shows the obstacle avoidance model when the robot detects obstacle information in a dynamic environment. Assume that the robot's detection radius is d r , process the obstacle information collected within the detection range, and find the triangular cone that can completely surround the obstacles within the detection range, so that the robot mass point and the two intersection points of the triangular cone and the obstacle can form a triangle, and by obtaining the information of the inscribed circle of the triangle, the relevant parameter information similar to the collision cone can be obtained. Considering that the robot has a certain size, the triangular cone is expanded according to the size of the robot to obtain an expanded triangular cone.

[0102] The specific steps include:

[0103] Step S2-2-2-1, discretize the threat within the detection range and obtain the corresponding coordinates:

[0104]

[0105] The robot obtains its own position through the sensor. Suppose the position of the robot at a certain moment is calculate To each coordinate vector, get the angle set between every two vectors;

[0106]

[0107] Find the largest angle and get its corresponding two coordinates: The robot's own position is obtained by the sensor. Suppose the robot's position at this moment is Then the incircle P of the triangle formed by the three points O =[x o ,y o ] and the radius of the triangular cone threat r TC It can be concluded.

[0108] Calculate the semi-apex angle β1 of the triangular cone and the angle between the relative velocity and the center of the inscribed circle

[0109]

[0110]

[0111] Step S2-2-2-2, considering that the robot has size and dimensions, in order to make the robot move safely, the triangular cone should be expanded according to the size of the robot when calculating it. Then, the part that needs to be expanded is:

[0112]

[0113] Among them, the size radius of the robot is R r , The current robot R i With obstacles O s1 , O s2 Therefore, the half vertex angle β of the expanded triangular pyramid is β = β1 + 2*β2.

[0114] Step S2-2-2-3, design an avoidance strategy based on collision cone and expanded triangular cone

[0115] The premise for a group of robots to complete a task is to ensure their own safe movement. Therefore, safe and effective robot avoidance is the primary consideration. Avoidance not only requires avoiding static and dynamic non-convex obstacles, but also ensuring that robots can avoid each other safely and effectively.

[0116] The robot in this embodiment has a radius of R r For robots, collision avoidance between robots can use the collision cone model, while static and dynamic non-convex obstacles use the expansion triangular cone obstacle avoidance model. Figure 5is the avoidance model of the robot within its detection range at a certain moment. According to the surrounding threats detected by its own sensors, the robot's next feasible range is calculated according to the expansion triangle cone or collision cone model, and a direction is randomly selected within the feasible area as the next speed direction. The CC threat radius used for robot collision avoidance is r = 2 × R r .

[0117] Step S2-2-3, roaming search strategy based on RDC

[0118] When a group of robots conducts mobile search in a task environment, the biggest reason for the low search efficiency is repeated search, which mainly comes from two sources: 1) repeated search generated by the robot itself; 2) repeated search generated by other robots.

[0119] In this embodiment, in order to avoid repeated searches by the robot, that is, when the robot performs task search in a dynamic environment, the same robot repeatedly searches the same area. For this problem, the roaming direction cone solves it by increasing the step length of each step of the robot and avoiding the previous position direction of the robot. Therefore, during the roaming search process, the robot will move with the maximum step length and avoid the position of the previous moment.

[0120] In this embodiment, in order to avoid repeated searches by other robots, that is, the area explored by the robot may be an area that has been explored by other robots, the roaming direction cone takes the approach of avoiding the robot d m All robots in range, where d m is the maximum distance that the robot can detect the target, avoiding d m The robots within the range can prevent missing the target and can also make the robots evenly distributed in the developed area. The avoidance angle range of the robot is related to the distance between the robots, that is, the smaller the distance, the larger the avoidance range, where d0 = V rm +R r Before deciding the next moving direction, the robot can use the roaming direction cone to determine the necessity of turning in advance. If the speed direction at the previous moment falls within the feasible range of the roaming direction cone, there is no need to turn.

[0121]

[0122] like Figure 6 As shown, the robot R at time t i The roaming direction cone, at the roaming moment, the robot moves at the maximum speed V rm And it can avoid the surrounding robots. The measures taken are: the robot takes the position at time t-1 as the center of the roaming direction cone, the semi-vertex angle is 90 degrees, and the robot R at time t is j1 , Rj2 , R j3 The position is the center of the circle, and the semi-vertex angle is β * Follow formula (16).

