Multi-AGV textile workshop material distribution path dynamic planning method and system

Abstract

The invention belongs to the technical field of path planning, and particularly relates to a multi-AGV textile workshop material distribution path dynamic planning method and system, and the method comprises the following steps: S1, obtaining the position and speed of each AGV, and the Lidar and visual data of the environment, and recognizing and classifying the barriers, such as textile equipment, raw material stacks, workers and the like, in the environment through a semantic segmentation network; and fusing the position information of the Lidar point cloud and the semantic category information of the obstacle, constructing an asymmetric probabilistic risk potential field for different categories of obstacles, and ensuring that the risk potential field gradient of a region corresponding to a worker is the highest. According to the method, the AGV can accurately distinguish different obstacles such as working personnel and equipment by utilizing the semantic segmentation network and constructing the asymmetric risk potential field, so that the operation safety in a complex man-machine co-fusion environment is improved, and particularly, the safety of the personnel is guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of path planning technology, specifically relating to a dynamic planning method and system for material delivery paths in textile workshops with multiple AGVs. Background Technology

[0002] In a typical textile workshop environment, multiple AGVs need to work collaboratively in a complex scenario. Therefore, it is crucial to plan efficient, collision-free paths for the AGV cluster that can adapt to changing conditions. Currently, mainstream methods for multi-robot path planning include algorithms such as Dijkstra's algorithm based on centralized planning, and velocity-based obstacle avoidance (VO) methods and their variants based on distributed obstacle avoidance. While centralized methods can obtain globally optimal solutions, they suffer from high computational complexity and struggle to handle real-time environmental changes. In contrast, the VO method and its derivatives, such as reciprocal velocity obstacle avoidance and hybrid reciprocal velocity obstacle avoidance, guide individual robots to perform local obstacle avoidance by constructing conflict zones in velocity space, exhibiting good real-time performance and scalability.

[0003] However, traditional methods typically treat all obstacles as indistinguishable geometric entities, lacking semantic understanding of the environment and failing to assess risk based on obstacle type. When constructing speed obstacles, most methods employ symmetrical, equal collision avoidance responsibility allocation strategies, neglecting heterogeneous states such as task urgency, remaining battery power, and right-of-way for each AGV. For example, an AGV with sufficient battery power and a less urgent task should assume more avoidance responsibility to ensure priority passage for an AGV with depleted battery power performing a high-priority task. Furthermore, traditional optimization objectives are relatively singular when selecting execution speed, typically only considering minimizing the deviation from the desired speed while ignoring potential risks to the trajectory in the short term, lacking a comprehensive trade-off between safety and efficiency. Therefore, a novel dynamic path planning method that integrates environmental semantic information, achieves asymmetric collision avoidance responsibility allocation, and combines multiple optimization objectives is urgently needed. Summary of the Invention

[0004] This invention provides a method and system for dynamic planning of material delivery paths in textile workshops with multiple AGVs, in order to solve the technical problem that traditional methods have relatively singular optimization objectives and lack a comprehensive balance between safety and efficiency.

[0005] In a first aspect, the present invention provides a dynamic planning method for material delivery paths in a textile workshop using multiple AGVs, comprising the following steps: S1: Acquire the position, speed, and environmental LiDAR and visual data of each AGV; use a semantic segmentation network to identify and classify obstacles in the environment such as textile equipment, raw material stacks, and workers; fuse the position information of LiDAR point clouds with the semantic category information of obstacles to construct an asymmetric probabilistic risk potential field for different categories of obstacles, and ensure that the risk potential field gradient of the area corresponding to the workers is the highest. S2, For a pair of AGVs with collision risk, calculate an anisotropic velocity barrier with nonlinearly adjusted boundary according to the passage rules of the functional area of ​​the workshop and their respective task priorities; S3. Based on the remaining power and task time margin of each pair of AGVs, calculate an unbalanced reciprocal responsibility allocation coefficient, and use the reciprocal responsibility allocation coefficient to offset the anisotropic speed barrier to generate a hybrid reciprocal speed barrier. S4. Under the kinematic constraints of the AGV, search for an optimal speed that minimizes the weighted cost function in the allowed speed space outside the hybrid reciprocal speed obstacle. The weighted cost function is a weighted combination of the expected speed deviation and the integral of the predicted trajectory risk potential field, where the weights are adjusted according to the current task priority of the AGV.

