Multi-robot system collaborative comprehensive scheduling method in complex and narrow environment

Task allocation is performed through particle coding rules and adaptive particle swarm genetic algorithm, and combined with the improved hybrid A* search algorithm, the task allocation and path planning problems of multi-robot systems in small spaces are solved, and the collaborative synthesis of multi-robot systems is realized Scheduling and path planning improve the efficiency and accuracy of the system.

CN119005556APending Publication Date: 2024-11-22BEIJING INST OF TECH
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410892029.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-04
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing multi-robot systems have fewer applications in small spaces, making it difficult to effectively perform task allocation and path planning in complex and narrow environments, especially when considering robot size and kinematic constraints, which lead to difficult to resolve congestion and conflict problems.

Method used

Particle coding rules and adaptive particle swarm genetic algorithm are used for task allocation, combined with the improved hybrid A* search algorithm for path planning, considering robot motion mode switching and conflict detection to ensure conflict-free path generation.

Benefits of technology

It realizes collaborative comprehensive scheduling of multi-robot systems in complex and narrow environments, improves the accuracy and efficiency of task allocation, and is suitable for factories with high equipment density or intensive roll-on dock environments with vehicles, reducing congestion and conflicts between robots. .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119005556A_ABST
    Figure CN119005556A_ABST
Patent Text Reader

Abstract

The invention discloses a multi-robot system collaborative comprehensive scheduling method in a complex and narrow environment, and belongs to the technical field of multi-robot system scheduling planning. The implementation method comprises the following steps: designing a particle coding rule; determining a fitness value calculation method of each particle on the particle swarm; determining speed and position updating criteria of particles in an iteration process; determining a crossover and variation criterion; initializing the particle swarm and starting iteration to obtain tasks allocated to each robot and an execution sequence, namely obtaining final task allocation; determining a single robot path planning algorithm; and generating a conflict-free path of the plurality of robots, and realizing collaborative comprehensive scheduling of the multi-robot system according to the conflict-free path. According to the method, the conflict-free paths of the multiple robots are generated and used for the four-wheel independent driving and independent steering robots, and task allocation and conflict-free path planning can be performed according to the existing map, the position and orientation information of the robots and the position and orientation information of the tasks needing to be executed by the robots.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a multi-robot system coordinated comprehensive scheduling method in a complex and narrow environment, belonging to the technical field of multi-robot system scheduling planning. Background Art

[0002] Multi-robot systems are being widely used in smart warehouses, automated ports, unmanned factories, smart agriculture and other fields due to their advantages such as high efficiency, low cost, high safety and standardized operations, thus significantly improving the overall intelligence level.

[0003] At present, the research on multi-robot systems mainly focuses on coordinated scheduling. Through reasonable task allocation and path planning, the system can complete tasks quickly, efficiently and with low energy consumption. At present, although the existing multi-robot systems perform well in open environments, they are rarely used in small spaces. In terms of task allocation, existing solutions usually transform it into a multi-constrained optimization problem. The main difficulty is still how to establish a more practical model and encoding and decoding mechanism according to the actual situation, improve the accuracy and efficiency of the solution, and reduce congestion and conflicts between robots during task execution. The research on multi-vehicle path planning mainly focuses on search and optimization algorithms, but most of them only consider the movement of the intelligent body in the basic direction, ignoring the size and kinematic constraints of the intelligent body. Although some studies consider the kinematic constraints of the vehicle, they usually abstract the scene into a topological graph, or there are a small number of obstacles in the unstructured environment, so it is difficult to apply it to factories with high equipment density or roll-on / roll-off terminal environments with a large number of vehicles. Summary of the invention

[0004] The purpose of the present invention is to provide a method for coordinated comprehensive scheduling of multi-robot systems in complex and narrow environments, which is applicable to four-wheel independent drive and independent steering robots. The method can perform task allocation and conflict-free path planning based on existing maps and the position and orientation information of the robots, and the position and orientation information of the tasks that the robots need to perform, and realize coordinated comprehensive scheduling of multi-robot systems based on conflict-free paths.

[0005] The purpose of the present invention is achieved through the following technical solutions.

[0006] The present invention discloses a method for coordinated comprehensive scheduling of a multi-robot system in a complex and narrow environment, comprising the following steps:

[0007] Step 1: Design particle encoding rules.

[0008] The encoding method of particles is integer encoding. Each position of the particle represents a task allocation scheme. Each position contains m dimensions, and each dimension represents a task. The dimensions contained in the particle position are evenly divided according to the number of robots performing tasks. According to the total number of tasks m, the number of robots and the dimension of the task, the robot number performing each task and the order information of each robot performing the task are obtained;

[0009] Each position contains m dimensions, that is, there are m tasks to be performed. When the number of robots is n, the 1st to m / nth dimensions are the tasks to be performed by the first robot, the m / n+1th to 2m / nth dimensions are the tasks to be performed by the second robot, and so on. The order of task execution of the first robot is the tasks on the 1st dimension, the 2nd dimension,..., the m / nth dimension.

[0010] Step 2: Construct a method to calculate the fitness value of each particle in the particle swarm.

