MCPP system capable of adaptively changing number of robots

Abstract

The invention discloses an MCPP system capable of adaptively changing the number of robots, and relates to the technical field of robots, and the technical key points are as follows: modeling an MCPP problem of a version with variable number of robots, converting the MCPP problem into a series of MCPP problems with fixed number of robots, and solving the MCPP problem with fixed number of robots through a method of combining inverted search with evolutionary solution. And the optimal number of the robots and the coverage path of each robot are quickly calculated. According to the invention, by multiplexing the obtained optimization information, the optimal number of robots and the coverage path with the least time consumption are obtained; the performance of the method is verified through a feasibility experiment and a contrast experiment. The result shows that compared with the prior art, the number of the robots can be effectively reduced, and the minimum single machine coverage time consumption and calculation time consumption are provided.

Description

Technical Field

[0001] The present invention relates to the field of robotics technology, and in particular to a MCPP system with adaptively changing number of robots. Background Art

[0002] In the field of multi-robot collaborative coverage path planning (MCPP), as mission requirements become increasingly complex, efficiently utilizing limited robot resources to complete designated coverage tasks has become a key issue. In real-world applications, mission scenarios are often dynamic. For example, in scenarios such as environmental monitoring and search and rescue, the size of the mission area, the distribution of obstacles, and the urgency of the mission can all change at any time. This requires the robot system to dynamically adjust its number based on mission requirements to achieve optimal resource allocation while ensuring efficient mission completion.

[0003] Currently, research on MCPP focuses on two versions: those with a fixed number of robots and those with a variable number of robots. Fixed-number MCPP methods assume a fixed, pre-set number of robots and calculate the shortest path for each robot by minimizing the maximum robot time. These methods are typically modeled as minimum-maximum balanced connected partitioning problems, MTSP problems, and MVRP problems. Variable-number MCPP methods, on the other hand, assume a variable number of robots and simultaneously optimize both the number of robots and the maximum robot coverage time, ultimately generating a Pareto optimal solution.

[0004] Existing techniques for MCPP problems with adaptively changing robot numbers have several limitations. First, fixed-number MCPP methods cannot dynamically adjust the number of robots based on task requirements. This results in either wasted robot resources or failure to meet task requirements when task requirements change. Second, while flexible-number MCPP methods consider optimizing the number of robots, their practical application is relatively limited, and the optimization process often overlooks the useful knowledge gained during each optimization pass, which can improve subsequent optimization passes. Summary of the Invention

[0005] The purpose of the present invention is to solve the above problems and provide an MCPP system with adaptive number of robots, aiming to solve the problems in the prior art such as the inability to adaptively adjust the number of robots and the underutilization of historical knowledge in the optimization process.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: a MCPP system with adaptively changing number of robots, comprising the following modules:

[0007] The problem modeling module receives the two-dimensional grid map E of the target environment sent by the user, decomposes E to obtain a number of units and the cost matrix CM corresponding to the unit vertices, and sends CM to the evolutionary operation module;

[0008] The reverse search module receives the target area information, number of robots N, maximum working time T, turning radius R, and forward speed V sent by the user, and accepts the individual set I sent by the evolutionary operation module. It sends the current number of robots n (1≤n≤N) and individual set I to the evolutionary operation module, and sends the optimal number of robots M and the corresponding Dubins coverage path Path to the user.

[0009] The evolutionary operation module receives the cost matrix CM sent by the problem modeling module and the individual set I and the number of robots n sent by the reverse search module, and sends the generated new individual set I to the reverse search module. It contains 6 sub-modules: population initialization, fitness evaluation, elite selection, crossover operator, 2-opt operator and mutation operator, which are used to determine the optimal coverage path and the corresponding optimal time cost when the number of robots is n.

[0010] The population initialization submodule in the evolutionary operation module is used to generate the initialized population I'. This submodule uses a multi-chromosome encoding method to encode the robot's driving path. Each individual is composed of multiple chromosomes, each representing a robot's driving path. The strategy is used to initialize the first generation of the population, and in subsequent generations, the population is initialized by combining outstanding individuals from the previous generation.