[0123] The speed update formula of the robot in the roaming state is:

[0124]

[0125] Where RDC represents the walk direction cone.

[0126] Step S2-2-4, collaborative search strategy based on target search cone

[0127] Traditional robots mostly use particle swarm algorithms when performing multi-target search, but particle swarms are prone to fall into local optimality and convergence speed is not fast enough. In this embodiment, based on the multi-target search strategy of the target search cone, when the robot performs task search in a dynamic non-convex obstacle environment, when the signal sent by the target is detected, the robot's state changes from the roaming state to the collaborative search state.

[0128] The target search cone is based on the area between the optimal direction of the group and the optimal direction of the individual as the next search range, so that the group robot can quickly approach the target while maintaining the diversity of the population. It is mainly divided into two aspects: direction and step length. In terms of direction, the target direction range is predicted and the robot is turned to the target direction range; the predicted target direction range does not point to a single specific factor. In the collaborative search process, in order to ensure the diversity of the population, the members of the alliance generally consider their inertia factors, cognitive factors and social factors comprehensively, and the target search cone also considers these factors according to different situations during the group robot search process. Including the following three situations:

[0129] (1) When the current position of the robot is the individual optimal or the population optimal, the robot considers its inertia factor more. Therefore, the robot first predicts the robot position at the next moment based on the current direction, and uses this position as the center of the target search cone, and the robot radius as the target search cone radius, such as Figure 7a ) as shown.

[0130] (2) When the robot's current position is neither the individual optimal nor the group optimal, and the individual optimal position is different from the group optimal position, the robot comprehensively considers its cognitive factors and social factors, that is, the angle between the individual optimal and the group optimal is used as the next search direction range, such as Figure 7b ).

[0131] (3) When the current position of the robot is neither the individual optimal position nor the population optimal position, but the individual optimal position is equal to the population optimal position, such as Figure 7c), in this case, cognitive factors and social factors are the same, and only one of them needs to be considered. In order to expand the diversity of the population, the robot will take the individual optimal position as the center, V rm The radius of the target search cone to expand its search range.

[0132] The robot follows the above method to make the speed direction fall into the target search cone range and approach the target. Each time it approaches, it does not approach a single direction, but randomly selects a direction within a certain range, which can avoid falling into the local optimum and increase the diversity of the population. In addition, the target search cone can be used to determine whether the direction needs to be turned, so as to reduce unnecessary corner turning processes.

[0133] Furthermore, in the process of multi-target search by swarm robots, a step length that is too long may cause the target to be missed, resulting in unnecessary repeated searches and increased energy consumption, while a step length that is too short will result in a longer search time. In this embodiment, a nonlinear decreasing step length function based on the target response signal is adopted. The step length is related to the detected target signal. When the detected target signal is smaller, it means that the robot is farther away from the target. At this time, the step length should be increased. When the detected target signal is larger, it means that the robot is closer to the target. At this time, the step length should be reduced to avoid missing the detected target. The nonlinear decreasing step length function based on the target response signal is:

[0134]

[0135] Among them, I l is the target signal threshold, I ij (t) is the robot R in the collaborative search state at time t i With target T j The signal strength of l is controlled within the range of (l min ,l max ).

[0136] The speed update formula of the robot in the roaming state is:

[0137]

[0138] Where TSC represents the target search cone.

[0139] Then the displacement update formula of the group robot in the two-dimensional environment is:

[0140]

[0141] Step S2-3, group robot multi-target search control strategy

[0142] like Figure 8As shown in the figure, it is a flowchart of multi-target search of swarm robots in a two-dimensional unknown complex dynamic non-convex obstacle environment. Before the task search is carried out, the task environment is initialized first, including the initialization of parameters such as the swarm robot system, obstacles, and targets. The parameterization of the swarm robot system includes the population size, robot position and speed information; the swarm robot maintains a roaming search state before detecting the target information, and based on RDC, the robot quickly develops the global environment; after detecting the target signal, the robot switches to a collaborative search state and uses the TSC model for local search; the obstacle avoidance process runs through the entire process and has the highest priority, that is, no matter whether the robot is in a roaming state or a collaborative search state, as long as the robot is in d m When a threat is detected within the range, an avoidance strategy combining a collision cone and an expanding triangular cone is used for predictive avoidance; when the target signal detected by the robot reaches the preset threshold, the target is considered to have been searched successfully, and when all targets have been searched successfully, the search is completed.

