Probability route map motion planning method and device for large-scale robot cluster

Abstract

In order to solve the problems of long time consumption and high calculation complexity of a motion planning algorithm of a large-scale robot cluster, the invention provides a probabilistic roadmap motion planning method and device of the large-scale robot cluster, which are divided into a macroscopic stage and a microscopic stage by combining a sampling algorithm and a hierarchical motion planning framework. In the macroscopic stage, a probability density function is used for representing the whole cluster, and a transport track of the probability density function is planned; in the microscopic stage, an individual robot tracks the probability density function planned in the macroscopic stage through an artificial potential field method. The method is superior to a leading-edge method in the aspects of calculation efficiency, expandability, track quality and the like, and the cluster motion track with the highest quality can be solved with the least planning time consumption. In addition, the scheme of the invention shows the risk perception characteristic, the risk preference of the cluster movement track can be changed by adjusting the collision risk of Gaussian distribution nodes in the Gaussian route map, and the adaptability of the cluster to the complex environment and the track flexibility are improved.

Description

Technical Field

[0001] The present invention belongs to the field of robotics, and in particular, relates to a probabilistic roadmap motion planning method and device for a large-scale robot cluster. Background Art

[0002] Large-scale robot swarm systems show great development potential and can perform many tasks such as environmental monitoring and exploration, search and rescue, smart agriculture, industrial logistics, military operations, etc. These diverse application scenarios have given rise to the urgent need to design efficient and robust motion planning algorithms suitable for large-scale robot swarms.

[0003] The paper "Scalable and informed asymptotically-optimal multi-robot motion planning" (published in the journal "Autonomous Robots", Vol. 44, No. 3, pp. 443-467) proposes a heuristic, asymptotically optimal sampling-based multi-robot motion planning algorithm. This algorithm does not need to explicitly represent the joint state of the multi-robot system. Instead, it constructs a probabilistic roadmap (PRM) for each individual robot to represent the joint state space, reducing the algorithm's spatial complexity. Based on the roadmap of the robot's individual space, a fast random search tree is constructed in the joint state space to solve the collision avoidance problem between robots. Although the above algorithm has a solid theoretical guarantee, it has poor scalability. As the size of the cluster increases, the system's joint state space grows exponentially. This method cannot overcome the curse of dimensionality, and the rate of convergence to the optimal solution is very slow. It is difficult to plan a high-quality solution within a given computing time.

[0004] The document "A generalized reduced gradient method for the optimal control of very-large-scale robotic systems" (published in the journal "IEEE Transactions on Robotics", Vol. 33, No. 5, pp. 1226-1232) uses a hierarchical motion planning framework. At the macro level, a parameterized probability density function is used to represent the macro state of the robot cluster, and the overall motion trajectory of the robot cluster is obtained. The planning time is independent of the cluster size. At the micro level, the robots track the probability density function planned at the macro level and avoid collisions with each other. The solution disclosed in the above document uses a complex parameterization method to represent the macro state of the cluster robot, and constructs a transport method for planning the probability density function of a non-convex optimization problem, resulting in high computational complexity and inability to achieve real-time motion planning of the robot cluster. Summary of the invention

[0005] The embodiment of the present application provides a motion planning algorithm that can be applied to large-scale robot clusters, with high solution quality and good scalability. It extends the classic PRM algorithm to the motion planning problem of large-scale robot clusters, overcoming the defect that the existing method cannot take into account both scalability and cluster trajectory quality.

[0006] In a first aspect, an embodiment of the present application provides a probabilistic roadmap motion planning method for a large-scale robot cluster, comprising: S101, at the macro stage, use the probability density function to represent the cluster as a whole and plan the transport trajectory of the probability density function; S103, at the microscopic stage, through the artificial potential field method, the robot individuals track the probability density function of the macroscopic stage planning.

[0007] In some embodiments, in step S101, a probability density function is used Modeling the macroscopic state of the robot cluster, the value of the probability density function represents the probability density of the cluster robot appearing at the corresponding position, and the position of each robot They are all random sampling based on probability density functions. . The Gaussian mixture model is used to represent the macroscopic state of the cluster. The probability density function of the Gaussian mixture model can be expressed as:

[0008] in represents the number of Gaussian components of the Gaussian mixture model, represent the weight, mean, and covariance of each Gaussian component, respectively.

