A Multimodal Robot Path Planning Method Based on the Improved Gray Wolf Algorithm with Multi-Strategy

Abstract

This invention discloses a multimodal robot path planning method based on a multi-strategy improved gray wolf algorithm. The method includes: capturing feature information of the map to be planned; transforming the coordinate system; constructing a total path cost function by integrating energy cost and obstacle threat cost; introducing a language gray wolf with a global perspective and setting algorithm parameters; updating the guidance strength of the language gray wolf and other guiding wolves using a nonlinear piecewise strategy adjustment factor; designing a chaotic Tent elite back-learning strategy to mutate the optimal position of the gray wolf; and performing optimal path planning based on the multi-strategy improved gray wolf algorithm and map information to obtain the optimal planned path. By using this invention, the planning efficiency and accuracy of the algorithm can be improved, while reducing the demand for computational resources. This invention, as a multimodal robot path planning method based on a multi-strategy improved gray wolf algorithm, can be widely applied in the field of robot path planning technology.

Description

Technical Field

[0001] This invention relates to the field of robot path planning, and in particular to a multimodal robot path planning method based on a multi-strategy improved gray wolf algorithm. Background Technology

[0002] Path planning is a crucial component of robot software systems and a hot research area in robotics. From a typological perspective, path planning methods can be categorized into global path planning and local path planning. Robots typically begin with global path planning based on known environmental information, aiming for the shortest time, shortest distance, least energy consumption, or lowest threat level—essentially a multi-dimensional optimization problem. However, in real-world scenarios, robots cannot fully grasp all environmental information, thus requiring dynamic adjustments to local paths based on feedback from various sensors to meet the planning objectives. Currently, global path planning methods for robots can be divided into graph search-based planning methods, sampling-based planning methods, and bio-inspired algorithm-based path planning methods. The first two require greater computational resources and are slower. Bio-inspired algorithm-based path planning methods, on the other hand, offer advantages such as simple structure, fast planning speed, and strong optimization capabilities, attracting attention from researchers both domestically and internationally. These algorithms draw inspiration from nature, solving optimization problems by simulating biological behavior.

[0003] The gray wolf optimization algorithm, whale optimization algorithm, and hippo optimization algorithm are all common bio-inspired algorithms, but they suffer from varying degrees of poor population diversity, slow convergence in later stages, and susceptibility to getting trapped in local optima. When used for mobile robot path planning, they struggle to find the optimal path in a given environment. Large language models, however, have significant advantages in decision-making. Combining their superior decision-making capabilities with traditional bio-inspired algorithms can improve the planning efficiency of the algorithm, demonstrating broad research prospects. Summary of the Invention

[0004] 1. Purpose of the invention:

[0005] To address the problems mentioned above, the present invention aims to provide a multimodal robot path planning method based on a multi-strategy improved gray wolf algorithm. This method combines a large language model with a traditional bio-inspired algorithm, utilizing the former's decision-making capabilities to accelerate the algorithm's convergence speed. Furthermore, it introduces a chaotic Tent elite back-learning strategy to update the optimal gray wolf's position, thereby improving the algorithm's planning accuracy.

[0006] 2. Technical Solution:

[0007] To achieve the above objectives, this invention provides a multimodal robot path planning method based on a multi-strategy improved gray wolf algorithm, comprising the following steps:

[0008] Step 1: Acquire map feature information through multimodal robot sensors and discretize it;

[0009] Step 2: Rotate the coordinate system around the origin so that the horizontal axis after rotation coincides with the line connecting the start and end points, and draw D+1 perpendicular lines to this line, one of which passes through the end point E. D is the total number of path points of the robot, let D = 5.