[0011] The minimum driving distance cost of a single robot, the total driving distance cost of all robots, and the total motion mode switching cost of all robots are used as evaluation indicators of the task allocation schemes represented by different positions of particles. The fitness value of each particle position in the particle swarm is calculated according to formula (1);

[0012] f allcost =α1f mindisence +α2f alldisance +α3f switch (1)

[0013] Where: f mindisence is the minimum travel distance cost of a single robot, f alldisance is the total travel distance cost of all robots, f switch is the total motion mode switching cost of the robot, α i (i=1, 2, 3) are the weights of the three costs;

[0014] The calculation process of the minimum travel distance cost of a single robot is as follows: first, according to the execution order of all robot tasks determined on a single particle, the distance required for each robot to complete all tasks from the starting point is obtained, and the minimum distance among all robots is selected as the minimum travel distance cost of a single robot;

[0015] The calculation process of the total driving distance of all robots is as follows: first, according to the execution order of all robot tasks determined on a single particle, the distance required for each robot to complete all tasks from the starting point is obtained, and the distance required by all robots is added up as the total driving distance cost of all robots;

[0016] The method for obtaining the distance required for each robot to complete all tasks from the starting point is: based on the map, an A* search method is used to obtain the cost value of the robot from the starting point to the first task point, the first task point to the second task point, ..., the m-1th task point to the mth task point, and all the cost values ​​are accumulated as the distance required for the robot to complete all tasks from the starting point.

[0017] The calculation process of the total motion mode switching cost of all robots is as follows: according to the execution order of all robot tasks determined on a single particle, the difference in orientation angles of each robot from the starting point to the first task point, as well as between task points in the execution order, is obtained. The differences in all orientation angles of the individual robots are accumulated to estimate the motion mode switching cost of each robot. The motion mode switching costs of all robots are summed up as the total motion mode switching cost.

[0018] Step 3: Determine the particle velocity and position update criteria during the iteration process to update the particle velocity and position.

[0019] When the particle swarm is iteratively updated, the speed and position of the particle iteration are updated by formula (2), where the inertia weight and learning factor are adaptively adjusted according to the number of iterations and the degree of particle aggregation during the iterative evolution process;

[0020]

[0021] in: They represent the position and velocity of the ith particle at the t+1th update iteration, They represent the position and velocity of the ith particle at the tth update iteration, and d is the dimensional information of the particle’s position; They represent the local optimal position of the ith particle and the global optimal position in the particle swarm as of the tth iteration; r1 and r2 are uniform random numbers in [0,1]; ω is the inertia weight; c1 is the learning factor of the local position, and c2 is the learning factor of the global position;

[0022] During the particle iterative update calculation process, when the calculated particle position When there is a non-integer value in the dimension, the probability p of rounding up or down is determined according to the difference between the non-integer and its two adjacent integer values. The specific formula is (3):

[0023]

[0024] in, is the value rounded up, is the value after rounding down;

[0025] After the particle is updated, if the same value exists in different dimensions of the same particle, that is, the task is assigned repeatedly, then one of the tasks is randomly selected and deleted, and a task is randomly selected from the unassigned tasks to fill it in to ensure the integrity of the task.

[0026] Step 4: Determine the crossover and mutation criteria.

[0027] Randomly select particles from the current particle group and form N / 2 pairs of particles as the parent generation. Randomly select two fragments at any position of the two particles to exchange and generate new particles. Randomly mutate the crossed particles with a set probability. The random mutation is to change the task of a certain dimension on the randomly selected particle. The changed task should be included in all the tasks to be executed. After the particle cross-mutation, if the same value exists in different dimensions of the same particle, that is, the task is repeatedly assigned, then randomly select and delete one of the tasks, and randomly select a task from the unassigned tasks to fill it to ensure the integrity of the task.

[0028] Step 5: Initialize the particle swarm and start iteration to obtain the task allocation results.

[0029] S51: Initialize the particle swarm, generate N particles, and randomly assign the tasks to be executed to different dimensions of the same particle according to the particle encoding criteria of step 1. When the N particles have completed the random assignment of tasks, the initial positions of the N particles are obtained, that is, the N task assignment schemes are obtained;

[0030] S52: using the method of step 2, calculating the fitness values ​​of the initial positions of the N particles obtained in step S51; and comparing the fitness values ​​of the particles, setting the position of the particle with the smallest fitness value as the global optimal position in the particle group, and setting the initial position of each particle as the local optimal position of each particle;

[0031] S53: Calculate the iteration speed of each particle according to the local optimal position and the global optimal position obtained in step S52 and the particle speed update criterion in the iteration process determined in step 3, and update the position of each particle according to the calculated particle iteration speed and the particle position update criterion in the iteration process determined in step 3;

[0032] S54: According to the particle position information of the particle swarm updated in S53 and the crossover and mutation criteria determined in step 4, the particle swarm is crossovered and mutated, and after the mutation and crossover are completed, the N / 2 particles generated after the crossover and mutation and the N particles in the original particle swarm are sorted according to the fitness value, and according to the elite retention principle, new N particles are selected from the smallest to the largest fitness value to enter the iteration process of the next generation particle swarm;

[0033] At the same time, the minimum fitness value in the newly generated particle swarm is compared with the fitness value of the particle at the global optimal position in the original particle swarm, and the position of the particle with the smaller fitness value is set as the global optimal position in the particle swarm; among the selected new N particles, if the particle is a newly generated particle after cross-compilation, the current position of the particle is the local optimal position of the particle; if the particle is a particle that has not undergone cross-mutation, the fitness value of the particle position is compared with the fitness value of the particle at its local optimal position; when the fitness value of the particle position is less than the fitness value of the particle at its local optimal position, the position of the particle is set as the local optimal position of the particle; when the fitness value of the particle position is greater than the fitness value of the particle at its local optimal position, the local optimal position of the particle is kept unchanged;

[0034] S55: Based on the N particles obtained in step S54, the global optimal position in the particle swarm and the local optimal position of each particle, the particles are repeatedly iteratively updated and cross-mutated according to the particle speed and position update criteria in the iterative process determined in step 3 and the crossover and mutation criteria determined in step 4, that is, steps S53 and S54 are repeated; when the number of iterations reaches the set value, the particle update and crossover mutation process is stopped, and the particles in the global optimal position are decoded according to the encoding method of step 1 to obtain the tasks assigned to each robot and the execution order.