[0009] In some embodiments, in step S101, at the macro stage, the initial and target distributions of the robot cluster are modeled as two probability density functions: , Establish the following optimization problem to plan the transport trajectory of the Gaussian mixture model ,

[0010]

[0011]

[0012]

[0013] Among them, st represents the constraint condition, and the objective function The transport cost of the Gaussian mixture model trajectory is quantified. Constraint (1) represents the initial state constraint of the Gaussian mixture model, constraint (2) represents the target state constraint of the Gaussian mixture model, and constraint (3) represents the collision risk constraint. Indicates risk preference, and Represent risk measurement and safety zone threshold respectively; Under the preset risk preference, The risk measurement result cannot exceed the preset safety zone threshold.

[0014] In some embodiments, in step S101, a risk-aware Gaussian roadmap is constructed at the macro level, where each node in the map represents a Gaussian distribution. First, samples are taken to obtain low collision risk. nodes, construct a node set ; Then, for each Gaussian node in the set , calculate all the The set of nodes in the neighborhood ; For each node ,if and If the collision risk between the shortest path and the obstacle is lower than the preset threshold, the edge and Add to edge set In the end, we got the constructed Gaussian roadmap ; Using a five-dimensional parameter vector represents a two-dimensional Gaussian distribution with a mean of , the covariance matrix is ​​expressed as:

[0015] in They represent random variables in The standard deviation in direction, represents the correlation coefficient; by sampling parameter vector Randomly generate Gaussian distribution nodes, and use various sampling strategies to Sampling and building a node set , Sampling strategies include random sampling and deterministic sampling methods; The Wasserstein distance between Gaussian distributions has an analytical expression:

[0016] in Represents Gaussian distribution The mean and covariance of ; using Wasserstein distance Define two Gaussian distribution nodes The transportation cost between ; and generate a set of neighboring nodes for each Gaussian node ; Conditional Value at Risk (CVaR) can calculate the random variable exist The worst-case conditional expected value,

[0017] For a Gaussian distribution A random variable , the conditional value at risk CVaR has an analytical expression:

[0018] in is the probability density function of the standard normal distribution, is the cumulative density function of the standard normal distribution; Use CVaR to quantify the collision risk of Gaussian distribution nodes and constrain the collision risk to not exceed the preset threshold ; Considering the Gaussian distribution transport trajectory of the robot cluster, at any time, the macroscopic state of the cluster is represented by a Gaussian distribution ; Through linear approximation, robots can be clustered to environmental obstacles The signed distance function The probability distribution of is modeled as a Gaussian distribution; the signed distance function is calculated using CVaR. The worst-case conditional expectation determines whether the expected value is within the safe area. If it is within the safe area, it is considered that the collision risk of the Gaussian distribution is low and can be used as the macro state of the swarm robot.

[0019] In some embodiments, in step S101, a series of transport trajectories from the Gaussian components of the starting Gaussian mixture model to the Gaussian components of the target Gaussian mixture model are planned through graph search, and the corresponding transport costs are calculated. ,Then, a linear programming problem is designed to transform the Gaussian mixture model trajectory ,planning the GMM trajectory that meets the requirements is designed.

[0020]

[0021]

[0022] Where st represents the constraint condition, Represents the weight assigned to each Gaussian transport trajectory. The weight constraints of the initial and target Gaussian mixture models need to be satisfied during the optimization process.

[0023] In some embodiments, in step S103, after the upper-level planning outputs the probability density function of the cluster at the next moment, at the micro-control level, each robot calculates the optimal control input and tracks the macro-state of the cluster; uses the artificial potential field method to centrally control the robot cluster and construct a gravitational field Realize the robot's tracking of the cluster's macroscopic state; define the cluster robot's current position as , for each set of control input , can calculate the microscopic state of the cluster robot at the next moment, and estimate the probability density function of the cluster through the kernel density estimation method ; The gravitational field guides the cluster to the target probability density function Where the difference meets the preset conditions:

[0024] in is a scalar parameter that measures the degree of dispersion of robots in the cluster; Constructing a repulsive field Prevent robots from colliding with each other or with obstacles; when the distance between robots or between robots and obstacles is less than the preset threshold, the repulsive field will produce a repulsive effect to prevent robots from getting closer or closer to each other.

[0025] in is the number of robots in the cluster, Is the obstacle to The repulsive field of a robot, It is The repulsive field between the robots; The weights of the gravitational field and the repulsive field are defined as and , constructing an artificial potential field by weighted combination of gravitational field and repulsive field ,Right now

[0026] The corresponding control input for each robot for: .