[0010] Step 3: Taking into account and standardizing the energy cost and obstacle threat cost of the multimodal robot, the total cost of path planning is calculated using the following formula:

[0011]

[0012] Among them, C total C represents the total cost of the path; e Energy cost; C t k1 is the cost of obstacle threats; k2 is the energy cost coefficient, which can be manually set according to mission objectives and requirements; Φ is the obstacle threat cost coefficient, which can be manually set according to mission objectives and requirements. e The maximum energy cost, set based on experience, can be manually set according to mission objectives and requirements; Φ t The maximum threat cost, set based on experience, can be manually set according to mission objectives and requirements;

[0013] Step 4: Set the parameters of the multi-strategy improved gray wolf algorithm; the number of wolves is N+W, consisting of α wolves, β wolves, δ wolves, ω wolves and ε wolves, where α wolves are the continuous guide wolves, responsible for gradually approaching the prey; β wolves are the suboptimal solutions that are definitely inferior to α wolves; δ wolves are the suboptimal solutions that are definitely inferior to β wolves; ordinary gray wolves are ω wolves; the total number of α wolves, β wolves, δ wolves and ω wolves is N; ε wolves are language gray wolves, which can provide solutions from a global perspective under the guidance of the large language model, continuously guide ω wolves, and their number is W; the number of iterations is EP, and the maximum number of iterations is MaxEP.

[0014] Step 5: Under the simultaneous guidance of α, β, δ and ε wolves, the next position of an ω wolf is X(t+1). The guidance strength of α, β, δ and ε wolves is updated through a nonlinear piecewise strategy adjustment factor.

[0015] Step 6: At the optimal position of the ε wolf, use Chaotic Tent Elite Backward Learning Mutation to improve the Elite Backward Learning process.

[0016] Step 7: Minimize the total cost C of the path total To achieve the goal of path planning, we propose to plan the cost-optimal path based on a multi-strategy improved Grey Wolf algorithm and map feature information.

[0017] Furthermore, in step 1, the feature information includes the robot's starting position, ending position, and the position information and range of the obstacle after it has been expanded into a circular obstacle.

[0018] Furthermore, in step 2, there is a path planning waypoint on each vertical line. The 6 vertical lines correspond to 5 process waypoints and 1 endpoint, thus transforming the path planning problem into a 5-dimensional function optimization problem.

[0019] Furthermore, in step 3, the coordinates of the robot and the obstacle, after rotation of the coordinate system, can be expressed as:

[0020]

[0021] Where x and y are the original x and y coordinates; x′ and y′ are the new x and y coordinates after the coordinate system has been rotated; x E and y E The original x and y coordinates of the endpoint E; x S and y S Let S be the original x and y coordinates of the starting point S.

[0022] Furthermore, in step 4, the energy cost is defined by the following formula:

[0023]

[0024] Where D is the total number of waypoints for the robot; E is the energy consumption per unit distance for the robot at a constant speed; x i Let y be the x-coordinate of the robot at the i-th path point; i Let be the ordinate of the robot at the i-th path point.

[0025] Connect adjacent path points pairwise, and take 5 points at equal intervals from each segment. Assuming that the sub-path connected to path points i and i+1 is within the expanded circular obstacle range, the obstacle threat cost is calculated using the following formula:

[0026]

[0027] Where M is the number of circular obstacles; O i d is the path length of the i-th sub-path; 4 0.1,i,m d 4 0.3,i,m d 4 0.5,i,m d 4 0.7,i,m d 4 0.9,i,mR represents the distances from the 10%, 30%, 50%, 70%, and 90% points of the i-th sub-route segment to the center of the m-th obstacle circle; m Let R be the threat level of the m-th obstacle, the value of which is determined by the semantic segmentation result: when the obstacle is an enemy radar, missile, anti-aircraft gun, or electronic jamming equipment, R is... m =6; R in other cases m =4.

[0028] Furthermore, in step 5, the position of the ε wolf at the beginning of each iteration is initialized to the position X of the α wolf. α Draw a line V that passes through the current position and is perpendicular to the starting point S and the ending point E. EP In the iteration, ε wolf will V EP The information about the region between the endpoint and the vertical line is provided to a large language model that has been pre-configured with specific prompt words (Prompt). This process is expressed as follows:

[0029] (x ω,i+1 ,y ω,i+1 )=LLM(Prompt,Map_Information,θ);

[0030] Among them, (x ω,i+1 ,y ω,i+1 ) represents the next position of ε Wolf planned by the large language model, LLM() represents the output of the large language model, Prompt is the preset prompt word, Map_Information is the original information provided by ε Wolf, i.e. the feature information of the region, and θ is the key parameters of the large language model, including top_p, top_k, Temperature, Max_Tokens, etc.