[0035] Step 6: Construct a single robot path planning algorithm based on the improved hybrid A* search algorithm.

[0036] The single robot path planning algorithm is an improved hybrid A* search algorithm, such as formula (4) to formula (6), which introduces turning penalty and switching penalty between different motion modes into the cost function g; introduces the orientation cost from the current node to the end point into the cost function h; in addition, when the distance between the current node and the end point is less than the set distance, the Reeds-Shepp curve generation functions for straight and lateral movement are respectively called to calculate a Reeds-Shepp curve path from the current point to the end point without considering static obstacles. If the curve path does not collide with environmental obstacles, the path is considered feasible and a path from the node to the end point is directly generated. Otherwise, the search for a drivable path continues; an adaptive heuristic weight ξ factor is introduced, ξ is adaptively adjusted according to the current position of the robot and the position of the target point, so as to balance the trade-off between optimality and solution speed in the solution process;

[0037] The principle of adaptively adjusting ξ according to the current position of the robot and the position of the target point is: when the robot moves toward the target point, different ξ values ​​are set in different distance intervals, and the principle is that as the distance of the robot to the target point decreases, ξ will also decrease;

[0038] The turning penalty is the penalty added when the robot is in the forward left turn state, the forward right turn state, the backward left turn state or the backward right turn state; the switching penalty between different motion modes is the penalty added when the robot switches between the straight, backward, lateral, rotation and turning motion states.

[0039] g(s) = g(s f )+λ turn ·dist(s f , s)+λ Switch ·dist(s f ,s) (4)

[0040] h(s,s goal )=h o (s,s goal )+h h (s,s goal ) (5)

[0041] f(s)=g(s)+ξ*h(s,s goal ) (6)

[0042] In the formula, f(s) is the cost, g(s) is the path cost of the robot from the starting point to the current position, and h(s, s goal ) is the estimated cost of the robot from the current position to the end point, g(s f ) represents the cumulative cost of the parent node, dist(s f , s) represents the distance cost from the current node to the parent node, λ turn represents the steering penalty factor, λ Switch represents the penalty factor for switching sports modes, h o (s,s goal ) is the Manhattan distance from the current position to the end point; h h (s,s goal ) represents the directional cost from the current node to the end point; ξ is the heuristic weight factor.

[0043] Step 7: Generate conflict-free paths for multiple robots and implement coordinated integrated scheduling of the multi-robot system based on the conflict-free paths.

[0044] S71: Based on the existing map, the task allocation result obtained in step 5, and the single robot path planning algorithm in step 6, generate drivable paths for all robots;

[0045] S72: Perform conflict detection on the drivable paths of all robots generated in step S71 to determine whether there are two or more robots at the same position on the map at the same time. If not, generate a multi-robot conflict-free path; if so, randomly reserve the path of a robot and generate constraints on the conflicting positions of other robots at the conflicting moment, i.e., do not allow them to appear at the position at the conflicting moment.

[0046] S73: Based on the existing map, the task allocation result obtained in step 5, the single robot path planning algorithm in step 6 and the position constraints generated in step S72, the single robot drivable path is generated again, and conflict detection is performed according to the conflict detection method in step S72. If there is a conflict, the position constraints are added again, and the process of generating single robot paths, conflict detection, and adding position constraints is repeated until conflict-free paths for multiple robots are obtained, and coordinated comprehensive scheduling of multi-robot systems is achieved based on the conflict-free paths.

[0047] Beneficial effects:

[0048] 1. The present invention discloses a method for coordinated integrated scheduling of a multi-robot system in a complex and narrow environment. The method adopts a coordinated integrated scheduling framework suitable for four-wheel independent drive and independent steering robots. It is the first time that four-wheel independent drive and independent steering robots are used in a multi-robot system. At the same time, the size and kinematic constraints of the robots are fully considered. The method can be directly applied to factories with high equipment density or roll-on / roll-off terminal environments with dense vehicles.

[0049] 2. The present invention discloses a method for coordinated integrated scheduling of multi-robot systems in a complex and narrow environment. It uses an adaptive particle swarm genetic algorithm to solve the problem of task allocation in a narrow space, and uses the cost of A* search to replace the traditional distance calculation method to obtain the distance between the robot and the task and between the tasks when solving the fitness. The algorithm parameters are adaptively adjusted according to the iterative evolution process and the degree of particle aggregation. This method solves the problem of task allocation in a narrow space more effectively than traditional particle swarm algorithms and genetic algorithms.

[0050] 3. The present invention discloses a method for coordinated integrated scheduling of a multi-robot system in a complex and narrow environment, which adopts an improved hybrid A* robot path search algorithm, considers motion mode switching, and introduces adaptive heuristic weights to accelerate the single-vehicle search process. The method uses a conflict search algorithm to detect and eliminate conflicts between robots to obtain the optimal task driving path for all robots. The present invention is not only applicable to four-wheel independent drive and independent steering robots in a small space, but can also be applied to four-wheel independent drive and independent steering robots in other scenarios, such as inspection and transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1It is the overall flow chart of the present invention "A method for coordinated comprehensive scheduling of multi-robot systems in a complex and narrow environment";

[0052] Figure 2 It is a schematic diagram of the encoding and decoding process of the present invention "a method for coordinated comprehensive scheduling of multi-robot systems in a complex and narrow environment";

[0053] Figure 3 It is a schematic diagram of the cross-variation of the present invention "a method for coordinated comprehensive scheduling of multi-robot systems in a complex and narrow environment";

[0054] Figure 4 It is a flowchart of step 5 of the present invention "a method for coordinated comprehensive scheduling of multi-robot systems in a complex and narrow environment";

[0055] Figure 5 It is a flowchart of step 7 of the present invention "a method for coordinated comprehensive scheduling of multi-robot systems in a complex and narrow environment";

[0056] Figure 6 This is a scene diagram of an example of the present invention's "A method for coordinated comprehensive scheduling of multi-robot systems in a complex and narrow environment." DETAILED DESCRIPTION

[0057] In order to better illustrate the purpose and advantages of the present invention, the following is a diagram of the present invention. Figure 6 The constructed scene further illustrates the content of the invention.