[0006] Furthermore, semantic segmentation networks are used to identify and classify obstacles in the environment, such as textile equipment, raw material stacks, and workers, including: The 640×480 pixel RGB image captured by the AGV's front camera is input into the pre-trained DeepLabv3+ semantic segmentation model; The DeepLabv3+ semantic segmentation model classifies each pixel in an RGB image and outputs a semantic mask of the same size as the input image, where different pixel values ​​represent textile equipment, raw material stacks, workers, or passable ground areas. Based on the outlines and center points of various obstacle pixel clusters in the semantic mask, the bounding boxes and categories of obstacles are determined.

[0007] Furthermore, asymmetric probabilistic risk potential fields are constructed for different types of obstacles, including: Map the semantic categories of obstacles to the basic risk coefficient; A two-dimensional Gaussian potential field model is constructed with the geometric center of the obstacle as the mean. According to the movement direction of the AGV, the standard deviation of the two-dimensional Gaussian potential field in the AGV's forward direction is set to 1.5 times the obstacle size, and the standard deviation in the vertical direction is set to 1.2 times the obstacle size, forming an asymmetric distribution. Multiplying the basic risk coefficient by a two-dimensional Gaussian potential field yields a probabilistic risk potential field, ensuring that the gradient magnitude of the risk potential field in the area corresponding to the staff is the largest among all obstacles.

[0008] Furthermore, the calculation of an anisotropic velocity barrier with a nonlinearly adjusted boundary includes: Based on the relative position and size of a pair of AGVs, calculate a standard anisotropic velocity barrier represented as an ellipse; If the functional area where the AGV is located is the main channel, then the length of the major half-axis of the standard anisotropic velocity barrier in the direction of AGV movement is multiplied by an adjustment factor of 1.2, and the length of the minor half-axis in the vertical direction is multiplied by an adjustment factor of 0.8. If the task priority of the AGV is lower than that of the other AGV, the Sigmoid function is used to perform nonlinear deformation on the standard anisotropic velocity barrier boundary, so that it bulges outward in the velocity direction of the other AGV. The amount of deformation is positively correlated with the difference in task priority.

[0009] Furthermore, the calculation of a non-equilibrium reciprocal responsibility allocation coefficient includes: For a pair of AGVs Responsibility allocation factor The calculation formula is:

[0010] in, for The remaining battery percentage, for Task time margin in seconds; calculate Responsibility allocation factor ; Will By normalizing the responsibility allocation factors, we obtain the reciprocal responsibility allocation coefficient. :

[0011] Using the reciprocal responsibility allocation coefficient Migrate anisotropic velocity barriers to generate Hybrid reciprocal speed barrier.

[0012] Furthermore, the weighted cost function is a weighted combination of the integral of the expected velocity deviation and the risk potential field of the predicted trajectory. Defined as:

[0013] Where v is the candidate velocity, For the desired speed, This represents the maximum speed of the AGV. The predicted position of candidate velocity v at time t Risk potential value on the field, To predict the time domain, For efficiency weighting, For safety weights.

[0014] Furthermore, under the kinematic constraints of the AGV, an optimal speed that minimizes the weighted cost function is searched within the allowable speed space outside the hybrid reciprocal speed barrier, including: A series of candidate velocities are generated within the allowable velocity space discretized with a velocity resolution of 0.1 meters per second and an angular resolution of 5°. For each candidate velocity, extrapolate the trajectory 3 seconds ahead, i.e., predict the time domain. The trajectory is sampled every 0.2 seconds, for a total of 15 sampling points. The integral of the trajectory risk potential field is approximately predicted by calculating the weighted sum of the risk potential field values ​​of the sampling points. Calculate the weighted cost function value for each candidate velocity by combining the expected velocity deviation; The candidate speed that minimizes the weighted cost function value is selected as the optimal speed.

[0015] Furthermore, in S1, the semantic segmentation model is the YOLOv8-seg model.

[0016] Furthermore, in S1, the global position of the AGV is obtained through a real-time differential GPS or UWB module, and the current speed is obtained through fusion calculation of wheel odometer and inertial measurement unit.

[0017] Secondly, the present invention provides a dynamic planning system for material delivery paths in a textile workshop with multiple AGVs, including a memory and a processor. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned dynamic planning method for material delivery paths in a textile workshop with multiple AGVs is implemented.