[0031] Furthermore, in step 6, the next position X(t+1) of the ω wolf is:

[0032]

[0033] Among them, X α,i X β,i X δ,i X ε,i The guidance positions generated by ω wolf under the guidance of α wolf, β wolf, δ wolf, and ε wolf are respectively; λ is a nonlinear piecewise strategic adjustment factor; in the early stage, ε wolf is in a dominant leadership position, quickly determining the optimal solution and optimal encirclement area based on a global perspective; during the iteration process, ε wolf based on the language model is always independently guided and constrained by the large language model, and cannot be continuously optimized in each iteration; the language model cannot provide fine-grained decision-making, so in the later stage, the leadership position of ε wolf is much weaker than that of α wolf, β wolf, and δ wolf, and the algorithm focuses on concentrating on the local encirclement area to improve convergence accuracy; the formula used is:

[0034]

[0035] Where MaxEP is the maximum number of iterations; To optimize the adjustment parameters, set f(t) is the fitness function, expressed by the formula:

[0036]

[0037] Maxf(t) is the maximum fitness during the iteration process.

[0038] Furthermore, in step 7, the chaotic Tent elite reverse learning mutation is performed at the optimal position of the ε wolf, using the following formula:

[0039] X * best =Y D (X l best +X h best )-X best (9)

[0040] Where X * best The optimal position for reverse learning of chaotic Tent elites; X h best and X l best The optimal position X for the gray wolf is shown below. best The maximum and minimum values ​​of Y; d For the d-th chaotic sequence value between 0 and 1, the mapping sequence model is:

[0041]

[0042] Where k is the chaos coefficient.

[0043] 3. The multimodal robot path planning method based on the multi-strategy improved gray wolf algorithm of this invention has the following advancements and advantages compared with the prior art:

[0044] (1) Introducing ε wolf with a global perspective and participating in the position guidance of ω wolf improves the planning efficiency and mining efficiency of the algorithm and speeds up the early convergence speed.

[0045] (2) Since ε wolf only has a significant guiding effect on macro-global planning, the guiding effect of ε wolf gradually weakens as the iteration progresses, thus effectively avoiding the situation where the algorithm gets stuck in local optima in the early stage.

[0046] (3) The ε wolf improves the planning accuracy by using the chaotic Tent elite reverse learning strategy to mutate at the optimal position.

[0047] In summary, this method is a robot planning method with high planning efficiency and high planning accuracy. Attached Figure Description

[0048] Figure 1 This is a flowchart of the steps of the multimodal robot path planning method based on the multi-strategy improved gray wolf algorithm provided by the present invention;

[0049] Figure 2 This invention relates to the hierarchy of the gray wolf pack in the early iteration process of the multimodal robot path planning method based on the improved gray wolf algorithm using a multi-strategy approach.

[0050] Figure 3 This invention relates to the hierarchy of the gray wolf pack in the later iterations of the multimodal robot path planning method based on the improved gray wolf algorithm using a multi-strategy approach.

[0051] Figure 4 This invention provides a 6×6 grid map for a multimodal robot path planning method based on a multi-strategy improved gray wolf algorithm. Detailed Implementation

[0052] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and specific examples:

[0053] Figure 1 This is a flowchart illustrating the steps of the multimodal robot path planning method based on the multi-strategy improved gray wolf algorithm described in this invention. The specific implementation steps of this method include:

[0054] S1. The multimodal robot acquires feature information from a 6×6 map, including the robot's starting position, ending position, and the position and range of obstacles after they are expanded into circular obstacles, and then discretizes them.