[0143] Method validation

[0144] Verifying the search method in this embodiment includes the following steps:

[0145] Step 1: Establish system parameter settings and swarm robot task search performance evaluation index model. The task environment parameter settings are shown in Table 2.

[0146] Table 2 Task environment parameter settings

[0147] symbol significance Settings <![CDATA[N R ]]> Number of robots in the group 30～100 <![CDATA[N T ]]> Target quantity 10 <![CDATA[N O ]]> Number of obstacles 7 S Search area(unit) 1000×1000 <![CDATA[R r ]]> Robot radius 2 <![CDATA[V rm ]]> Maximum speed of robot 10 <![CDATA[V om ]]> Dynamic obstacle running speed 1 <![CDATA[d m ]]> Maximum sensing radius of targets and threats 30 <![CDATA[d com ]]> Robot maximum communication distance 300 Q Target signal energy <![CDATA[10 5 ]]> m Target signal attenuation coefficient 0.1 <![CDATA[I l ]]> Target signal threshold 400 <![CDATA[n m ]]> Class I Robot Number Threshold 4 <![CDATA[l max ]]> Maximum step length 10 <![CDATA[l min ]]> Minimum step size 5

[0148] The performance of group robot task search is evaluated from two aspects: system time consumption T r and the number of corners N a , where the system time consumption is the number of iterations when the task is completed, and the number of corner turns refers to the number of times the robot changes its direction at the previous moment during the search process. Let T r1 、T r2 、T r3 , N a1t 、N a2t 、N a3t , N R1t 、N R2t 、N R3t are the time consumption, number of corner turns and number of robots at the roaming time, collaborative search time and declaration time at time t, respectively, where N R|t =N R1|t +N R2|t +N R3|t , t k For a single target T k If the search is successful, then t k With T r , Na With N a1|t 、N a2|t 、N a3|t , there is the following relationship:

[0149]

[0150]

[0151]

[0152] Among them, N Rk The search target T k The number of sub-coalition members formed. Since the robot in the declaration state no longer participates in other search tasks, its speed is 0 and its position remains unchanged. Therefore, the number of corners that the robot in the declaration state searches for subsequent tasks is N. a No impact, that is, N a3 =0.

[0153] Step 2: Verification test

[0154] In the search process of two-dimensional unknown dynamic environment, complex static and dynamic non-convex obstacles are considered, with N R =50 as an example, the experimental process of multi-target search demonstration of swarm robots based on the multi-target search strategy (MSTC) is specifically described, as shown in Figure 9.

[0155] Step 2.1: Simulation experiment process based on expanded triangular cone (ETC)

[0156] Figure 9(c) , 9(d) The avoidance process based on ETC is demonstrated. In Figure 9(c), the robot R 12 The vehicle detects obstacle information within its detection range and determines based on ETC that it will encounter obstacles if it continues to move forward in the current direction. Therefore, R 12 According to ETC, the robot calculates its feasible range and avoids obstacles. In Figure 9(d), robot R6 detects obstacle information, but according to ETC, it determines that there will be no collision if it moves at the current moment, so robot R6 does not need to change direction; 36 , R 44 When obstacle information is detected, ETC determines that a collision will occur if the vehicle moves in the current direction. 36 , R 44 Calculate the feasible range according to ETC and randomly select a direction to move in the feasible range; similarly, R 45 When the obstacle information is detected, the robot R 45 Change direction according to ETC; and R 49It is determined that no collision will occur, so no direction change is required.