[0058] The space size of the design scene in Example 1 is 100 meters × 60 meters, in which 18 obstacles (shelves) of 35 meters × 1.5 meters are set. At the same time, the size of the robot is set to 2 meters × 3 meters, and the minimum turning radius is 3 meters. Compared with traditional car-like robots, this robot has added rotation and lateral movement modes, a smaller turning radius, and is more suitable for narrow spaces. In the above environment, 4 robots are set to perform 16 tasks. The overall process of coordinated comprehensive scheduling is as follows: Figure 1 As shown, a more detailed description is as follows:

[0059] like Figure 1 As shown, this embodiment discloses a method for coordinated comprehensive scheduling of a multi-robot system in a complex and narrow environment, and the specific implementation steps are as follows:

[0060] Step 1: Design particle encoding rules;

[0061] Integer coding is used for particle encoding. Each position of the particle represents a task allocation plan. Each position contains m dimensions, and each dimension represents a task. The dimensions contained in the particle position are evenly divided according to the number of robots performing tasks. According to the total number of tasks m, the number of robots and the dimension of the task, the robot number performing each task and the order information of each robot performing the task are obtained.

[0062] In this embodiment, each position contains 16 dimensions, that is, there are 16 tasks to be performed, and the number of robots is 4. Then, dimensions 1 to 4 are the tasks to be performed by the first robot, dimensions 5 to 8 are the tasks to be performed by the second robot, and so on. The order of task execution of the first robot is tasks in the 1st dimension, 2nd dimension, 3rd dimension, and 4th dimension. Figure 2 The process of using the coding criteria to determine the tasks assigned to the robots and the execution order of each robot is illustrated in the example.

[0063] Step 2: Construct a method for calculating the fitness value of each particle in the particle swarm;

[0064] The minimum driving distance cost of a single robot, the total driving distance cost of all robots, and the total motion mode switching cost of all robots are used as evaluation indicators of the task allocation schemes represented by different positions of particles. The fitness value of each particle position in the particle swarm is calculated according to formula (7);

[0065] f all cost =α1f min disence +α2f alldisance +α3f switch (7)

[0066] Among them, f min disence is the minimum driving distance cost of a single robot; f alldisance is the total travel distance cost of all robots; f switch is the total motion mode switching cost of the robot; α i (i=1, 2, 3) are weights of the three costs, and in this embodiment, α1=α2=α3=1.

[0067] The calculation process of the minimum driving distance cost of a single robot is as follows: first, according to the execution order of all robot tasks determined on a single particle, the distance required for each robot to complete all tasks from the starting point is obtained, and the minimum distance among all robots is selected as the minimum driving distance cost of a single robot.

[0068] The calculation process of the total driving distance of all robots is as follows: first, according to the execution order of all robot tasks determined on a single particle, the distance required for each robot to complete all tasks from the starting point is obtained, and the distances required by all robots are added up as the total driving distance cost of all robots.

[0069] The method for obtaining the distance required for each robot to complete all tasks from the starting point is: based on the map, the A* search method is used to obtain the cost of the robot from the starting point to the first task point, the first task point to the second task point, ..., the m-1th task point to the mth task point, and all the cost values ​​are accumulated as the distance required for the robot to complete all tasks from the starting point. In this embodiment, each robot uses the A* search method based on the map to obtain the cost of the robot from the starting point to the first task point, the first task point to the second task point, the second task point to the third task point, and the third task point to the fourth task point, and the four values ​​are accumulated as the distance required for the robot to complete all tasks from the starting point.

[0070] The calculation process of the total motion mode switching cost of all robots is as follows: according to the execution order of all robot tasks determined on a single particle, the difference in the orientation angles of each robot from the starting point to the first task point, and between the task points in the execution order, is obtained, and all the differences in the orientation angles of the obtained single robots are accumulated to estimate the motion mode switching cost of each robot, and the motion mode switching costs of all robots are added up as the total motion mode switching cost. In this embodiment, the difference in the orientation angle between the starting point and the first task point of the robot, the difference in the orientation angle between the first task point and the second task point, the difference in the orientation angle between the second task point and the third task point, and the difference in the orientation angle between the third task point and the fourth task point are calculated, and the four values ​​are accumulated to estimate the motion mode switching cost of each robot.

[0071] Step 3: Determine the particle velocity and position update criteria during the iteration process to update the particle velocity and position;

[0072] When the particle swarm is iteratively updated, the speed and position of the particle iteration are updated by formula (8), where the inertia weight and learning factor are adaptively adjusted according to the number of iterations and the degree of particle aggregation during the iterative evolution process.

[0073]

[0074] in: They represent the position and velocity of the ith particle at the t+1th update iteration, They represent the position and velocity of the ith particle at the tth update iteration respectively; d is the dimensional information of the particle’s position; They represent the local optimal position of the ith particle and the global optimal position in the particle swarm as of the tth iteration; r1 and r2 are uniform random numbers in [0,1]; ω is the inertia weight; c1 is the learning factor of the local position, and c2 is the learning factor of the global position;

[0075] During the particle iterative update calculation process, when the calculated particle position When there is a non-integer value in the dimension, the probability p of rounding up or down is determined according to the difference between the non-integer and its two adjacent integer values. The specific formula is (9):

[0076]

[0077] in, is the value rounded up, The value is rounded down.