[0018] The beneficial effects are as follows: By utilizing semantic segmentation networks and constructing an asymmetric risk potential field, this invention enables AGVs to accurately distinguish between different obstacles such as workers and equipment, improving operational safety in complex human-machine collaborative environments, especially for personnel safety. By integrating multi-dimensional state information such as the task priority, remaining battery power, and passage rules of each AGV, an unbalanced reciprocal collision avoidance responsibility allocation mechanism is established, granting priority passage to high-priority or urgent AGVs, thus improving the collaborative operation efficiency and material delivery smoothness of the entire AGV cluster. By employing a weighted cost function that balances desired speed and predicted trajectory risk to optimize speed, the AGV's decision-making achieves a better balance, ensuring the overall reliability of the path planning scheme. Attached Figure Description

[0019] Figure 1 A flowchart for a dynamic planning method for material delivery paths in a textile workshop with multiple AGVs. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] An embodiment of the dynamic planning method for material delivery paths in a textile workshop using multiple AGVs provided by this invention: like Figure 1 As shown, the dynamic planning method for material delivery paths in a textile workshop with multiple AGVs includes the following steps: S1: Acquire the position, speed, and environmental LiDAR and visual data of each AGV. Use a semantic segmentation network to identify and classify obstacles in the environment, such as textile equipment, raw material stacks, and workers. Integrate the position information of LiDAR point clouds with the semantic category information of obstacles to construct an asymmetric probabilistic risk potential field for different categories of obstacles, and ensure that the risk potential field gradient is highest in the area corresponding to the workers.

[0022] Specifically, the global position of the AGV is obtained through an onboard real-time differential GPS or UWB module, and the current speed is calculated by fusing wheel odometers and inertial measurement units (IMUs). Simultaneously, a forward-facing RGB industrial camera acquires environmental image information, and a 360° field-of-view LiDAR scanner is used to scan and obtain 3D point cloud data of the surrounding environment. The real-time images captured by the camera are input into a semantic segmentation model, such as the YOLOv8-seg model, pre-trained on a textile workshop scene dataset. This model can classify each pixel in the image and output a semantic mask map of the same size as the original image, where different colors or grayscale values ​​represent different categories, thereby identifying static textile equipment, raw material stacks, and moving workers and other AGVs.

[0023] By using joint calibration parameters of the camera and LiDAR, the 3D LiDAR point cloud is projected onto a 2D semantic mask image plane, and each LiDAR point is assigned a corresponding semantic category label, forming a semantic point cloud. Based on this semantic point cloud, a 2D raster map-like risk potential field is constructed. The potential field value U of any point p in the risk potential field is contributed by all surrounding obstacles, and the calculation formula is: [Formula for all obstacle points]. The sum of the generated potential fields. For a single obstacle, the generated potential field values ​​are modeled using a Gaussian function. Where d is the distance between point p and obstacle point distance, and This is a parameter related to obstacle category c. To achieve asymmetry, for this type of obstacle, the potential field standard deviation is... Stretching in the direction of motion and compression in the perpendicular direction forms elliptical equipotential lines. To ensure the highest risk for workers, the potential field amplitude corresponding to the category is... Set it to the highest value among all categories, for example, set it to 1.0, while setting textile equipment to 0.6 and raw material stacking to 0.3, so that the surrounding potential field value is the highest and the gradient change is the most dramatic.

[0024] In an optional embodiment, a semantic segmentation network is used to identify and classify obstacles in the environment, such as textile equipment, raw material stacks, and workers, including: The 640×480 pixel RGB image captured by the AGV's front camera is input into the pre-trained DeepLabv3+ semantic segmentation model; The DeepLabv3+ semantic segmentation model classifies each pixel in an RGB image and outputs a semantic mask of the same size as the input image, where different pixel values ​​represent textile equipment, raw material stacks, workers, or passable ground areas. Based on the outlines and center points of various obstacle pixel clusters in the semantic mask, the bounding boxes and categories of obstacles are determined.

[0025] Specifically, the AGV's front-facing camera captures a 640×480 pixel color image stream at 30 frames per second. Each frame is fed in real-time into the DeepLabv3+ model running in the onboard computing unit. This model, trained on a large number of textile workshop scene images, is able to recognize the content in the images. For example, after processing a frame, the model generates a 640×480 matrix, i.e., a semantic mask. In this matrix, all pixels belonging to workers are assigned a value of 1, pixels belonging to raw material stacks are assigned a value of 2, pixels belonging to textile equipment are assigned a value of 3, and the ground that is safe to pass through is assigned a value of 0.