[0027] On the second aspect, the present application provides a probabilistic roadmap motion planning device for a large-scale robot cluster, which is configured as follows: at the macroscopic stage, a probability density function is used to represent the cluster as a whole, and the transport trajectory of the probability density function is planned; at the microscopic stage, through the artificial potential field method, individual robots track the probability density function planned at the macroscopic stage.

[0028] In a third aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any of the above methods when executed by a processor.

[0029] In a fourth aspect, an embodiment of the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the above methods when executing the program.

[0030] The probabilistic roadmap motion planning method and device for large-scale robot clusters in the embodiment of the present application have the following beneficial effects: By setting different cluster sizes, map environments, and navigation tasks, a large number of simulation experiments were carried out to verify the excellent performance of the SwarmPRM algorithm. The experimental results show that the SwarmPRM algorithm of this application is superior to the cutting-edge methods in terms of computational efficiency, scalability, and trajectory quality, and can solve the highest quality cluster motion trajectory with the least planning time. In addition, the SwarmPRM algorithm of this application exhibits risk awareness characteristics, and can change the risk preference of the cluster motion trajectory by adjusting the collision risk of the Gaussian distribution nodes in the Gaussian roadmap, thereby increasing the cluster's adaptability to complex environments and trajectory flexibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A schematic flow chart of a probabilistic roadmap motion planning method for a large-scale robot cluster according to an embodiment of the present application; Figure 2 This is the framework diagram of the SwarmPRM algorithm in this application. DETAILED DESCRIPTION

[0032] The present application is further described below in conjunction with the accompanying drawings and embodiments.

[0033] In the following introduction, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance. The following introduction provides multiple embodiments of the present invention, and different embodiments can be replaced or combined, so this application can also be considered to include all possible combinations of the same and / or different embodiments recorded. Therefore, if one embodiment includes features A, B, and C, and another embodiment includes features B and D, then this application should also be considered to include embodiments containing one or more of all other possible combinations of features A, B, C, and D, even though the embodiment may not be clearly recorded in the following text.

[0034] like Figure 1-2 As shown, in order to enable the swarm robot to reach the target position from the initial position through the complex environment, the present invention proposes the SwarmPRM algorithm, which combines the sampling algorithm and the hierarchical motion planning framework and is divided into two stages: macro and micro. In the macro stage (step S101), the probability density function is used to represent the entire cluster, and the transport trajectory of the probability density function is planned; in the micro stage (step S103), the robot individual tracks the probability density function planned in the macro stage through the artificial potential field method. Next, the SwarmPRM algorithm is introduced from the two aspects of macro planning and micro control: 1. Macro planning Using the Probability Density Function (PDF) Modeling the macroscopic state of a robot cluster, where the PDF value represents the probability density of the cluster robots appearing at the corresponding position, and the position of each robot All are based on random sampling of PDF. . The Gaussian mixture model (GMM) is used to represent the macroscopic state of the cluster. GMM can fit almost any probability density function and is suitable for characterizing the macroscopic state of the robot cluster. The probability density function of GMM can be expressed as: , in represents the number of Gaussian components of GMM, represent the weight, mean, and covariance of each Gaussian component, respectively.

[0035] In the macroscopic stage, the initial and target distributions of the robot swarm are modeled as two probability density functions: , Establish the following optimization problem to plan the transport trajectory of GMM ,

[0036]

[0037]

[0038]

[0039] Among them, st represents the constraint condition, and the objective function The transport cost of the GMM trajectory is quantified. Constraints (1) and (2) represent the initial and target state constraints of the GMM, respectively. Constraint (3) represents the collision risk constraint. Indicates risk preference, and Represent risk measurement and safety zone threshold respectively. Under the preset risk preference, The risk measurement result cannot exceed the preset safety zone threshold. The SwarmPRM algorithm proposed in this application constructs a risk-aware Gaussian roadmap at the macro level. Each node in the map represents a Gaussian distribution. First, we sample the low-collision risk nodes, construct a node set . Then, for each Gaussian node in the set , calculate all the The set of nodes in the neighborhood For each node ,if and If the collision risk between the shortest path and the obstacle is lower than the preset threshold, the edge and Add to edge set Finally, we get the constructed Gaussian roadmap .