[0011] The fitness evaluation submodule in the evolutionary operation module receives the initialized population I' sent by the population initialization submodule and calculates the time consumed by the n chromosomes of all individuals in I'; for a single individual, the maximum time consumed by the n chromosomes in the individual is selected as the fitness function of the individual;

[0012] The elite selection submodule in the evolutionary operation module receives I' and its corresponding fitness C sent by the fitness evaluation submodule, uses the gene selection operator to select the best individual set I' in I', and sends I'' to three submodules: the crossover operator, the mutation operator, and the 2-opt operator.

[0013] After receiving I", the three submodules of the evolutionary operation module, namely the crossover operator, mutation operator and 2-opt operator, complete the crossover, mutation and optimization operations through the strategy to generate new individuals. These individuals constitute the set I, which is sent to the reverse search module.

[0014] Furthermore, the problem modeling module adopts the Semi-BCD region decomposition method to decompose the target area into a number of vertical strip rectangular units, each of which is composed of the coordinates of the upper and lower endpoints and the connecting lines between the endpoints; the cost between each endpoint pair is calculated, each cost represents the time cost of the Dubins path between the robot leaving the first endpoint to the second endpoint in the endpoint pair, and these costs constitute the cost matrix CM.

[0015] Furthermore, the population initialization submodule uses nearest neighbor insertion, minimum cost insertion and random insertion to generate the first generation population, and uses nearest neighbor insertion, minimum cost insertion, random insertion and individual reuse to generate the second generation and subsequent populations.

[0016] Furthermore, the individual reuse method of the population initialization submodule recombines multiple chromosomes in the previous generation of individuals into a single chromosome and generates new individuals by equally dividing the cost, so as to retain some useful fragments in the previous generation of population.

[0017] Furthermore, the crossover operator submodule in the evolutionary operation module uses a greedy crossover operator to recombine two chromosomes in a single individual to form two new sub-chromosomes; and uses a sequence-based crossover operator to recombine two individuals to form a new individual.

[0018] Furthermore, the mutation operator submodule in the evolutionary operation module uses a flip mutation operator to generate new individuals.

[0019] Furthermore, the 2-opt operator submodule in the evolutionary operation module uses the 2-opt operator to optimize the optimal individual to generate a new individual.

[0020] Furthermore, the reverse search module uses a reverse search method to determine the optimal number of robots M and the corresponding Dubins cover path in [1, N]. The reverse search module first sets n = N and sends n to the evolutionary operation module. If no feasible solution is found in the evolutionary operation feedback I when the number of robots is n, it means that at least n + 1 robots can complete the specified task, and N is set to n + 1. Otherwise, n is set to n - 1 and n is sent to the evolutionary operation module again.

[0021] Compared with the existing technology, this solution has the following beneficial effects:

[0022] Adaptive Adjustment of the Number of Robots: This invention utilizes a problem modeling module to transform an undefined MCPP problem into a series of fixed MCPPs. A reverse search combined with evolutionary operations automatically determines the optimal number of robots and the corresponding Dubins coverage path. This design enables the system to dynamically adjust the number of robots based on task requirements, avoiding the resource waste and task incompletion issues associated with fixed-number MCPP approaches when task requirements change. In this way, the system can automatically optimize the number of robots in different task scenarios, ensuring efficient task completion.

[0023] Knowledge reuse in the optimization process: When generating the second and subsequent initialized populations, the population initialization module of the present invention recombines multiple chromosomes in the previous generation of individuals into a single chromosome, and generates a new chromosome set in an equal-cost manner, thereby retaining some useful fragments in the previous generation of population. The crossover operator in the genetic operation module utilizes a greedy-based crossover operator to recombine the two parent chromosomes, and utilizes a sequence-based crossover operator to recombine the two parent individuals, so that the recombined individuals effectively retain some useful fragments of the parent generation. These mechanisms enable the system to fully utilize historical optimization knowledge during the optimization process, thereby improving the speed and quality of solution. Compared with the prior art, the present invention can more effectively utilize historical information, accelerate the optimization process, and improve the efficiency and accuracy of path planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 1 is a diagram of the MCPP system architecture with adaptively changing number of robots in Example 1 of the present invention;