[0078] In this embodiment, it is assumed that the calculated particle position If the value of a certain dimension is 5.6, then the probability of its existence is 0.4 and the value is 5, and the probability of its existence is 0.6 and the value is 6.

[0079] After the particle is updated, if the same value exists in different dimensions of the same particle, that is, the task is assigned repeatedly, then one of the tasks is randomly selected and deleted, and a task is randomly selected from the unassigned tasks to fill it in to ensure the integrity of the task.

[0080] In the process of updating the particle speed and position, in order to avoid entering the local optimal solution too early in the iteration process and improve the search speed of the particle swarm optimal solution, the inertia parameter and learning factor are adaptively adjusted according to the iterative evolution process and the degree of particle aggregation. The process is as follows: Formula (10) to Formula (12), where the degree of particle aggregation is measured by the closeness of the fitness values ​​of the particles in the population.

[0081]

[0082] Among them, ω max and ω min are the maximum and minimum values ​​of the inertia weight respectively. In this embodiment, ω max =1,ω min = 0; T is the maximum number of iterations, which is 200 in this embodiment; N is the number of particles, which is 30 in this embodiment; F g,t is the fitness value of the global optimal position in the particle swarm in the tth iteration; F a,t is the average fitness value of the particle swarm at the tth iteration; γ is the particle aggregation coefficient, which is 0.2 in this embodiment.

[0083] Step 4: Determine the crossover and mutation criteria;

[0084] In order to avoid falling into the local optimal solution, particles are randomly selected from the current particle group and N / 2 pairs of particles are formed as the parent generation. Two fragments are randomly selected at any position of the two particles to exchange and generate new particles; and the crossover particles are randomly mutated with a set probability. The random mutation is to change the task of a certain dimension on the randomly selected particle, and the changed task should be included in all the tasks to be executed; after the particle crossover mutation, if the same value exists in different dimensions of the same particle, that is, the task is repeatedly assigned, then one of the tasks is randomly selected and deleted, and a task is randomly selected from the unassigned tasks to fill in, to ensure the integrity of the task. The crossover process is as follows: Figure 3 shown.

[0085] Step 5: Initialize the particle swarm and start iteration to obtain the task allocation results;

[0086] S51: Initialize the particle swarm and generate 30 particles. According to the particle encoding criteria in step 1, randomly assign the 16 tasks to be executed to different dimensions of the same particle. When all 30 particles complete the random assignment of tasks, the initial positions of the 30 particles are obtained, that is, 30 task assignment schemes are obtained.

[0087] S52: using the method of step 2, calculate the fitness values ​​of the initial positions of the 30 particles obtained in step S51; and compare the fitness values ​​of the particles, set the position of the particle with the smallest fitness value as the global optimal position in the particle group, and set the initial position of each particle as the local optimal position of each particle;

[0088] S53: Calculate the iteration speed of each particle according to the local optimal position and the global optimal position obtained in step S52 and the particle speed update criterion in the iteration process determined in step 3, and update the position of each particle according to the calculated particle iteration speed and the particle position update criterion in the iteration process determined in step 3;

[0089] S54: According to the particle position information of the particle swarm updated in S53 and the crossover and mutation criteria determined in step 4, the particle swarm is crossovered and mutated. After the mutation and crossover are completed, the 15 particles generated after the crossover and mutation and the 30 particles in the original particle swarm are sorted according to the fitness value. According to the elite retention principle, 30 new particles are selected from the smallest to the largest fitness value to enter the iteration process of the next generation particle swarm; at the same time, the minimum fitness value in the newly generated particle swarm is compared with the fitness value of the particle in the global optimal position in the original particle swarm, and the particle with the smallest fitness value is positioned at the top. Set it to the global optimal position in the particle swarm; among the new 30 particles selected, if the particle is a newly generated particle after cross-compilation, the current position of the particle is the local optimal position of the particle; if the particle is a particle that has not undergone cross-mutation, compare the fitness value of the particle position with the particle at its local optimal position. When the fitness value of the particle position is less than the fitness value of the particle at its local optimal position, set the position of the particle to the local optimal position of the particle; when the fitness value of the particle position is greater than the fitness value of the particle at its local optimal position, keep the local optimal position of the particle unchanged;

[0090] S55: Based on the 30 particles obtained in step S54, the global optimal position in the particle swarm and the local optimal position of each particle, the particles are repeatedly iteratively updated and cross-mutated according to the particle speed and position update criteria in the iterative process determined in step 3 and the crossover and mutation criteria determined in step 4, that is, steps S53 and S54 are repeated; when the number of iterations reaches 200, the particle update and crossover mutation process is stopped, and the particles in the global optimal position are decoded according to the encoding method of step 1 to obtain the tasks assigned to each robot and the execution order.

[0091] The specific process of this step is as follows Figure 4 shown.

[0092] Step 6: Construct a single robot path planning algorithm based on the improved hybrid A* search algorithm;

[0093] The single robot path planning algorithm is an improved hybrid A* search algorithm, such as formula (13) to formula (15), which introduces turning penalty and switching penalty between different motion modes into the cost function g; introduces the orientation cost from the current node to the end point into the cost function h; in addition, when the distance between the current node and the end point is less than the set distance, the Reeds-Shepp curve generation functions for straight and lateral movement are respectively called to calculate a Reeds-Shepp curve path from the current point to the end point without considering static obstacles. If the curve path does not collide with environmental obstacles, the path is considered feasible and a path from the node to the end point is directly generated. Otherwise, the search for a drivable path continues; an adaptive heuristic weight factor ξ is introduced, and ξ is adaptively adjusted according to the current position of the robot and the position of the target point to balance the tradeoff between optimality and solution speed during the solution process;

[0094] The principle of adaptive adjustment of ξ according to the current position of the robot and the position of the target point is: in the process of the robot moving toward the target point, different ξ values ​​are set in different distance intervals, and the principle is that as the distance of the robot to the target point decreases, ξ will also decrease. In this embodiment, when the current position of the robot and the position of the target point are greater than 30 meters, ξ is set to 2, when the current position of the robot and the position of the target point are greater than 20 meters and less than 30 meters, ξ is set to 1.5, when the current position of the robot and the position of the target point are greater than 10 meters and less than 20 meters, ξ is set to 1.2, and in other cases, ξ is set to 1;

[0095] The turning penalty is the penalty added when the robot is in the forward left turn state, the forward right turn state, the backward left turn state or the backward right turn state; the switching penalty between different motion modes is the penalty added when the robot switches between the straight, backward, lateral, rotation and turning motion states.