[0040] Using a five-dimensional parameter vector represents a two-dimensional Gaussian distribution with a mean of , the covariance matrix is ​​expressed as

[0041] in They represent random variables in The standard deviation in direction, Represents the correlation coefficient. By sampling the parameter vector Randomly generate Gaussian distribution nodes. Various sampling strategies can be used, including random sampling and deterministic sampling methods. Take samples.

[0042] In optimal transport theory, Wasserstein distance intuitively shows the transport cost of transferring from one distribution to another, and has properties such as positive definiteness, symmetry, and triangle inequality. The Wasserstein distance between Gaussian distributions has an analytical expression

[0043] in Represents Gaussian distribution The mean and covariance of . Using Wasserstein distance Define two Gaussian distribution nodes The transportation cost between , and generate a set of neighboring nodes for each Gaussian node .

[0044] Value-at-Risk (VaR) Calculation of Random Variables In a given confidence interval The worst case scenario,

[0045] in Represents the probability of an event occurring. Compared with VaR, Conditional Value-at-Risk (CVaR) can more effectively identify rare events in fat-tailed distributions and calculate the probability of a random variable. exist The worst-case conditional expected value,

[0046] For a Gaussian distribution A random variable , CVaR has an analytical expression,

[0047] in and are the probability density function and cumulative density function of the standard normal distribution, respectively.

[0048] Use CVaR to quantify the collision risk of Gaussian distribution nodes and constrain the collision risk to not exceed the preset threshold Considering the Gaussian distribution transport trajectory of a robot cluster, at any time, the macroscopic state of the cluster is represented by a Gaussian distribution Through linear approximation, robots can be clustered to environmental obstacles Signed Distance Function (SDF) The probability distribution of is modeled as a Gaussian distribution. Then, CVaR is used to calculate the SDF in The worst-case conditional expectation determines whether the expected value is within the safe area. If it is within the safe area, it is considered that the collision risk of the Gaussian distribution is low and can be used as the macro state of the swarm robot.

[0049] After constructing the Gaussian route map, the SwarmPRM algorithm plans a series of transport trajectories from the Gaussian components of the starting GMM to the Gaussian components of the target GMM through graph search, and calculates the corresponding transport costs ,Then, a linear programming problem is designed to transform the GMM trajectory planning problem in the macro stage into an optimization problem of a series of Gaussian transport trajectory allocation weights, which greatly improves the computational efficiency and plans the optimal GMM trajectory.

[0050]

[0051]

[0052] in Represents the weight assigned to each Gaussian transport trajectory, and the weight constraints of the starting and target GMMs need to be satisfied during the optimization process.

[0053] 2. Micro-control When the upper-level planning outputs the probability density function of the cluster at the next moment, at the micro-control level, each robot calculates the optimal control input and tracks the macroscopic state of the cluster. The artificial potential field method is used to centrally control the robot cluster and construct a gravitational field. Realize the robot's tracking of the cluster's macroscopic state. Define the current position of the cluster robot as , for each set of control input , can calculate the microscopic state of the cluster robot at the next moment, and estimate the probability density function of the cluster through the kernel density estimation method The gravitational field guides the cluster to the target probability density function The position where the difference is as small as possible,

[0054] in is a scalar parameter that measures the degree of dispersion of the robots in the cluster.

[0055] In addition, constructing a repulsive field Prevent robots from colliding with each other or with obstacles. When the distance between robots or between robots and obstacles is less than the preset threshold, the repulsive field will produce a repulsive effect, preventing robots from getting closer or closer to obstacles.

[0056] in is the number of robots in the cluster, Is the obstacle to The repulsive field of a robot, It is The repulsive field between the robots.

[0057] The weights of the gravitational field and the repulsive field are defined as and , constructing an artificial potential field by weighted combination of gravitational field and repulsive field ,Right now

[0058] The corresponding control input for each robot for:

[0059] This application combines sampling methods and hierarchical strategies to propose a novel sampling-based hierarchical motion planning algorithm SwarmPRM, which is used to solve the motion planning problem of large-scale robot clusters. In the macroscopic stage, the robot cluster is represented as a GMM, and the optimal GMM transport trajectory is planned. In the microscopic stage, the individual robots efficiently track the GMM trajectory generated in the macroscopic stage based on the artificial potential field method. The innovative points of the probabilistic roadmap motion planning method for large-scale robot clusters in this application are mainly reflected in three aspects: (1) We propose to sample in the parameter space of Gaussian distribution and construct a Gaussian roadmap, where each node represents a Gaussian distribution. Based on the optimal transport theory and Wasserstein distance, we calculate the transport cost and shortest transport path between Gaussian distribution nodes.