[0025] Figure 2 This is an example diagram of the regional modeling and decomposition process in Example 1 of the present invention;

[0026] Figure 3 This is an example diagram of multi-chromosome encoding representation in Example 1 of the present invention;

[0027] Figure 4 This is an example diagram of a new individual generated by knowledge reuse in Example 1 of the present invention; DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the present invention, the technical solution of the present invention will be further described in detail below in conjunction with the embodiments of the present invention and the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0029] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other. The present invention will be described in detail below with reference to the embodiments.

[0030] Example 1: A MCPP system with adaptive change of number of robots, as shown in the attached Figure 1 As shown, it includes the following modules:

[0031] A problem modeling module is used to convert the MCPP problem with a variable number of robots into a series of MCPP problems with a fixed number of robots;

[0032] The Semi-BCD region decomposition method is used to decompose the target area into several vertical strip rectangular units. Each rectangular unit is composed of the coordinates of the upper and lower endpoints and the lines between the endpoints. The costs between each endpoint pair are calculated. Each cost represents the time cost of the Dubins path between the first endpoint and the second endpoint of the endpoint pair. These costs constitute the cost matrix CM.

[0033] The present invention assumes that the target area A is a two-dimensional area with known boundaries. There are a limited number of obstacles distributed inside the area, but the boundaries and positions are known. We use the ox-plow decomposition to decompose the target area and obtain a number of K strip rectangular units with vertical directions, as shown in the following figure. Figure 2 As shown in the middle left figure. The subsequent processing of decomposing the target area into longitudinal strip rectangular units or transverse strip rectangular units based on bull tooth decomposition is equivalent. The present invention adopts the longitudinal decomposition method. Each rectangular unit is composed of three elements, namely the upper and lower endpoint coordinates and the line between the endpoints, as shown in the attached figure. Figure 2 As shown in the middle right figure. Let the set of K strip rectangular units obtained by region decomposition be S = {s1,s2,…,s K}, K is a positive integer. Each unit in S corresponds to two upper and lower endpoints, so the K units in S correspond to 2K endpoints, and these 2K endpoints constitute the endpoint set P = {p1, p2, ..., p 2K}, where endpoint p 2k+1 and p 2k+2 Corresponding strip rectangular unit s k The upper and lower endpoints are 1≤k≤K.

[0034] The present invention uses N homogeneous Dubins robots to perform the covering task. These N robots constitute a robot set RN = {R1, R2, ..., R N Each curvature-constrained robot R in RN n (1≤n≤N) are equipped with a task sensor that can cover a square area with a side length of r1. Each robot has a maximum working time, which is set as T.n The forward speed is set to V and the minimum turning radius is set to R.

[0035] The present invention first transforms the MCPP problem with an adaptively varying number of robots into a series of MCPP problems with a fixed number of robots. Specifically, it is necessary to determine the minimum number of curvature-constrained robots and plan the shortest Dubins covering path for each robot to ensure that every point on all cells is covered within a given time frame.

[0036] The coverage task requires the robot set RN = {R1, R2, ..., R N} select a minimum number (let the number be M, M ≤ N) of Dubins robots, and plan the shortest Dubins coverage path for each Dubins robot. The robots execute the corresponding coverage paths in parallel, and can ensure that the task sensors they carry can cover every point on every unit in the strip rectangular unit set S within a given time T.

[0037] The present invention abstracts the strip rectangular unit set S into a graph G = {V, E}, where V is a set of K vertices, each vertex in V corresponds to a unit in S, and A represents the arc set between vertices. Let CM = (c ij ) is the cost matrix associated with A, c ij Represents the robot leaving the i-th unit s i After that, it reaches and covers the jth unit s j The time cost, so let in Leave s for robots i Arrived after j The time cost, To reach s j After covering j The time cost of c is due to the Dubins path characteristic. ij Not necessarily equal to c ji ,Right now Therefore, the cost matrix CM is asymmetric. Obviously, the range of the number of robots in this problem is [1, N], which can be converted into a series of optimization problems, namely

[0038]

[0039] The M in the constraint is the number of robots actually called. M robots correspond to M paths, and these M paths must satisfy the following constraints:

[0040]

[0041] Among them, the constraint introduces a binary variable xijm Formulas (1)-(2) represent the starting point constraints, that is, each robot must start from the starting point and finally return to the starting point; Formulas (3)-(4) have only one robot visit and leave each coverage point; Formula (5) is the MTZ sub-loop deletion constraint, and Formula (6) ensures that the path time cost of each robot is below the specified boundary L.