[0055] S2. Rotate the coordinate system around the origin so that the horizontal axis coincides with the line connecting the start and end points. The formula is as follows:

[0056]

[0057] Draw 6 perpendicular lines to the line connecting them, one of which passes through the endpoint E;

[0058] S3. Taking into account and standardizing the energy cost and obstacle threat cost of the multimodal robot, the total cost of path planning is calculated using the following formula:

[0059]

[0060] Among them, C total C represents the total cost of the path; eEnergy cost; C t K1 represents the obstacle threat cost; K2 represents the energy cost coefficient; Φ represents the obstacle threat cost coefficient. e Let Φ be the maximum energy cost set based on experience. e =30; Φ t Let Φ be the maximum threat cost set based on experience. t =20;

[0061] Furthermore, the energy cost is calculated using the following formula:

[0062]

[0063] Where D is the total number of waypoints for the robot, let D = 5; E is the energy consumption per unit distance for the robot at a constant speed, let E = 5 (UEC / UL), that is, the robot consumes 5 units of energy per unit distance moved; x i Let y be the x-coordinate of the robot at the i-th path point; i Let be the ordinate of the robot at the i-th path point;

[0064] Connect adjacent path points pairwise, and take 5 points at equal intervals from each segment. Assuming that the sub-path connected to path points i and i+1 is within the expanded circular obstacle range, the obstacle threat cost is calculated using the following formula:

[0065]

[0066] Where M is the number of circular obstacles, let M = 9; O i d is the path length of the i-th sub-path; 4 0.1,i,m d 4 0.3,i,m d 4 0.5,i,m d 4 0.7,i,m d 4 0.9,i,m R represents the distances from the 10%, 30%, 50%, 70%, and 90% points of the i-th sub-route segment to the center of the m-th obstacle circle; m Let R be the threat level of the m-th obstacle, the value of which is determined by the semantic segmentation result: when the obstacle is an enemy radar, missile, anti-aircraft gun, or electronic jamming equipment, R is... m =6; R in other cases m =4.

[0067] S4. Set the parameters for the multi-strategy improved gray wolf algorithm; the number of wolves N is N+W, consisting of α wolves, β wolves, δ wolves, ω wolves, and ε wolves, where α wolves are the continuous guide wolves responsible for gradually approaching the prey; β wolves are the suboptimal solutions that are always inferior to α wolves; δ wolves are the suboptimal solutions that are always inferior to β wolves; ordinary gray wolves are ω wolves; the total number of α wolves, β wolves, δ wolves, and ω wolves is N, let N = 40; ε wolves are language gray wolves, which can provide solutions from a global perspective under the guidance of the large language model, continuously guide ω wolves, and their number is W, let W = 1; the number of iterations is EP, which affects the search capability of the algorithm, and the maximum number of iterations is MaxEP, let MaxEP = 500.

[0068] Furthermore, the position of wolf ε at the beginning of each iteration is initialized to the position X of wolf α. α Draw a line V that passes through the current position and is perpendicular to the starting point S and the ending point E. EP In the iteration, ε wolf will V EP The information about the region between the endpoint and the vertical line is provided to a large language model that has been pre-configured with specific prompt words (Prompt). This process is expressed as follows:

[0069] (x ω,i+1 ,y ω,i+1 )=LLM(Prompt,Map_Information,θ);

[0070] Among them, (x ω,i+1 ,y ω,i+1 ) represents the next position of εwolf planned by the large language model, LLM() represents the output of the large language model, Prompt is the preset prompt word, Map_Information is the original information provided by εwolf, i.e. the feature information of the region, and θ is the key parameter of the large language model.

[0071] S5. Under the guidance of α, β, δ and ε wolves, the next position of an ω wolf is X(t+1). The guidance strength of α, β, δ and ε wolves is updated by a nonlinear piecewise strategy adjustment factor.

[0072] Furthermore, the next position X(t+1) of the ω wolf is:

[0073]

[0074] Among them, X α,i X β,i X δ,i X ε,iThe guidance positions generated by ω wolf under the guidance of α wolf, β wolf, δ wolf, and ε wolf are respectively; λ is a nonlinear piecewise strategic adjustment factor; in the early stage, ε wolf is in a dominant leadership position, quickly determining the optimal solution and optimal encirclement area based on a global perspective; during the iteration process, ε wolf based on the language model is always independently guided and constrained by the large language model, and cannot be continuously optimized in each iteration; the language model cannot provide fine-grained decision-making, so in the later stage, the leadership position of ε wolf is much weaker than that of α wolf, β wolf, and δ wolf, and the algorithm focuses on concentrating on the local encirclement area to improve convergence accuracy; the formula used is:

[0075]

[0076] Where MaxEP is the maximum number of iterations; To optimize the adjustment parameters, f(t) is the fitness function, expressed by the formula:

[0077] f(t) = ||X(t) - X(t-1)||;

[0078] Maxf(t) is the maximum fitness during the iteration process.