[0060] (2) A collision risk detection method is systematically proposed, using CVaR as a risk metric to quantify the collision risk of each Gaussian distribution node when constructing a Gaussian roadmap. Specifically, the signed distance function (SDF) between the Gaussian distribution and environmental obstacles is calculated. During collision risk detection, the CVaR of the SDF distribution is constrained to be within the threshold range of the safe area.

[0061] (3) A linear programming problem was designed to transform the GMM trajectory planning problem in the macroscopic stage into an optimization problem of allocating weights to a series of Gaussian transport trajectories, which greatly improved the computational efficiency and planned the optimal GMM trajectory.

[0062] By setting different cluster sizes, map environments, and navigation tasks, a large number of simulation experiments were carried out to verify the excellent performance of the SwarmPRM algorithm. The experimental results show that the SwarmPRM algorithm of this application is superior to the cutting-edge methods in terms of computational efficiency, scalability, and trajectory quality, and can solve the highest quality cluster motion trajectory with the least planning time. In addition, the SwarmPRM algorithm of this application exhibits risk awareness characteristics, and can change the risk preference of the cluster motion trajectory by adjusting the collision risk of the Gaussian distribution nodes in the Gaussian roadmap, thereby increasing the cluster's adaptability to complex environments and trajectory flexibility.

[0063] The present application also provides a probabilistic roadmap motion planning device for a large-scale robot cluster, which is configured as follows: at the macroscopic stage, a probability density function is used to represent the cluster as a whole, and the transport trajectory of the probability density function is planned; at the microscopic stage, through the artificial potential field method, individual robots track the probability density function planned at the macroscopic stage.

[0064] In the present application, an embodiment of a probabilistic roadmap motion planning device for a large-scale robot cluster is basically similar to an embodiment of a probabilistic roadmap motion planning method for a large-scale robot cluster. For relevant matters, please refer to the introduction of the embodiment of the probabilistic roadmap motion planning method for a large-scale robot cluster.

[0065] The present application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the above methods when executing the program.

[0066] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned probabilistic roadmap motion planning method steps of a large-scale robot cluster are implemented. Among them, the computer-readable storage medium may include but is not limited to any type of disk, including a floppy disk, an optical disk, a DVD, a CD-ROM, a micro drive and a magneto-optical disk, a ROM, a RAM, an EPROM, an EEPROM, a DRAM, a VRAM, a flash memory device, a magnetic card or an optical card, a nanosystem (including a molecular memory IC), or any type of medium or device suitable for storing instructions and / or data.

[0067] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A probabilistic roadmap motion planning method for large-scale robot swarms, characterized in that: include: S101, at the macro stage, use the probability density function to represent the cluster as a whole and plan the transport trajectory of the probability density function; S103, at the microscopic stage, through the artificial potential field method, the robot individuals track the probability density function of the macroscopic stage planning.

2. The probabilistic roadmap motion planning method for large-scale robot clusters according to claim 1, characterized in that: In step S101, the probability density function is used Modeling the macroscopic state of the robot cluster, the value of the probability density function represents the probability density of the cluster robot appearing at the corresponding position, and the position of each robot They are all random sampling based on probability density functions. . The Gaussian mixture model is used to represent the macroscopic state of the cluster. The probability density function of the Gaussian mixture model can be expressed as: in represents the number of Gaussian components of the Gaussian mixture model, represent the weight, mean, and covariance of each Gaussian component, respectively.

3. The probabilistic roadmap motion planning method for large-scale robot clusters according to claim 1 or 2, characterized in that: In step S101, at the macro stage, the initial and target distributions of the robot cluster are modeled as two probability density functions: ， Establish the following optimization problem to plan the transport trajectory of the Gaussian mixture model , Among them, st represents the constraint condition, and the objective function The transport cost of the Gaussian mixture model trajectory is quantified. Constraint (1) represents the initial state constraint of the Gaussian mixture model, constraint (2) represents the target state constraint of the Gaussian mixture model, and constraint (3) represents the collision risk constraint. Indicates risk preference, and Represent risk measurement and safety zone threshold respectively; Under the preset risk preference, The risk measurement result cannot exceed the preset safety zone threshold.

