Mobile robot multi-target path planning method based on multilayer algorithm

Abstract

The invention discloses a multi-layer algorithm-based mobile robot multi-target path planning method. The method comprises the following steps of S1, obtaining an upper layer algorithm; in the multi-target planning process, we need to plan a large number of paths, and in the large number of paths, not each path spreads all over the whole map, so that on the basis of a DBVSB-P-RRT * algorithm, the search range is limited in an oval range on the basis of the idea of an Informed-RRT * optimization sampling area, and S2, middle-layer algorithm acquisition is carried out; and S3, obtaining a lower-layer algorithm. The three-layer algorithm with feedback is used for solving the problem of multi-target planning with obstacles, the multi-target planning method can plan a mobile robot path with better path quality on the premise of ensuring safety, and the three-layer algorithm with feedback can be used for solving the problem of multi-target planning with obstacles through a middle-layer feedback mechanism and a lower-layer feedback mechanism. And the length of a planned path in an environment of a complex target point can be obviously shortened.

Description

Technical Field

[0001] This invention belongs to the field of robot path planning, specifically relating to a multi-objective path planning method for mobile robots based on a multi-layer algorithm. Background Technology

[0002] With the increasing complexity of mobile robot applications, multi-objective path planning technology is facing unprecedented technical challenges. Traditional single-layer planning architectures are severely inadequate in addressing the multi-dimensional optimization needs of modern industrial environments: First, when simultaneously optimizing multiple objectives such as path length, safety, and energy consumption, the overall performance score of single-layer algorithms is often insufficient; second, existing methods have poor adaptability to dynamic environmental changes, and replanning time is generally long in scenarios with sudden obstacles; finally, the computational complexity of the algorithms increases exponentially, resulting in long average processing times for objective planning tasks on conventional computing platforms. These limitations directly restrict the depth and breadth of mobile robot applications in key areas such as intelligent manufacturing and smart logistics.

[0003] Existing technical solutions mainly fall into three categories: heuristic search algorithms (such as A* variants), which have a fast response speed but weak multi-objective optimization capabilities and insufficient Pareto front coverage; evolutionary algorithms (such as MOEA / D), which have good optimization effects but consume a lot of computational resources and cannot guarantee real-time performance; and traditional hierarchical methods, which have a clear structure but suffer from low inter-layer collaboration efficiency and imperfect feedback mechanisms.

[0004] To address these technical bottlenecks, this invention proposes a multi-layered, multi-objective robot path planning method with a feedback mechanism. This method first decomposes the multi-objective planning problem into two sub-problems: the TSP (Tracking, Path Planning) problem and the path planning problem. Then, a multi-layered algorithm with a feedback mechanism is used to solve them. By providing a fast, safe path with obstacle avoidance capabilities through the upper-layer algorithm, and through the effective cooperation between the middle and lower-layer algorithms under the feedback mechanism, this method can quickly output a safe path that passes through all target points with the shortest distance. Summary of the Invention

[0005] The technical problem to be solved by this invention is: for multi-objective path planning of mobile robots, a multi-layer algorithm with feedback mechanism is adopted. The upper-layer algorithm quickly provides a safe path with obstacle avoidance performance, and the middle and lower-layer algorithms cooperate effectively, thereby realizing the timeliness and reliability of multi-objective path planning.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a multi-objective path planning method for mobile robots based on multi-layer algorithms, comprising the following steps:

[0007] Step S1: Obtaining information via upper-level algorithm;

[0008] To ensure that effective path information can be quickly initialized in various spaces, the DBVSB-P-RRT* algorithm is improved and used as the main body of the upper-level algorithm;

[0009] In multi-objective programming, we need to plan a large number of paths. Not every path covers the entire map. Therefore, based on the DBVSB-P-RRT* algorithm, and based on the idea of ​​Informed-RRT* to optimize the sampling area, we limit its search range to an elliptical range, which greatly reduces the computational cost of sampling, expansion and collision detection.

[0010] The improved algorithm first initializes an elliptical region with a start and end point, then initializes two random trees, T1 and T2, at the start and target points respectively. Starting from T1, a pre-designed sampling function is used to sample and guide the random trees to expand towards the unobstructed region. After obtaining the nearest point, a pre-designed expansion function is used for expansion. Through the adaptive attraction and variable step size strategy of the function, the tree is guided towards the target point while adapting to multiple obstacle regions by changing the step size. When a new node is obtained, collision detection is performed to determine whether to add the new node to the tree. After collision detection, parent and child nodes are reselected, inheriting the advantages of the RRT* algorithm to ensure path quality. Subsequently, collision and distance checks are performed. If the conditions are not met, T1 and T2 switch to continue expansion, ensuring bidirectional tree expansion to improve planning efficiency. When the termination condition is met, the path is successfully planned and output.