[0157] Step 2.2: Simulation experiment process based on RDC

[0158] At the initial moment, the robots are in 100×100 units, their positions are random, and their current states are all roaming search states. Therefore, they will spread around to develop the environment according to the RDC method, such as Figure 9a ). Figure 9(b) shows the process of the robot in the roaming state developing the global environment according to RDC at time T=30-40. At this time, the robot is rapidly developing the unknown environment, and the target 9 has been 23 The search was successful. Figure 9(c) shows the process of the robot in the roaming state developing the global environment according to RDC during T = 60-70. It can be seen that the robot is exploring the unknown environment. This strategy can traverse the search space at the maximum speed without obtaining prior information, and has a strong ability to develop unknown environments.

[0159] Step 2.3: Simulation experiment process based on target search cone (TSC)

[0160] Figure 9(d) shows the process of searching for target 10 based on TSC from T = 70 to 89. When T = 70, robot R 38 , R 16 As a type I robot, it detects the signal from target 10 and quickly forms a sub-alliance sub10. At this time, the number of sub-alliance members is less than Nm, so sub10 urgently publishes recruitment information to the surrounding areas, while R8, R 15 After receiving the information, the robot R in the sub-alliance sub10 joins the sub-alliance sub10 as a Class II robot and searches for target 10 together according to the TSC strategy. 15 After successfully searching for target 10, sub10 disbands and the remaining members switch to roaming search mode to continue searching for other robots.

[0161] Figure 9(e) shows the search process of the sub-alliance sub7 with T=140-167 according to the TSC strategy. When T=140, robot R 34 The signal from target 7 is detected and the robot interacts with it within the communication range. R6, R 11 , R 44 By interacting with R 34 The sub-coalition sub7 is formed and the TSC strategy is used to jointly search for target 7. When T=167, robot R 34 Target 7 was successfully searched and sub-alliance sub7 was disbanded.

[0162] When T=229, robot R 11The signal from target 4 is detected, and R6, R 44 Received Class I Robot R 11 The information sent by the robot R forms an alliance with sub4 to search for target 4, as shown in Figure 9(f). 44 When the target response signal reaches the threshold, it is considered that the search for target 4 is successful, as shown in Figure 9(g). At this time, the search for all targets is completed, marking the completion of this search task.

[0163] Step 2.4: Simulation experiment process of the multi-target search strategy (MSTC strategy) of swarm robots based on triangular cones

[0164] Figure 9(h) shows the entire path of the robot that has found the target position during the entire search process. The static and dynamic obstacles in the figure are the final states when the task is completed. Although part of the robot trajectory shown in the figure overlaps with the obstacles, the robot has never encountered any obstacles during the actual operation of the simulation experiment. For details, see the robot's collision avoidance process in Figure 9(c) and (d). In Figure 9(h), the robot develops the global environment based on RDC in the roaming state. Through RDC, it can first predict whether the current direction needs to be changed before deciding the next moving direction. When the target is detected, the feasibility of the current direction is first determined based on TSC. If the current direction falls within the TSC range, there is no need to change the direction, allowing the robot to search for the target more quickly. For example, robot R that has searched for targets 9 and 8 23 , R 49 , based on the TSC judgment criteria, the target is quickly searched without too many turns. The other robots also choose not to turn in order to quickly approach the target when judging that the current direction falls within the feasible area. In the process of movement, in addition to the threat of obstacles, there is also the threat of collision from other robots, and avoiding threats has the highest priority. That is, when encountering a threat, ETC is used to determine whether it is necessary to avoid collision and then decide the next moving direction to ensure the safe movement of the robot. Therefore, the other robots in Figure 9(h) also have an inevitable turning process.

[0165] Step 2.5: Simulation comparison experiment

[0166] This embodiment sets up six groups of comparative experiments to analyze two performance indicators of multi-target search in a dynamic non-convex obstacle environment, as shown in Table 3. Among them, SVF, LSDN, and KCPSO are respectively a simplified virtual force obstacle avoidance method, a straight line search method different from the nearest neighbor, and a particle swarm optimization algorithm with kinematic constraints, while ETC, RDC, and TSC are respectively an obstacle avoidance method, a roaming search strategy, and a collaborative search strategy in the invention, wherein Mode 6 is a multi-target search strategy (MSTC strategy) formed by combining the three methods in this embodiment.