[0042] Due to the different constraints, the search spaces of these N sub-problems are completely disjoint and different. The present invention finds the minimum number of robots M, and L1, L2..., L M Satisfy all constraints and minimize the time. This problem is a multi-objective optimization problem with a dimension of N+1.

[0043] The evolutionary operation module uses a multi-chromosome encoding strategy to encode the robot's travel path. Each individual is composed of multiple chromosomes, each representing the travel path of a robot. The total number of chromosomes is equal to the number of robots. If the initial estimate is that three robots are needed, then the individual will be composed of three chromosomes.

[0044] Simplified encoding: Based on the Semi-BCD region decomposition method, the target area is divided into several rectangular sub-units, each of which consists of two upper and lower endpoints. The robot's orientation when covering a unit is determined by setting the order of the endpoints. This simplifies the mixed encoding of orientation and access sequence to an encoding that only considers the access sequence of the unit endpoints.

[0045] The population initialization submodule is used to initialize the first generation of population using a strategy, and to initialize the population in subsequent generations by combining the outstanding individuals in the previous generation of population;

[0046] Initialization of the first generation population: Use random generation, minimum cost insertion, and nearest neighbor insertion strategies to initialize the population. Assuming the initial population contains NUM individuals, use random generation, minimum cost insertion, and nearest neighbor insertion strategies to generate NUM / 3 individuals, which constitute the first generation population.

[0047] Subsequent population initialization: For the second generation and subsequent populations, some individuals in the new population are assumed to come from the previous generation. Select the best individuals from the previous generation, recombine multiple chromosomes from the previous generation into a single chromosome, and generate a new chromosome set by cost averaging. Figure 3 An example is shown in which we attach Figure 3 The individuals with 3 chromosomes are combined and decomposed to generate attached Figure 4 corresponds to an individual with two chromosomes. Figure 4 Retained the attached Figure 3At the same time, in order to ensure the diversity of the population, the remaining individuals in each round are generated by random generation, minimum cost insertion method and nearest neighbor insertion method.

[0048] The fitness evaluation submodule counts the chromosome with the longest travel time for each individual and uses this travel time as the individual's fitness value. For example, if an individual consists of three chromosomes, representing the paths of three robots, with travel times T1, T2, and T3, respectively, then the individual's fitness is max(T1, T2, T3). Lower fitness values indicate better individual performance, meaning the robots spend less time reaching their maximum coverage.

[0049] The elite selection submodule in the evolutionary operation module selects several individuals from the current population with a certain probability based on the fitness of the individuals.

[0050] The crossover operator, mutation operator and 2-opt operator in the evolutionary operation module complete the crossover, mutation and optimization operations through strategies to generate new individuals.

[0051] The crossover operator submodule uses a greedy crossover operator to recombine two chromosomes in a single individual to form two new sub-chromosomes; it uses a sequence-based crossover operator to recombine two individuals to form a new individual.

[0052] The mutation operator submodule in the evolutionary operation module uses the flip mutation operator to generate new individuals.

[0053] The 2-opt operator submodule in the evolutionary operation module uses the 2-opt operator to optimize the optimal individuals and generate new individuals.

[0054] Algorithm termination and result output

[0055] Termination condition judgment: The termination condition for a single evolution operation is set to reach the maximum number of iterations (the default maximum number of iterations is 400). The termination condition for the reverse search is to find a positive integer M (1≤M≤N) that satisfies the following conditions: when the number of robots is M, the genetic operation can obtain a feasible solution, and when the number of robots is M-1, the genetic operation cannot obtain a feasible solution.