[0026] The semantic mask matrix is ​​processed, and an image connected component analysis algorithm is used to find clusters of pixels with values ​​of 1, 2, or 3. For example, if a connected region consisting of pixels with a value of 1 is found, the minimum bounding rectangle of this region is calculated. Assuming the top-left corner of the rectangle is (120, 250), the bottom-right corner is (180, 400), and the center point is (150, 325), the rectangle is recorded as an obstacle of the category "worker," and its position and size information are passed to the subsequent path planning module. This process is repeated for all identified obstacle categories in the image.

[0027] In an optional embodiment, an asymmetric probabilistic risk potential field is constructed for different categories of obstacles, including: Map the semantic categories of obstacles to the basic risk coefficient; A two-dimensional Gaussian potential field model is constructed with the geometric center of the obstacle as the mean. According to the movement direction of the AGV, the standard deviation of the two-dimensional Gaussian potential field in the AGV's forward direction is set to 1.5 times the obstacle size, and the standard deviation in the vertical direction is set to 1.2 times the obstacle size, forming an asymmetric distribution. Multiplying the basic risk coefficient by a two-dimensional Gaussian potential field yields a probabilistic risk potential field, ensuring that the gradient magnitude of the risk potential field in the area corresponding to the staff is the largest among all obstacles.

[0028] Specifically, suppose the AGV's sensing system identifies a worker with a width of 0.6 meters and a fixed textile machine with a width of 2 meters in front of it. A basic risk coefficient is assigned based on category: the worker is assigned the highest risk coefficient of 1.0, while the textile machine is assigned a lower risk coefficient of 0.4. The risk coefficient determines the overall strength of the respective risk potential field, reflecting the prior judgment of the danger level of different obstacles.

[0029] A two-dimensional Gaussian distribution is generated for each obstacle to simulate the diffusion of risk. Taking a worker as an example, if the AGV approaches along the positive X-axis, the standard deviation of the Gaussian distribution in the X-axis direction is set to 0.9 meters, while the standard deviation in the Y-axis direction is 0.72 meters. This setting makes the risk potential field extend further along the AGV's path, while converging relatively laterally. Multiplying the asymmetric Gaussian distribution function by the base risk coefficient of 1.0 constitutes the risk potential field for the worker. The same operation is applied to textile equipment, but due to the different base risk coefficient and size, the generated risk potential field has a lower strength and a wider range.

[0030] S2, for a pair of AGVs with collision risk, calculate an anisotropic velocity barrier with nonlinearly adjusted boundary based on the passage rules of the functional area of ​​the workshop and their respective task priorities.

[0031] Specifically, determine whether there is a risk of collision between a pair of AGVs, denoted as A and B. Compared to speed Whether the AGV falls within a circular velocity obstacle VO with the line connecting the geometric centers of the two vehicles as its axis and its radius being the sum of the radii of the two vehicles. If a risk exists, the shape of the VO is adjusted according to the functional area where the AGV is located, such as a narrow passage, intersection, or loading / unloading area, using preset traffic rules. For example, in a narrow passage where right-hand traffic is permitted, the left half of the circular velocity obstacle VO is nonlinearly expanded outward, while the right half is contracted inward, forming a teardrop-shaped anisotropic velocity obstacle AVO, making the cost of choosing the speed solution to avoid from the left higher. Simultaneously, based on task priority, if... The priority is higher than ,but The perceived size of AVO will be scaled up proportionally, forcing By taking avoidance actions, the AGV path for high-priority tasks is made smoother.

[0032] In an optional embodiment, calculating an anisotropic velocity barrier with a boundary nonlinearly adjusted includes: Based on the relative position and size of a pair of AGVs, calculate a standard anisotropic velocity barrier represented as an ellipse; If the functional area where the AGV is located is the main channel, then the length of the major half-axis of the standard anisotropic velocity barrier in the direction of AGV movement is multiplied by an adjustment factor of 1.2, and the length of the minor half-axis in the vertical direction is multiplied by an adjustment factor of 0.8. If the task priority of the AGV is lower than that of the other AGV, the Sigmoid function is used to perform nonlinear deformation on the standard anisotropic velocity barrier boundary, so that it bulges outward in the velocity direction of the other AGV. The amount of deformation is positively correlated with the difference in task priority.