[0011] Step S2: Obtaining intermediate layer algorithms;

[0012] The core of the mid-level algorithm is the ACS algorithm, used to solve the business travel problem and generate a sequence of planned points. The ACS algorithm differs from the ACO algorithm in three main ways: the utilization of search experience, state transition rules, and pheromone update rules. Regarding state transition rules, the ACS algorithm utilizes the search experience accumulated by ants more strongly than the ACO algorithm. The decision rule for ant k to move from node i to node j is determined by a pseudo-random proportional rule. The equation for the pseudo-random proportional rule is shown below:

[0013]

[0014] Where, p ij k τ is the probability that ant k moves from city i to city j. ij η represents the pheromone level along the edge between city i and city j. ijN represents heuristic information about the edges between city i and city j (e.g., the reciprocal of the distance). α is the exponent of pheromones, which determines the relative importance of pheromones and heuristic information. β is the exponent of heuristic information, which determines the relative importance of heuristic information and pheromones. i Let be the set of neighboring cities that ant k can move from city i. q is a random number, and q0 is a definite value in the range [0,1]. Regarding the utilization of search experience, pheromone evaporation and deposition only occur on paths belonging to the best possible path to date. Therefore, pheromone updates in ACS are implemented using the following equation:

[0015]

[0016] Where, Δτ ij L is the total amount of pheromone deposited by all ants on edge (i, j), which is usually inversely proportional to the path length of the ants. a is the pheromone carried by each ant. k (Q i Q j The distance () represents the length of the path traversed by the ant. Regarding pheromone evaporation, each time an ant passes by, it carries away a portion of the pheromone to increase the alternatives for path exploration. Local pheromone updates are achieved through the following equation:

[0017] τ ij (t+1)←(1-α)·τ ij (t)+α·τ0(t);

[0018] In this context, τ0 is a parameter. The effect of the local update rule is that whenever an ant traverses a path, its pheromone trace τ is updated. ij The amount of pheromone will be reduced. Therefore, this path becomes less attractive to another ant. Each ant chooses a path based on pheromones. To ensure that ants generate a large amount of pheromones in the early stages of iteration for rapid convergence, a variable pheromone strategy is adopted, as follows:

[0019]

[0020] Where iter represents the number of iterations, iter max This indicates the maximum number of iterations.

[0021] The ACS algorithm is commonly used to handle the TSP problem, but the path length used in the ACS algorithm is often expressed as the straight-line distance between two points. However, this does not satisfy the condition in real-world environments with obstacles. To handle the TSP problem in environments with obstacles, such as... Figure 1As shown, the ACS algorithm is used as the main algorithm in the middle layer, and the path information of the upper layer algorithm is introduced as the initial path, making the target sequence planned by the ACS algorithm effective in multi-obstacle environments. In this way, the upper and middle layer algorithms form an algorithm that can solve the TSP problem with obstacles.

[0022] Step S3: Obtaining from the lower-level algorithm;

[0023] The lower-level algorithm, based on IBIPF-RRT*, forms a feedback mechanism with the middle-level ACS algorithm to plan paths with optimal performance and good motion characteristics. Improvements to F-RRT* enhance its environmental adaptability and enable rapid path planning. During the sampling phase, the sampling space is optimized, referencing the Informed-RRT* algorithm. In the expansion phase, an adaptive gravity strategy is introduced to guide the expansion tree towards the target point, preventing the random tree from expanding excessively in other directions. Furthermore, the gravity adapts adaptively when approaching obstacles, overcoming the APF algorithm's tendency to get stuck in minima. In the connection phase, a bidirectional search strategy using shortest path optimization is employed. The path between the two target points is first searched, and then the shortest path optimization strategy ensures optimal overall path quality.

[0024] Specifically, for the mobile robot path planning described in this invention, planning only needs to be done within a small portion of the space, eliminating the need to spend a significant amount of time exploring the entire space. Therefore, during the sampling phase, the random tree is restricted to a single elliptical region for sampling, thereby reducing the randomness of sampling, significantly decreasing planning time, and lowering the upper limit of path length.