[0056] Output: Output the optimal solution, including the optimal number of robots M and the corresponding Dubins coverage path with the shortest time. Output the optimal number of robots as 4, along with the paths and times for each of the four robots, demonstrating the algorithm's optimization results.

[0057] Example 2: Verifying the effectiveness of the system described in Example 1

[0058] 1. Performance comparison experiment

[0059] 1.1 Experimental scenario and parameter settings

[0060] Scenario setting: Experiments are conducted in four different scenarios to comprehensively evaluate the performance of the present invention in various environments.

[0061] Parameter setting: The present invention is compared with the DCRC algorithm (i.e., Karapetyan, Nare, et al. "Multi-robot dubins coverage with autonomous surface vehicles." 2018 IEEE International Conference on Robotics and Automation (ICRA)), and the CMD algorithm (i.e., Li L, Shi D, Jin S, et al. Exact and heuristic multi-robot dubins coverage path planning for known environments [J]. Sensors, 2023). The three comparison methods all use the same robot parameters, as shown in columns 1 to 5 of Table 1. The parameters set for the evolutionary operation of the present invention are shown in columns 6 to 8 of Table 1.

[0062] Table 1 Experimental parameter settings

[0063]

[0064] 2.2 Experimental process

[0065] Test the number of robots: Test the solutions generated by the DCRC algorithm and the CMD algorithm when the number of robots is [R, R-1, R-2, ...] in sequence until a suitable number of robots M is found, such that the paths planned by the DCRC or CMD algorithms when the number of robots is M meet all constraints, while the paths planned by the algorithms when the number of robots is M-1 do not meet all constraints.

[0066] Performance Metrics Evaluation: The performance of the compared algorithms is estimated based on three performance metrics: the number of robots, the maximum robot coverage time, and the computation time. The fewer the number of robots, the better the performance of the representative method. Given the same number of robots, the shorter the maximum robot coverage time and computation time, the better the performance of the representative method. By comparing these three metrics, the performance of different algorithms is evaluated under a fixed speed condition.

[0067] 2.3 Experimental Results

[0068] Comparison of Robot Quantity: Table 2 shows that in scenario 1, the proposed method calculates 3 robots, while the other two algorithms calculate 4 robots. In scenarios 2-4, the three compared methods calculate the same number of robots. This comparison demonstrates that the proposed method outperforms the other two algorithms in terms of robot quantity.

[0069] Comparison of Maximum Single-Machine Coverage Time: Table 3 shows that in scenarios 2-4, the present invention generates the same number of robots as the other algorithms, but the maximum single-machine coverage time is shorter than the other algorithms, indicating superior performance. In scenario 1, the present invention generates the fewest robots, resulting in a higher single-machine coverage time. However, as described in the performance evaluation criteria, a smaller number of robots indicates better performance, indicating that the present invention performs better.

[0070] Comparison of computational time: As can be seen from Table 4, in scenarios 1-4, the computational time of the present invention is less than that of other algorithms, indicating that the present invention has better performance.

[0071] Table 2 Comparison results of the number of robots

[0072] Scenario 1 Scenario 2 Scenario 3 Scene 4 The present invention 3 3 4 5 DCRC algorithm 4 3 4 5 CMD algorithm 4 3 4 5

[0073] Table 3 Comparison of the maximum coverage time of a single machine

[0074] Scenario 1 Scenario 2 Scenario 3 Scene 4 The present invention 348.42 301.55 143.82 146.25 DCRC algorithm 261.4 306.6 144.05 154.52 CMD algorithm 276.90 306.02 146.66 149.69

[0075] Table 4 Comparison of calculation time

[0076] Scenario 1 Scenario 2 Scenario 3 Scene 4 The present invention 6.58 6.54 0.78 0.62 DCRC algorithm 8.35 10.66 1.55 0.84 CMD algorithm 14.37 21.46 1.19 0.87

[0077] The above specific embodiments are merely explanations of the present invention and are not limitations of the present invention. After reading this specification, those skilled in the art may make non-creative modifications to the embodiments as needed. However, as long as they are within the scope of the claims of the present invention, they are protected by patent law.