[0079] S6. At the optimal position of the ε wolf, adopt chaotic Tent elite reverse learning mutation;

[0080] Furthermore, the chaotic Tent elite reverse learning mutation at the optimal position of the ε wolf is performed using the following formula:

[0081] X * best =Y d (X l best +X h best )-X best

[0082] Where X * best The optimal position for reverse learning of chaotic Tent elites; X h best and X l best The optimal position X for the gray wolf is shown below. best The maximum and minimum values ​​of Y; d For the d-th chaotic sequence value between 0 and 1, the mapping sequence model is:

[0083]

[0084] Where k is the chaos coefficient, let k = 0.6.

[0085] S7. Minimize the total cost C of the path. totalPlan the objectives for path planning.

[0086] The present invention has been described according to specific embodiments, but is not limited to the above-described embodiments. Equivalent modifications or substitutions made by those skilled in the art without departing from the scope of the claims are all within the scope of protection of the present invention, and the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A multimodal robot path planning method based on a multi-strategy improved gray wolf algorithm, characterized in that, Includes the following steps: Step 1: Acquire map feature information through multimodal robot sensors and discretize it; Step 2: Rotate the coordinate system around the origin so that the horizontal axis after rotation coincides with the line connecting the start and end points, and draw D+1 perpendicular lines to this line, one of which passes through the end point E. D is the total number of path points of the robot. Step 3: Taking into account and standardizing the energy cost and obstacle threat cost of the multimodal robot, the total cost of path planning is calculated using the following formula: Among them, C total C represents the total cost of the path; e Energy cost; C t K1 represents the obstacle threat cost; K2 represents the energy cost coefficient; Φ represents the obstacle threat cost coefficient. e Maximum energy cost set based on experience; Φ t Maximum threat cost set based on experience; Step 4: Set the parameters of the multi-strategy improved gray wolf algorithm; the wolf pack size is N+W, consisting of α wolves, β wolves, δ wolves, ω wolves, and ε wolves, where α wolves are the continuous guide wolves responsible for gradually approaching the prey; β wolves are always suboptimal solutions below α wolves; δ wolves are always suboptimal solutions below β wolves; ordinary gray wolves are ω wolves; the total number of α wolves, β wolves, δ wolves, and ω wolves is N; ε wolves are language gray wolves, which can provide solutions from a global perspective under the guidance of the large language model, continuously guiding ω wolves, and their number is W; the number of iterations is EP, and the maximum number of iterations is MaxEP; Step 5: Under the simultaneous guidance of α, β, δ and ε wolves, the next position of an ω wolf is X(t+1). The guidance strength of α, β, δ and ε wolves is updated through a nonlinear piecewise strategy adjustment factor. Step 6: At the optimal position of the ε wolf, use chaotic Tent elites to reverse learn mutation; Step 7: Minimize the total cost C of the path total To achieve the goal of path planning, we propose to plan the cost-optimal path based on a multi-strategy improved Grey Wolf algorithm and map feature information.

2. The multimodal robot path planning method based on the multi-strategy improved gray wolf algorithm according to claim 1, characterized in that: In step 1, the feature information includes the robot's starting position, ending position, and the position information and range of the obstacle after it has been expanded into a circular obstacle.

3. The multimodal robot path planning method based on the multi-strategy improved gray wolf algorithm according to claim 1, characterized in that: In step 2, there is a path planning point on each vertical line. The D+1 vertical lines correspond to D process path points and 1 endpoint, thus transforming the path planning problem into a D-dimensional function optimization problem.