[0025] Specifically, the formulas for the gravitational field Ua and gravitational force Fa in the adaptive phase are as follows:

[0026]

[0027] Among them, U a Let λ be the gravitational field, and D(q, q) be the gravitational coefficient. goal D(q, q) is the distance to the target point. goal ) is the distance to the center of the obstacle. D obs It is a high-risk area near the obstacle, and also the range of the obstacle's repulsive force.

[0028] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0029] 1. This invention proposes a three-layer algorithm with feedback to solve multi-objective planning problems with obstacles.

[0030] 2. The multi-objective planning method proposed in this invention can plan a mobile robot path with better path quality while ensuring safety.

[0031] 3. The three-layer algorithm with feedback proposed in this invention can significantly shorten the length of the planned path in complex target point environments through the feedback mechanism of the middle and lower layers.

[0032] 4. The three-layer algorithm with feedback proposed in this invention can handle multi-objective path planning problems in complex environments. Attached Figure Description

[0033] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0034] Figure 1 This is a flowchart of a three-layer algorithm with a feedback mechanism.

[0035] Figure 2 This is a schematic diagram of path optimization.

[0036] Figure 3 This is a schematic diagram for shortest path optimization. Detailed Implementation

[0037] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Example 1

[0039] The following describes in detail a multi-objective path planning method for mobile robots based on a multi-layer algorithm, comprising the following steps:

[0040] Step S1: Obtaining information via upper-level algorithm;

[0041] To ensure that effective path information can be quickly initialized in various spaces, the DBVSB-P-RRT* algorithm is improved and used as the main body of the upper-level algorithm;

[0042] In multi-objective programming, we need to plan a large number of paths. Not every path covers the entire map. Therefore, based on the DBVSB-P-RRT* algorithm, and based on the idea of ​​Informed-RRT* to optimize the sampling area, we limit its search range to an elliptical range, which greatly reduces the computational cost of sampling, expansion and collision detection.

[0043] The improved algorithm first initializes an elliptical region with a start and end point, then initializes two random trees, T1 and T2, at the start and target points respectively. Starting from T1, a pre-designed sampling function is used to sample and guide the random trees to expand towards the unobstructed region. After obtaining the nearest point, a pre-designed expansion function is used for expansion. Through the adaptive attraction and variable step size strategy of the function, the tree is guided towards the target point while adapting to multiple obstacle regions by changing the step size. When a new node is obtained, collision detection is performed to determine whether to add the new node to the tree. After collision detection, parent and child nodes are reselected, inheriting the advantages of the RRT* algorithm to ensure path quality. Subsequently, collision and distance checks are performed. If the conditions are not met, T1 and T2 switch to continue expansion, ensuring bidirectional tree expansion to improve planning efficiency. When the termination condition is met, the path is successfully planned and output.

[0044] Step S2: Obtaining intermediate layer algorithms;

[0045] The core of the mid-level algorithm is the ACS algorithm, used to solve the business travel problem and generate a sequence of planned points. The ACS algorithm differs from the ACO algorithm in three main ways: the utilization of search experience, state transition rules, and pheromone update rules. Regarding state transition rules, the ACS algorithm utilizes the search experience accumulated by ants more strongly than the ACO algorithm. The decision rule for ant k to move from node i to node j is determined by a pseudo-random proportional rule. The equation for the pseudo-random proportional rule is shown below:

[0046]

[0047] Where, p ij k τ is the probability that ant k moves from city i to city j. ij η represents the pheromone level along the edge between city i and city j. ij N represents heuristic information about the edges between city i and city j (e.g., the reciprocal of the distance). α is the exponent of pheromones, which determines the relative importance of pheromones and heuristic information. β is the exponent of heuristic information, which determines the relative importance of heuristic information and pheromones. i Let be the set of neighboring cities that ant k can move from city i. q is a random number, and q0 is a definite value in the range [0,1]. Regarding the utilization of search experience, pheromone evaporation and deposition only occur on paths belonging to the best possible path to date. Therefore, pheromone updates in ACS are implemented using the following equation:

[0048]

[0049] Where, Δτ ijL is the total amount of pheromone deposited by all ants on edge (i, j), which is usually inversely proportional to the path length of the ants. a is the pheromone carried by each ant. k (Q i Q j The distance () represents the length of the path traversed by the ant. Regarding pheromone evaporation, each time an ant passes by, it carries away a portion of the pheromone to increase the alternatives for path exploration. Local pheromone updates are achieved through the following equation:

[0050] τ ij (t+1)←(1-α)·τ ij (t)+α·τ0(t);

[0051] In this context, τ0 is a parameter. The effect of the local update rule is that whenever an ant traverses a path, its pheromone trace τ is updated. ij The amount of pheromone will be reduced. Therefore, this path becomes less attractive to another ant. Each ant chooses a path based on pheromones. To ensure that ants generate a large amount of pheromones in the early stages of iteration for rapid convergence, a variable pheromone strategy is adopted, as follows:

[0052]

[0053] Where iter represents the number of iterations, iter max This indicates the maximum number of iterations.