Claims

1. A MCPP system with adaptively changing number of robots, characterized by: Includes the following modules: Problem modeling module: used to receive the two-dimensional grid map E of the target environment input by the user, decompose the map E, obtain the cost matrix CM corresponding to several units and unit vertices, and send the cost matrix CM to the evolutionary operation module; Reverse search module: This module receives the target area information, number of robots N, maximum robot working time T, robot turning radius R, and forward speed V input by the user. It also receives the individual set I sent by the evolutionary operation module, sends the current number of robots n and the individual set I to the evolutionary operation module, and sends the final optimal number of robots M and the corresponding Dubins coverage path Path to the user. Evolutionary operation module: used to receive the cost matrix CM sent by the problem modeling module and the individual set I and the number of robots n sent by the reverse search module, generate a new individual set I and send it to the reverse search module. The evolutionary operation module includes six sub-modules: population initialization, fitness evaluation, elite selection, crossover operator, 2-opt operator and mutation operator, which are used to determine the optimal coverage path and the corresponding optimal time cost when the number of robots is n.

2. The system according to claim 1, wherein: The problem modeling module uses the Semi-BCD region decomposition method to decompose the target area into several vertical strip rectangular units, each of which is composed of the coordinates of the upper and lower endpoints and the lines connecting the endpoints; the cost between each pair of endpoints is calculated, and each cost represents the time cost of the robot's Dubins path from the first endpoint to the second endpoint in the endpoint pair. These costs constitute the cost matrix CM.

3. The system according to claim 1, wherein: The population initialization submodule in the evolutionary operation module uses a multi-chromosome encoding method to encode the robot's driving path. Each individual is composed of multiple chromosomes, and each chromosome represents the driving path of a robot. The population initialization submodule uses the following method to generate the population: The first generation population is generated using nearest neighbor insertion, minimum cost insertion, and random insertion; The second generation and subsequent populations are generated using the nearest neighbor insertion, minimum cost insertion, random insertion and individual reuse methods.

4. The system according to claim 3, wherein: The individual reuse method of the population initialization submodule recombines multiple chromosomes in the previous generation of individuals into a single chromosome and generates new individuals by equally dividing the cost, so as to retain some useful fragments in the previous generation of population.

5. The system according to claim 1, wherein: The fitness evaluation submodule in the evolutionary operation module receives the initialized population I sent by the population initialization submodule, and calculates the time consumed by the n chromosomes of all individuals in I; for a single individual, the maximum time consumed by the n chromosomes of the individual is selected as the fitness function of the individual.

6. The system according to claim 1, wherein: The elite selection submodule in the evolutionary operation module receives the population I' and its corresponding fitness C sent by the fitness evaluation submodule, uses the gene selection operator to select the best individual set I' in the population I', and sends the individual set I'' to the crossover operator, mutation operator and 2-opt operator submodules.

7. The system according to claim 1, wherein: The crossover operator submodule in the evolutionary operation module adopts the following strategy: The greedy crossover operator is used to recombine the two chromosomes in a single individual to form two new daughter chromosomes; A sequence-based crossover operator is used to recombine two individuals to form a new individual.

8. The system according to claim 1, wherein: The mutation operator submodule in the evolutionary operation module uses a flip mutation operator to generate new individuals.

9. The system according to claim 1, wherein: The 2-opt operator submodule in the evolutionary operation module uses the 2-opt operator to optimize the optimal individual and generate a new individual.

10. The system according to claim 1, wherein: The reverse search module uses a reverse search method to determine the optimal number of robots M and the corresponding Dubins coverage path in [1, N]. The specific steps are as follows: Initialize the number of robots n = N and send n to the evolutionary operation module; If the number of robots is n, and no feasible solution can be obtained from the individual set I fed back by the evolutionary operation module, it means that at least n+1 robots are needed to complete the specified task, and let N=n+1; Otherwise, let n = n-1, continue to send n to the evolutionary operation module, and repeat the above process until the optimal number of robots M and the corresponding Dubins coverage path Path are determined.

Images