4. The multimodal robot path planning method based on the multi-strategy improved gray wolf algorithm according to claim 1, characterized in that: In step 2, the coordinates of the robot and the obstacle, after rotation of the coordinate system, can be expressed as: Where x and y are the original x and y coordinates; x′ and y′ are the new x and y coordinates after the coordinate system has been rotated; x E and y E The original x and y coordinates of the endpoint E; x S and y S Let S be the original x and y coordinates of the starting point S.

5. The multimodal robot path planning method based on the multi-strategy improved gray wolf algorithm according to claim 1, characterized in that: In step 3, the energy cost is calculated using the following formula: Where D is the total number of waypoints for the robot; E is the energy consumption per unit distance for the robot at a constant speed; x i Let y be the x-coordinate of the robot at the i-th path point; i Let be the ordinate of the robot at the i-th path point; Connect adjacent path points pairwise, and take D points at equal intervals from each segment. Assuming that the sub-path connected to path points i and i+1 is within the expanded circular obstacle range, the obstacle threat cost is calculated using the following formula: Where M is the number of circular obstacles; O i d is the path length of the i-th sub-path; 4 0.1,i,m d 4 0.3,i,m d 4 0.5,i,m d 4 0.7,i,m d 4 0.9,i,m R represents the distances from the 10%, 30%, 50%, 70%, and 90% points of the i-th sub-route segment to the center of the m-th obstacle circle; m Let R be the threat level of the m-th obstacle, the value of which is determined by the semantic segmentation result: when the obstacle is an enemy radar, missile, anti-aircraft gun, or electronic jamming equipment, R is... m =6; R in other cases m =4.

6. The multimodal robot path planning method based on the multi-strategy improved gray wolf algorithm according to claim 1, characterized in that: In step 4, the position of wolf ε at the beginning of each iteration is initialized to the position X of wolf α. α Draw a line V that passes through the current position and is perpendicular to the starting point S and the ending point E. EP In the iteration, ε wolf will V EP The information about the region between the endpoint and the vertical line is provided to a large language model that has been pre-configured with specific prompt words (Prompt). This process is expressed as follows: (x ω,i+1 ,y ω,i+1 )=LLM(Prompt,Map_Information,θ); Among them, (x ω,i+1 ,y ω,i+1 ) represents the next position of ε Wolf planned by the large language model, LLM() represents the output of the large language model, Prompt is the preset prompt word, Map_Information is the original information provided by ε Wolf, i.e. the feature information of the region, and θ is the key parameters of the large language model, including top_p, top_k, Temperature, Max_Tokens, etc.

7. The multimodal robot path planning method based on the multi-strategy improved gray wolf algorithm according to claim 1, characterized in that: In step 5, the next position X(t+1) of the ω wolf is: Among them, X α,i X β,i X δ,i X ε,i The guidance positions generated by ω wolf under the guidance of α wolf, β wolf, δ wolf, and ε wolf are respectively; λ is a nonlinear piecewise strategic adjustment factor; in the early stage, ε wolf is in a dominant leadership position, quickly determining the optimal solution and optimal encirclement area based on a global perspective; during the iteration process, ε wolf based on the language model is always independently guided and constrained by the large language model, and cannot be continuously optimized in each iteration; the language model cannot provide fine-grained decision-making, so in the later stage, the leadership position of ε wolf is much weaker than that of α wolf, β wolf, and δ wolf, and the algorithm focuses on concentrating on the local encirclement area to improve convergence accuracy; the formula used is: Where MaxEP is the maximum number of iterations; To optimize the adjustment parameters; f(t) is the fitness function, and the formula is: f(t) = ||X(t) - X(t-1)||; Maxf(t) is the maximum fitness during the iteration process.

8. The multimodal robot path planning method based on the multi-strategy improved gray wolf algorithm according to claim 1, characterized in that: In step 6, the chaotic Tent elite reverse learning mutation is performed at the optimal position of the ε wolf, using the following formula: X * best =Y D (X l best +X h best )-X best Where X * best The optimal position for reverse learning of chaotic Tent elites; X h best and X l best The optimal position X for the gray wolf is shown below. best The maximum and minimum values ​​of Y; d For the d-th chaotic sequence value between 0 and 1, the mapping sequence model is: Where k is the chaos coefficient.

Images