[0096] g(s) = g(s f )+λ turn ·dist(s f , s)+λ Switch ·dist(s f ,s) (13)

[0097] h(s,s goal )=h o (s,s goal )+h h (s,s goal ) (14)

[0098] f(s)=g(s)+ξ*h(s,s goal ) (15)

[0099] Where f(s) is the cost; g(s) is the path cost of the robot from the starting point to the current position; h(s, s goal ) is the estimated cost of the robot from the current position to the end point; g(s f ) represents the cumulative cost of the parent node; dist(s f , s) represents the distance cost from the current node to the parent node; λ turn represents the turning penalty factor, which is 2 in this embodiment; Switch represents the penalty factor for switching the sports mode, which is 3 in this embodiment; h o (s,s goal ) is the Manhattan distance from the current point to the end point; h h (s,s goal ) represents the orientation cost from the current node to the end point, which is represented by the difference between the orientation angle of the robot at the current node and the orientation angle of the end point in this embodiment; ξ is the heuristic weight factor.

[0100] Step 7: Generate conflict-free paths for multiple robots, and implement coordinated integrated scheduling of the multi-robot system based on the conflict-free paths;

[0101] S71: Based on the existing map, the task allocation result obtained in step 5, and the single robot path planning algorithm in step 6, generate drivable paths for all robots;

[0102] S72: Perform conflict detection on the drivable paths of all robots generated in step S71 to determine whether there are two or more robots at the same position on the map at the same time. If not, generate a multi-robot conflict-free path; if so, randomly reserve the path of a robot and generate constraints on the conflicting positions of other robots at the conflicting moment, i.e., do not allow them to appear at the position at the conflicting moment.

[0103] S73: Based on the existing map, the task allocation result obtained in step 5, the single robot path planning algorithm in step 6 and the position constraints generated in step S72, the single robot drivable path is generated again, and conflict detection is performed according to the conflict detection method in step S72. If there is a conflict, the position constraints are added again, and the process of generating the single robot path, conflict detection and adding the position constraints is repeated until multiple robot conflict-free paths are obtained.

[0104] The specific process of this step is as follows Figure 5 shown.

[0105] To further illustrate the effectiveness and advancement of the method proposed in this patent, the method proposed in this article is combined with random allocation, particle swarm allocation, genetic algorithm and genetic particle swarm algorithm to perform 20 task allocations in the scenario of Example 1, and the average fitness value and the optimal fitness value in the allocation process are recorded. Then, a task allocation scheme is randomly selected from them for path planning, and the planned maximum driving distance, average driving distance, total driving distance and total cost are recorded. The relevant result data are shown in Table 1. It can be seen that the method proposed in this article ensures higher allocation scheme quality and completes multi-robot conflict-free path planning, realizing the coordinated comprehensive scheduling of multi-robot systems in complex and narrow environments.

[0106] Table 1 Comparison of calculation results

[0107]

[0108] The specific description above further illustrates the purpose, technical solutions and beneficial effects of the invention in detail. It should be understood that the above is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for coordinated comprehensive scheduling of multi-robot systems in complex and narrow environments, characterized by: Applicable to four-wheel independent drive and independent steering robots, the method can perform task allocation and conflict-free path planning based on the existing map and the position and orientation information of the robot and the position and orientation information of the task that the robot needs to perform; it includes the following steps: Step 1: Design particle encoding rules; integer encoding is used for particle encoding. Each position of the particle represents a task allocation scheme. Each position contains m dimensions, and each dimension represents a task. The dimensions contained in the particle position are evenly divided according to the number of robots performing tasks. According to the total number of tasks m, the number of robots, and the dimension of the task, the robot number performing each task and the order information of each robot performing the task are obtained; Step 2: The minimum driving distance cost of a single robot, the total driving distance cost of all robots, and the total motion mode switching cost of all robots are used as evaluation indicators of the task allocation schemes represented by different positions of particles, and a method for calculating the fitness value of each particle position in the particle swarm is constructed; Step 3: Determine the particle velocity and position update criteria during the iteration process to update the particle velocity and position; Step 4: Determine the crossover and mutation criteria, randomly select particles from the current particle group and form N / 2 pairs of particles as the parent generation, randomly select two fragments at any position of the two particles to exchange and generate new particles; and randomly mutate the crossed particles with a set probability. The random mutation is to change the task of a certain dimension on the randomly selected particle, and the changed task should be included in all the tasks to be executed; after the particle crossover mutation, if the same value exists in different dimensions of the same particle, that is, the task is repeatedly assigned, then randomly select and delete one of the tasks, and randomly select a task from the unassigned tasks to fill in, to ensure the integrity of the task; Step 5: Initialize the particle swarm, generate N particles, obtain the initial positions of the N particles, and obtain N task allocation schemes; calculate the fitness values ​​of the initial positions of the N particles, determine the global optimal position of the particles and the local optimal position of each particle; calculate the iteration speed of each particle according to the global optimal position of the particles and the local optimal position of each particle, and update the position of each particle; perform crossover and mutation on the updated particles, sort the N / 2 particles generated after the crossover and mutation and the N particles in the original particle swarm according to the fitness value, and select new N particles from the smallest to the largest fitness value to enter the iteration process of the next generation particle swarm according to the principle of elite retention; when the number of iterations reaches the set value, stop the particle update and crossover mutation process, and decode the particles in the global optimal position to obtain the tasks and execution order assigned to each robot, that is, obtain the task allocation result; Step 6: Construct a single robot path planning algorithm based on the improved hybrid A* search algorithm; Step 7: Based on the existing map and the obtained task allocation results, generate the drivable paths of all robots according to the single robot path planning algorithm; perform conflict detection on the drivable paths of all robots to determine whether there are more than two robots at the same position on the map at the same time. If not, generate a multi-machine conflict-free path; if so, randomly retain the path of a robot, and generate constraints on the conflicting positions of other robots at the conflict time, that is, they are not allowed to appear at the position at the conflict time; add constraints to the robots in conflict, and only retain the path of one robot passing through the conflicting position at the conflict time. Based on the map, the task allocation results of each robot and the position constraints of some robots, generate conflict-free paths for multiple robots, and realize the coordinated comprehensive scheduling of the multi-robot system according to the conflict-free paths.