[0167] Table 3 Multi-target search mode of group robots in two-dimensional unknown dynamic environment

[0168] model Obstacle Avoidance Strategy Roaming search strategy Collaborative search strategy Mode1 SVF LSDN KCPSO Mode2 SVF RDC KCPSO Mode3 SVF RDC TSC Mode4 ETC LSDN KCPSO Mode5 ETC RDC KCPSO Mode6 ETC RDC TSC

[0169] According to the number of robots 20, 40, 60, 80, and 100, the simulation experiments were carried out independently 30 times in the above six groups of modes. The robot search time and number of corner turns after each run were recorded, and the average value as well as the maximum and minimum values ​​were calculated. The obtained data are shown in Table 4.

[0170] Table 4 Performance comparison of swarm robot systems of different scales in six modes

[0171]

[0172]

[0173] Set the population size N R As the horizontal axis, T r 、N a As shown in Figure 10, as the population size N R Increase, search time T r It is gradually decreasing, and the number of corner turns is related to the number of iterations in addition to the population size.

[0174] Step 2.6: Simulation results analysis

[0175] As shown in Figure 10, the number of corner turns increases with the increase of time consumption. Therefore, in the subsequent data comparison, the number of corner turns is divided by the time consumption.

[0176] Step 2.6.1. ETC comparison results analysis

[0177] Three groups of comparative experiments were established, namely Mode 1 and Mode 4, Mode 2 and Mode 5, Mode 3 and Mode 6. These three groups of comparative experiments were conducted to compare ETC and SVF under the same roaming search strategy and collaborative search strategy. For Mode 1 and Mode 4, when the roaming search strategy is MSTC and the collaborative search strategy is KCPSO, ETC reduces the number of corner turns by 0.04% to 5.64% and the time consumption by 0.73% to 6.13% compared with SVF; compared with Mode 2 and Mode 5, the number of corner turns and time consumption are reduced by 0.19% to 10.09% and 0.81% to 6.13% respectively; compared with Mode 3 and Mode 6, the number of corner turns and time consumption are reduced by 1.68% to 8.68% and 3.50% to 13.85% respectively.

[0178] When a group of robots is searching for a task, they will abandon the originally planned optimal route in order to avoid threats. ETC can not only safely avoid threats, but also predict the necessity of avoidance, reducing unnecessary collision avoidance processes and reducing time consumption and the number of turns to a certain extent.

[0179] Step 2.6.2, RDC comparison result analysis

[0180] In order to compare the characteristics of RDC in search performance, two groups of comparative experiments were established, namely Mode 1 and Mode 2, Mode 4 and Mode 5. These two groups of comparative experiments were conducted when the collision avoidance strategy and the collaborative search strategy were the same but the roaming strategy was different. From Mode 1 and 2 in Table 4, it can be seen that when the obstacle avoidance strategy and the collaborative search strategy were both SVF and KCPSO, and the roaming strategy was changed from LSDN to RDC, the search efficiency was improved by 1.02%-13.32%; at this time, the number of corner turns was reduced by 24.96%-52.45%. For Mode 4 and 5, when the obstacle avoidance strategy and the collaborative search strategy were both ETC and KCPSO, and the roaming strategy was changed from LSDN to RDC, the search efficiency was improved by 1.4%-13.46%; at this time, the number of corner turns was reduced by 21.13%-54.52%.

[0181] The roaming search strategy based on RDC can evenly explore the unexplored environment without missing the target, which improves the search capability of the global environment to a certain extent, thereby improving the efficiency of completing the entire task search. Since RDC can first determine whether the current direction falls into the feasible area before deciding the next move, it can greatly reduce the number of corner turns, thereby reducing the time consumption, energy consumption and mechanical wear caused by the corner turning process.

[0182] Step 2.6.3. TSC comparison result analysis

[0183] In order to compare the characteristics of TSC in search performance, two groups of comparative experimental analyses were conducted between TSC and KCPSO. For Mode 2 and 3, when the obstacle avoidance strategy and roaming strategy are both SVF and RDC, and the collaborative search strategy is changed from KCPSO to TSC, the efficiency is improved by 2.6%-11.18%; at this time, the number of corner turns is reduced by 8.89%-18.60%. For Mode 5 and 6, when the obstacle avoidance strategy and roaming strategy are both ETC and RDC, and the collaborative search strategy is changed from KCPSO to TSC, the efficiency is improved by 10.61%-22.9%, and the number of corner turns is reduced by 3.68%-22.91%.