[0054] The ACS algorithm is commonly used to handle the TSP problem, but the path length used in the ACS algorithm is often expressed as the straight-line distance between two points. However, this does not satisfy the condition in real-world environments with obstacles. To handle the TSP problem in environments with obstacles, such as... Figure 1 As shown, the ACS algorithm is used as the main algorithm in the middle layer, and the path information of the upper layer algorithm is introduced as the initial path, making the target sequence planned by the ACS algorithm effective in multi-obstacle environments. In this way, the upper and middle layer algorithms form an algorithm that can solve the TSP problem with obstacles.

[0055] Step S3: Obtaining from the lower-level algorithm;

[0056] The lower-level algorithm, based on IBIPF-RRT*, forms a feedback mechanism with the middle-level ACS algorithm to plan paths with optimal performance and good motion characteristics. Improvements to F-RRT* enhance its environmental adaptability and enable rapid path planning. During the sampling phase, the sampling space is optimized, referencing the Informed-RRT* algorithm. In the expansion phase, an adaptive gravity strategy is introduced to guide the expansion tree towards the target point, preventing the random tree from expanding excessively in other directions. Furthermore, the gravity adapts adaptively when approaching obstacles, overcoming the APF algorithm's tendency to get stuck in minima. In the connection phase, a bidirectional search strategy using shortest path optimization is employed. The path between the two target points is first searched, and then the shortest path optimization strategy ensures optimal overall path quality.

[0057] Specifically, for the mobile robot path planning described in this invention, planning only needs to be done within a small portion of the space, eliminating the need to spend a significant amount of time exploring the entire space. Therefore, during the sampling phase, the random tree is restricted to a single elliptical region for sampling, thereby reducing the randomness of sampling, significantly decreasing planning time, and lowering the upper limit of path length.

[0058] Specifically, the formulas for the gravitational field Ua and gravitational force Fa in the adaptive phase are as follows:

[0059]

[0060]

[0061] Among them, U a Let λ be the gravitational field, and D(q, q) be the gravitational coefficient. goal D(q, q) is the distance to the target point. goal ) is the distance to the center of the obstacle. D obs It is a high-risk area near the obstacle, and also the range of the obstacle's repulsive force.

[0062] Specifically, the bidirectional search strategy for shortest path optimization first performs a search and connection between the two target points. The search and connection process is as follows: Figure 2 As shown, starting from point q start and target point q goal Two random trees are planned for expansion. The blue lines and dots represent the nodes and branches of random tree T1, and the green lines and dots represent the nodes and branches of random tree T2. When the two random trees are a certain distance apart, the search is completed, a connection is made, and a connection q is planned. start and q goal The path.

[0063] Then, perform shortest path optimization on any two segments of the path connecting the two target points, as follows: Figure 3 As shown, the blue path is first formed during the optimization process, and then gradually optimized into a red dashed line to achieve a better effect.