2. A method for coordinated integrated scheduling of a multi-robot system in a complex and narrow environment as claimed in claim 1, characterized in that: The specific implementation method of step 1 is as follows: The encoding method of particles is integer encoding. Each position of the particle represents a task allocation scheme. Each position contains m dimensions, and each dimension represents a task. The dimensions contained in the particle position are evenly divided according to the number of robots performing tasks. According to the total number of tasks m, the number of robots and the dimension of the task, the robot number performing each task and the order information of each robot performing the task are obtained; Each position contains m dimensions, that is, there are m tasks to be performed. When the number of robots is n, the 1st to m / nth dimensions are the tasks to be performed by the first robot, the m / n+1th to 2m / nth dimensions are the tasks to be performed by the second robot, and so on. The order of task execution of the first robot is the tasks on the 1st dimension, the 2nd dimension,..., the m / nth dimension.

3. The method for coordinated integrated scheduling of a multi-robot system in a complex and narrow environment as claimed in claim 1, characterized in that: The specific implementation method of step 2 is as follows: The minimum driving distance cost of a single robot, the total driving distance cost of all robots, and the total motion mode switching cost of all robots are used as evaluation indicators of the task allocation schemes represented by different positions of particles. The fitness value of each particle position in the particle swarm is calculated according to formula (1); f allcost =α1f mindisence +α2f alldisance +α3f switch (1) Where: f mindisence is the minimum driving distance cost of a single robot, f alldisance is the total travel distance cost of all robots, f switch is the total motion mode switching cost of the robot, α i (i=1, 2, 3) are the weights of the three costs; The calculation process of the minimum travel distance cost of a single robot is as follows: according to the execution order of all robot tasks determined on a single particle, the distance required for each robot to complete all tasks from the starting point is obtained, and the minimum distance among all robots is selected as the minimum travel distance cost of a single robot; The calculation process of the total driving distance of all robots is as follows: according to the execution order of all robot tasks determined on a single particle, the distance required for each robot to complete all tasks from the starting point is obtained, and the distance required by all robots is added up as the total driving distance cost of all robots; The method for obtaining the distance required for each robot to complete all tasks from the starting point is: based on the map, an A* search method is used to obtain the cost values ​​of the robot from the starting point to the first task point, the first task point to the second task point, ..., the m-1th task point to the mth task point, and all the cost values ​​are accumulated as the distance required for the robot to complete all tasks from the starting point; The calculation process of the total motion mode switching cost of all robots is as follows: according to the execution order of all robot tasks determined on a single particle, the difference in orientation angles of each robot from the starting point to the first task point, as well as between task points in the execution order, is obtained. The differences in all orientation angles of the individual robots are accumulated to estimate the motion mode switching cost of each robot. The motion mode switching costs of all robots are summed up as the total motion mode switching cost.

4. The method for coordinated integrated scheduling of a multi-robot system in a complex and narrow environment as claimed in claim 1, characterized in that: The specific implementation method of step 3 is as follows: When the particle swarm is iteratively updated, the speed and position of the particle iteration are updated by formula (2), where the inertia weight and learning factor are adaptively adjusted according to the number of iterations and the degree of particle aggregation during the iterative evolution process; in: They represent the position and velocity of the ith particle at the t+1th update iteration, They represent the position and velocity of the ith particle at the tth update iteration, and d is the dimensional information of the particle’s position; They represent the local optimal position of the ith particle and the global optimal position in the particle swarm as of the tth iteration; r1 and r2 are uniform random numbers in [0,1]; ω is the inertia weight; c1 is the learning factor of the local position, and c2 is the learning factor of the global position; During the particle iterative update calculation process, when the calculated particle position When there is a non-integer value in the dimension, the probability p of rounding up or down is determined according to the difference between the non-integer and its two adjacent integer values. The specific formula is (3): in, is the value rounded up, is the value after rounding down; After the particle is updated, if the same value exists in different dimensions of the same particle, that is, the task is assigned repeatedly, then one of the tasks is randomly selected and deleted, and a task is randomly selected from the unassigned tasks to fill it in to ensure the integrity of the task.