4. The probabilistic roadmap motion planning method for large-scale robot clusters according to claim 1 or 2, characterized in that: In step S101, a Gaussian roadmap of risk awareness is constructed at the macro level, where each node in the map represents a Gaussian distribution. First, samples are taken to obtain low collision risk. nodes, construct a node set ; Then, for each Gaussian node in the set , calculate all the The set of nodes in the neighborhood ; For each node ,if and If the collision risk between the shortest path and the obstacle is lower than the preset threshold, the edge and Add to edge set In the end, we got the constructed Gaussian roadmap ; Using a five-dimensional parameter vector represents a two-dimensional Gaussian distribution with a mean of , the covariance matrix is ​​expressed as: in They represent random variables in The standard deviation in direction, represents the correlation coefficient; By sampling the parameter vector Randomly generate Gaussian distribution nodes, and use various sampling strategies to Sampling and building a node set , Sampling strategies include random sampling and deterministic sampling methods; The Wasserstein distance between Gaussian distributions has an analytical expression: in Represents Gaussian distribution The mean and covariance of Using Wasserstein distance Define two Gaussian distribution nodes The transportation cost between ; And generate a set of neighboring nodes for each Gaussian node ; Conditional Value at Risk (CVaR) can calculate the random variable exist The worst-case conditional expected value, For a Gaussian distribution A random variable , the conditional value at risk CVaR has an analytical expression: in is the probability density function of the standard normal distribution, is the cumulative density function of the standard normal distribution; Use CVaR to quantify the collision risk of Gaussian distribution nodes and constrain the collision risk to not exceed the preset threshold ; Considering the Gaussian distribution transport trajectory of the robot cluster, at any time, the macroscopic state of the cluster is represented by a Gaussian distribution ; Through linear approximation, robots can be clustered to environmental obstacles The signed distance function The probability distribution of is modeled as a Gaussian distribution; the signed distance function is calculated using CVaR. The worst-case conditional expectation determines whether the expected value is within the safe area. If it is within the safe area, it is considered that the collision risk of the Gaussian distribution is low and can be used as the macro state of the swarm robot.

5. The probabilistic roadmap motion planning method for large-scale robot clusters according to claim 1 or 2, characterized in that:In step S101, a series of transport trajectories from the Gaussian components of the starting Gaussian mixture model to the Gaussian components of the target Gaussian mixture model are planned through graph search, and the corresponding transport costs are calculated. ,Then, a linear programming problem is designed to transform the Gaussian mixture model trajectory ,planning the GMM trajectory that meets the requirements is designed. Where st represents the constraint condition, Represents the weight assigned to each Gaussian transport trajectory. The weight constraints of the initial and target Gaussian mixture models need to be satisfied during the optimization process.

6. The probabilistic roadmap motion planning method for large-scale robot clusters according to claim 1 or 2, characterized in that: In step S103, after the upper-level planning outputs the probability density function of the cluster at the next moment, at the micro-control level, each robot calculates the optimal control input and tracks the macro-state of the cluster; Centralized control of robot swarms using artificial potential field methods and construction of gravitational fields Realize the robot's tracking of the cluster's macroscopic state; define the cluster robot's current position as , for each set of control input , can calculate the microscopic state of the cluster robot at the next moment, and estimate the probability density function of the cluster through the kernel density estimation method ; The gravitational field guides the cluster to the target probability density function Where the difference meets the preset conditions: in is a scalar parameter that measures the degree of dispersion of robots in the cluster; Constructing a repulsive field Prevent robots from colliding with each other or with obstacles; when the distance between robots or between robots and obstacles is less than the preset threshold, the repulsive field will produce a repulsive effect to prevent robots from getting closer or closer to obstacles. in is the number of robots in the cluster, Is the obstacle to The repulsive field of a robot, It is The repulsive field between the robots; The weights of the gravitational field and the repulsive field are defined as and , constructing an artificial potential field by weighted combination of gravitational field and repulsive field ,Right now The corresponding control input for each robot for: 。 7. A probabilistic roadmap motion planning device for a large-scale robot cluster, characterized in that: The device is configured as follows: at the macroscopic stage, a probability density function is used to represent the entire cluster and a transport trajectory of the probability density function is planned; at the microscopic stage, through an artificial potential field method, individual robots track the probability density function planned at the macroscopic stage.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method described in any one of claims 1 to 6 are implemented.

Images