[0064] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A multi-objective path planning method for mobile robots based on a multi-layer algorithm, characterized in that, Includes the following steps: Step S1: Obtaining information via upper-level algorithm; To ensure that effective path information can be quickly initialized in various spaces, the DBVSB-P-RRT* algorithm is improved and used as the main body of the upper-level algorithm; In multi-objective programming, we need to plan a large number of paths. Not every path covers the entire map. Therefore, based on the DBVSB-P-RRT* algorithm, and based on the idea of ​​Informed-RRT* to optimize the sampling area, we limit its search range to an elliptical range, which greatly reduces the computational cost of sampling, expansion and collision detection. The improved algorithm first initializes an elliptical region with a start and end point, then initializes two random trees, T1 and T2, at the start and target points respectively. Starting from T1, a pre-designed sampling function is used to sample and guide the random trees to expand towards the unobstructed region. After obtaining the nearest point, a pre-designed expansion function is used for expansion. Through the adaptive attraction and variable step size strategy of the function, the tree is guided towards the target point while adapting to multiple obstacle regions by changing the step size. When a new node is obtained, collision detection is performed to determine whether to add the new node to the tree. After collision detection, parent and child nodes are reselected, inheriting the advantages of the RRT* algorithm to ensure path quality. Subsequently, collision and distance checks are performed. If the conditions are not met, T1 and T2 switch to continue expansion, ensuring bidirectional tree expansion to improve planning efficiency. When the termination condition is met, the path is successfully planned and output. Step S2: Obtaining intermediate layer algorithms; The core of the mid-level algorithm is the ACS algorithm, used to solve the business travel problem and generate a sequence of planned points. The ACS algorithm differs from the ACO algorithm in three main ways: the utilization of search experience, state transition rules, and pheromone update rules. Regarding state transition rules, the ACS algorithm utilizes the search experience accumulated by ants more strongly than the ACO algorithm. The decision rule for ant k to move from node i to node j is determined by a pseudo-random proportional rule. The equation for the pseudo-random proportional rule is shown below: Where, p ij k τ is the probability that ant k moves from city i to city j. ij η represents the pheromone level along the edge between city i and city j. ij N represents heuristic information about the edges between city i and city j (e.g., the reciprocal of the distance). α is the exponent of pheromones, which determines the relative importance of pheromones and heuristic information. β is the exponent of heuristic information, which determines the relative importance of heuristic information and pheromones. i Let be the set of neighboring cities that ant k can move from city i. q is a random number, and q0 is a definite value in the range [0,1]. Regarding the utilization of search experience, pheromone evaporation and deposition only occur on paths belonging to the best possible path to date. Therefore, pheromone updates in ACS are implemented using the following equation: Where, Δτ ij L is the total amount of pheromone deposited by all ants on edge (i, j), which is usually inversely proportional to the path length of the ants. a is the pheromone carried by each ant. k (Q i Q j The distance () represents the length of the path traversed by the ant. Regarding pheromone evaporation, each time an ant passes by, it carries away a portion of the pheromone to increase the alternatives for path exploration. Local pheromone updates are achieved through the following equation: t ij (t+1)←(1-a)·t ij (t)+α·τ0(t); In this context, τ0 is a parameter. The effect of the local update rule is that whenever an ant traverses a path, its pheromone trace τ is updated. ij The amount of pheromone will be reduced. Therefore, this path becomes less attractive to another ant. Each ant chooses a path based on pheromones. To ensure that ants generate a large amount of pheromones in the early stages of iteration for rapid convergence, a variable pheromone strategy is adopted, as follows: Where iter represents the number of iterations, iter max This indicates the maximum number of iterations. The ACS algorithm is commonly used to solve the TSP problem, but the path length it uses is often expressed as the straight-line distance between two points. This doesn't satisfy the condition in real-world environments with obstacles. To address TSP problems in obstacle-filled environments, as shown in Figure 1, the middle-level algorithm uses the ACS algorithm as the core, incorporating path information from the upper-level algorithm as the initial path. This ensures that the target sequence planned by the ACS algorithm is effective in multi-obstacle environments. Thus, the upper and middle-level algorithms together form an algorithm capable of solving TSP problems with obstacles. Step S3: Obtaining from the lower-level algorithm; The lower-level algorithm, based on IBIPF-RRT*, forms a feedback mechanism with the middle-level ACS algorithm to plan paths with optimal performance and good motion characteristics. Improvements to F-RRT* enhance its environmental adaptability and enable rapid path planning. During the sampling phase, the sampling space is optimized, referencing the Informed-RRT* algorithm. In the expansion phase, an adaptive gravity strategy is introduced to guide the expansion tree towards the target point, preventing the random tree from expanding excessively in other directions. Furthermore, the gravity adapts adaptively when approaching obstacles, overcoming the APF algorithm's tendency to get stuck in minima. In the connection phase, a bidirectional search strategy using shortest path optimization is employed. The path between the two target points is first searched, and then the shortest path optimization strategy ensures optimal overall path quality. Specifically, for the mobile robot path planning described in this invention, planning only needs to be done within a small portion of the space, eliminating the need to spend a significant amount of time exploring the entire space. Therefore, during the sampling phase, the random tree is restricted to a single elliptical region for sampling, thereby reducing the randomness of sampling, significantly decreasing planning time, and lowering the upper limit of path length. Specifically, the formulas for the gravitational field Ua and gravitational force Fa in the adaptive phase are as follows: Among them, U a Let λ be the gravitational field, and D(q, q) be the gravitational coefficient. goal D(q, q) is the distance to the target point. goal ) is the distance to the center of the obstacle. D obs It is a high-risk area near the obstacle, and also the range of the obstacle's repulsive force.

Images