5. The method for coordinated integrated scheduling of a multi-robot system in a complex and narrow environment as claimed in claim 1, characterized in that: The specific implementation method of step 5 is as follows: S51: Initialize the particle swarm, generate N particles, and randomly assign the tasks to be executed to different dimensions of the same particle according to the particle encoding criteria of step 1. When the N particles have completed the random assignment of tasks, the initial positions of the N particles are obtained, that is, the N task assignment schemes are obtained; S52: using the method of step 2, calculating the fitness values ​​of the initial positions of the N particles obtained in step S51; and comparing the fitness values ​​of the particles, setting the position of the particle with the smallest fitness value as the global optimal position in the particle group, and setting the initial position of each particle as the local optimal position of each particle; S53: Calculate the iteration speed of each particle according to the local optimal position and the global optimal position obtained in step S52 and the particle speed update criterion in the iteration process determined in step 3, and update the position of each particle according to the calculated particle iteration speed and the particle position update criterion in the iteration process determined in step 3; S54: According to the particle position information of the particle swarm updated in S53 and the crossover and mutation criteria determined in step 4, the particle swarm is crossovered and mutated. After the mutation and crossover are completed, the N / 2 particles generated after the crossover and mutation are sorted with the N particles in the original particle swarm according to the fitness value. According to the elite retention principle, the new N particles are selected from the smallest to the largest fitness value to enter the iteration process of the next generation particle swarm; at the same time, the minimum fitness value in the newly generated particle swarm is compared with the fitness value of the particle in the global optimal position in the original particle swarm, and the particle with the smallest fitness value is positioned at the top. Set it to the global optimal position in the particle swarm; among the selected N new particles, if the particle is a newly generated particle after cross-compilation, the current position of the particle is the local optimal position of the particle; if the particle is a particle that has not undergone cross-mutation, compare the fitness value of the particle position with the particle at its local optimal position; when the fitness value of the particle position is less than the fitness value of the particle at its local optimal position, set the position of the particle to the local optimal position of the particle; when the fitness value of the particle position is greater than the fitness value of the particle at its local optimal position, keep the local optimal position of the particle unchanged; S55: Based on the N particles obtained in step S54, the global optimal position in the particle swarm and the local optimal position of each particle, the particles are repeatedly iteratively updated and cross-mutated according to the particle speed and position update criteria in the iterative process determined in step 3 and the crossover and mutation criteria determined in step 4, that is, steps S53 and S54 are repeated; when the number of iterations reaches the set value, the particle update and crossover mutation process is stopped, and the particles in the global optimal position are decoded according to the encoding method of step 1 to obtain the tasks assigned to each robot and the execution order.

6. The method for coordinated integrated scheduling of a multi-robot system in a complex and narrow environment as claimed in claim 1, characterized in that: The specific implementation method of step 6 is as follows: The single robot path planning algorithm is an improved hybrid A* search algorithm, such as formula (4) to formula (6), which introduces turning penalty and switching penalty between different motion modes into the cost function g; introduces the orientation cost from the current node to the end point into the cost function h; when the distance between the current node and the end point is less than the set distance, the Reeds-Shepp curve generation functions for straight and lateral movement are respectively called to calculate a Reeds-Shepp curve path from the current point to the end point without considering static obstacles. If the curve path does not collide with environmental obstacles, the path is considered feasible and a path from the node to the end point is directly generated. Otherwise, the search for a drivable path continues; an adaptive heuristic weight factor ξ is introduced, and ξ is adaptively adjusted according to the current position of the robot and the position of the target point to balance the tradeoff between optimality and solution speed during the solution process; The principle of adaptively adjusting ξ according to the current position of the robot and the position of the target point is: when the robot moves toward the target point, different ξ values ​​are set in different distance intervals, and the principle is that as the distance of the robot to the target point decreases, ξ will also decrease; The turning penalty is the penalty added when the robot is in the forward left turn state, the forward right turn state, the backward left turn state or the backward right turn state; the switching penalty between different motion modes is the penalty added when the robot switches between the straight, backward, lateral, self-rotation and turning motion states; g(s)=g(s f )+λ turn ·dist(s f ,s)+λ Switch ·dist(s f ,s) (4) h(s,s goal )=h o (s,s goal )+h h (s,s goal ) (5) f(s)=g(s)+ξ*h(s,s goal ) (6) In the formula, f(s) is the cost, g(s) is the path cost of the robot from the starting point to the current position, and h(s, s goal ) is the estimated cost of the robot from the current position to the end point, g(s f ) represents the cumulative cost of the parent node, dist(s f , s) represents the distance cost from the current node to the parent node, λ turn represents the steering penalty factor, λ Switch represents the penalty factor for switching sports modes, h o (s,s goal ) is the Manhattan distance from the current point to the end point; h h (s,s goal ) represents the directional cost from the current node to the end point; ξ is the heuristic weight factor.

7. The method for coordinated integrated scheduling of a multi-robot system in a complex and narrow environment as claimed in claim 1, characterized in that: The specific implementation method of step 7 is as follows: S71: Based on the existing map, the task allocation result obtained in step 5, and the single robot path planning algorithm in step 6, generate drivable paths for all robots; S72: Perform conflict detection on the drivable paths of all robots generated in step S71 to determine whether there are two or more robots at the same position on the map at the same time. If not, generate a multi-robot conflict-free path; if so, randomly reserve the path of a robot and generate constraints on the conflicting positions of other robots at the conflicting moment, i.e., do not allow them to appear at the position at the conflicting moment. S73: Based on the existing map, the task allocation result obtained in step 5, the single robot path planning algorithm in step 6 and the position constraints generated in step S72, the single robot drivable path is generated again, and conflict detection is performed according to the conflict detection method in step S72. If there is a conflict, the position constraints are added again, and the process of generating single robot paths, conflict detection, and adding position constraints is repeated until conflict-free paths for multiple robots are obtained, and coordinated comprehensive scheduling of multi-robot systems is achieved based on the conflict-free paths.

Citation Information

Cited By

  • Robot cluster scheduling method, device and equipment

    CN119902446A

  • Multi-agent cooperative scheduling obstacle avoidance method and system

    CN121613941A