[0184] The collaborative search strategy based on TSC can quickly approach the target while ensuring the diversity of the population, effectively improving the robot's local search ability. The direction strategy of TSC allows the robot to randomly select any direction within the feasible range before deciding the next move, avoiding the population from falling into the local optimum; the step length strategy of TSC effectively controls the step length of each step of the robot according to the nonlinear decreasing step length function based on the target response signal, so that the robot will neither miss the target due to too large a step length nor increase the time consumption due to too short a step length during the movement, so the robot can find the target quickly. In addition, TSC can first determine the necessity of the turning direction, so as not to miss the best direction for searching the target and reduce the number of corners.

[0185] Step 2.6.4: Analysis of MSTC comparison results

[0186] Comparing Mode6 with Mode1 as a whole, it can be seen that the search efficiency is improved by 15.93%-28.31%, and the number of corner turns is reduced by 34.51%-61.18%. Because ETC can safely and effectively avoid complex non-convex obstacles, RDC is more conducive to the development of unknown environments, and TSC's local search ability is more prominent, and ETC, RDC, and TSC can all judge the pros and cons of the current direction before deciding the next move. This can not only avoid missing the original optimal direction due to the direction of rotation, but also reduce the time loss, energy loss, and mechanical wear caused by the robot's angle rotation process to a certain extent. It can be clearly seen from Figure 10 that Mode6 is superior to Mode1 in improving search efficiency and reducing the number of corner turns. Therefore, the MSTC strategy of the present invention is more conducive to multi-target search of group robots than the other three strategies.

[0187] The above implementation cases are only preferred embodiments of the present invention and are not intended to limit the present invention in any form. Although the present invention has been disclosed as above with preferred embodiments, they are not intended to limit the present invention. Therefore, any simple modification, equivalent changes and modifications made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the scope of protection of the technical solution of the present invention.

Claims

1. A multi-target search method for swarm robots based on triangular pyramids, characterized in that: The multi-target search method initializes the task environment before performing task search, wherein the initialization parameters at least include group robot system parameters, obstacle parameters, and target parameters, wherein the group robot system parameters include population size, robot position, and speed information; the robot is in a roaming search state when no target information is detected, and the robot develops a global environment based on a roaming direction cone; after detecting a target signal, the robot switches to a collaborative search state and uses a target search cone model for local search; the robot is provided with an avoidance strategy, wherein the avoidance strategy has the highest priority, and the robot detects a threat within a sensing range and uses the avoidance strategy for predictive avoidance; when the target signal detected by the robot reaches a preset threshold, the search is deemed to be successful, and when all targets are successfully searched, the search is completed; the method specifically includes the following steps: Step S1, modeling: Construct a module for swarm robots to search for multi-target tasks in a two-dimensional task environment, including a task environment model, a target incentive model, and a task allocation model. The task environment model constructs the task environment required for the swarm robot's multi-target task search; the target incentive model is used to construct a signal model sent by the robot's perception of the target; and the task allocation model is used to construct the resource allocation and sub-alliance formation process of the swarm robot when searching for the target. Step S2, multi-target search strategy of swarm robots based on triangular cone: Design an expanded triangular cone model suitable for non-convex obstacles, a roaming direction cone search method and a target search cone search method for searching; the roaming direction cone search method is used for the robot to explore the unknown global environment in the roaming state, and the target search cone search method is used for the robot to search locally in the collaborative state; then design an avoidance strategy based on the collision cone and the expanded triangular cone for predictive avoidance; the robot searches according to the search strategy and finally completes all search targets; In step S2, the process of designing an expanded triangular cone model suitable for non-convex obstacles is: processing the obstacle information collected within the detection range of the robot, setting a triangular cone that can surround all obstacles within the detection range, the robot mass point and the two intersection points of the triangular cone and the obstacle form a triangle, and obtaining parameter information by obtaining the information of the inscribed circle of the triangle, and expanding the triangular cone in combination with the size of the robot to obtain an expanded triangular cone.

2. The multi-target search method of swarm robots based on triangular pyramid according to claim 1, characterized in that: In the step S1, when the group robots perform multi-target search in a non-convex obstacle environment, the task environment is first initialized, wherein the obstacles include static and dynamic non-convex obstacles; when the robots perform task search in an unknown dynamic environment, the next moving direction is selected according to the searched information, and obstacles and other mobile robots are avoided during the exploration process; when the robots explore in an unknown task environment, target search is performed according to the target incentive model; when the group robots perform task search in a dynamic environment, information is exchanged within the communication range, and a sub-alliance is formed to collaboratively search for the same target.

3. The multi-target search method of swarm robots based on triangular pyramid according to claim 1, characterized in that: In step S2, the process of designing an expanded triangular cone model suitable for non-convex obstacles is: processing the obstacle information collected within the detection range of the robot, setting a triangular cone that can surround all obstacles within the detection range, the robot mass point and the two intersection points of the triangular cone and the obstacle form a triangle, and obtaining parameter information by obtaining the information of the inscribed circle of the triangle, and expanding the triangular cone in combination with the size of the robot to obtain an expanded triangular cone.

4. The multi-target search method of swarm robots based on triangular pyramids according to claim 3, characterized in that: The design is applicable to the expanded triangular cone model of non-convex obstacles, and includes the following steps: (1) Discretize the threats within the detection range and obtain the corresponding coordinates: The robot obtains its own position through the sensor. Suppose the position of the robot at a certain moment is calculate To each coordinate vector, get the angle set between every two vectors: Find the largest angle and get its corresponding two coordinates: Assume the robot's position at this moment is Then the incircle of the triangle formed by the three points is P O =[x o ,y o ], the radius of the triangular cone threat is r TC ; Calculate the semi-apex angle β1 of the triangular cone and the angle between the relative velocity and the center of the inscribed circle (2) According to the size of the robot, the triangular cone is expanded, and the part that needs to be expanded is β2: Among them, the size radius of the robot is R r , The current robot R i With obstacles O s1 , O s2 The Euclidean distance between them; the half vertex angle of the expanded triangular pyramid β=β1+2*β2; (3) Design an avoidance strategy based on collision cone and expanding triangular cone: the collision cone model is used for collision avoidance between robots, and the expanding triangular cone model is used for static and dynamic non-convex obstacles.

5. The multi-target search method of swarm robots based on triangular pyramids according to claim 4, characterized in that: The collision cone model is constructed as follows: the robot is regarded as a particle, and the obstacle is expanded into a threat circle with a radius r according to the threat radius of the obstacle and the size of the robot. A ray l is drawn from the robot position, and the intersection of l and the threat circle is M. The set of all rays drawn from the robot position that can intersect with the threat circle is collectively called a collision cone. The relative velocity vector between the robot and the obstacle falls outside the collision cone area, and the robot avoids collision with the obstacle.

6. The multi-target search method of swarm robots based on triangular pyramid according to claim 1, characterized in that: The roaming direction cone search method adopts a roaming search strategy based on the roaming direction cone. The roaming search strategy is to increase the step length of each step of the robot and avoid the previous position direction of the robot; the roaming direction cone takes the approach of avoiding all robots within the maximum distance range that the robot can detect the target; before deciding on the next moving direction, the robot determines the necessity of turning the direction through the roaming direction cone. If the speed direction at the previous moment falls within the feasible range of the roaming direction cone, there is no need to turn the direction.

7. The multi-target search method of swarm robots based on triangular pyramid according to claim 1, characterized in that: The target search cone search method adopts a collaborative search strategy based on the target search cone, including the following three cases: (1) When the current position of the robot is the individual optimal or the population optimal, the robot first predicts the robot position at the next moment based on the current direction, takes the robot position at the next moment as the center of the target search cone, and takes the robot radius as the radius of the target search cone; (2) When the robot's current position is neither the individual optimal nor the group optimal and the individual optimal position is different from the group optimal position, the robot uses the angle between the individual optimal and the group optimal as the next search direction range; (3) When the current position of the robot is neither the individual optimal nor the population optimal, but the individual optimal position is equal to the population optimal position, the robot uses the individual optimal position as the center and the maximum speed of the robot as the target search cone radius to expand the search range.

Images