[0033] Specifically, when Detected that it is about to be with When a collision risk occurs, it calculates a basic elliptical velocity barrier area. It assumes the elliptical velocity barrier is in... The semi-axis length in the forward direction is 2.0 meters per second, and the semi-axis length in the vertical direction is 1.0 meter per second. If at this time... While driving on the factory's main road, the application scenario adjustment coefficient multiplies the forward half-axis length by 1.2 to 2.4 meters per second, and multiplies the vertical half-axis length by 0.8 to 0.8 meters per second. This adjustment makes the prohibited speed zone longer and narrower, encouraging... Conflicts are resolved by adjusting longitudinal speed rather than making large lateral maneuvers, adapting to the traffic characteristics of the main corridor.

[0034] Assumption The task priority is level 2, while The task is assigned an emergency priority level of 5, with a priority difference of 3. This difference of 3 serves as input to the Sigmoid function, which outputs a large value close to 1. This value is used to scale a deformed vector, the direction of which is... The velocity direction. Applying the deformation vector to the adjusted elliptical boundary will cause the velocity barrier region to face the direction of the velocity. It bulges outwards. Nonlinear deformation greatly increases... Choice and The cost of speed related to the direction of travel forces lower priority vehicles to... Adopt a conservative slowdown or steering strategy, prioritizing proactive measures. Give Way.

[0035] S3 calculates an unbalanced reciprocal responsibility allocation coefficient based on the remaining power and task time margin of each pair of AGVs, and uses the reciprocal responsibility allocation coefficient to offset the anisotropic speed barrier to generate a hybrid reciprocal speed barrier.

[0036] Specifically, for those with a risk of collision and Calculate their state weights w respectively. For example, state weights ,in This is the normalized remaining battery power. It is the normalized task time margin. and This refers to the weighting coefficient. This formula means that the lower the battery level and the more urgent the situation, the larger the state weight w value. Calculation The coefficient of liability to be borne Similarly The responsibility coefficient is Therefore, the AGV with a more urgent situation will be assigned a smaller responsibility coefficient. This coefficient is used to offset the anisotropic speed obstacle (AVO) calculated above, generating a hybrid reciprocal speed obstacle (HRVO). The vertex of the AVO, i.e., the vertex of the speed obstacle cone, is moved from the origin along the line connecting the current speeds of the two vehicles; the new vertex position is... This results in an unbalanced distribution of collision avoidance responsibility between the two parties.

[0037] In an optional embodiment, calculating a non-equilibrium reciprocal responsibility allocation coefficient includes: For a pair of AGVs Responsibility allocation factor The calculation formula is:

[0038] in, for The remaining battery percentage, for Task time margin in seconds; calculate Responsibility allocation factor ; Will By normalizing the responsibility allocation factors, we obtain the reciprocal responsibility allocation coefficient. :

[0039] Using the reciprocal responsibility allocation coefficient Migrate anisotropic velocity barriers to generate Hybrid reciprocity speed barrier.

[0040] Specifically, let's assume two AGVs, and There is a risk of collision. The remaining battery level is 80%, and the task time margin is 60 seconds. The remaining battery power is 50%, and the task time margin is 240 seconds. Calculate the responsibility allocation factor for both. Responsibility allocation factor It is 0.5. Responsibility allocation factor It is 0.65.

[0041] calculate Reciprocal responsibility allocation coefficient , It is approximately equal to 0.43. They will bear 43% of the responsibility for obstacle avoidance, while Then they will bear the remaining 57%. Reciprocal responsibility allocation coefficient. Used to adjust The speed obstacle zone. What was originally a conflict that needed to be resolved with both sides bearing half the responsibility has now become... Once it completes 43% of the avoidance maneuvers, the speed obstacle zone it faces will shrink accordingly, allowing it to choose a speed closer to the original path.

[0042] S4. Under the kinematic constraints of the AGV, search for an optimal speed that minimizes the weighted cost function in the allowed speed space outside the hybrid reciprocal speed obstacle. The weighted cost function is a weighted combination of the expected speed deviation and the integral of the predicted trajectory risk potential field, where the weights are adjusted according to the current task priority of the AGV.

[0043] Specifically, define the speed constraint set for the AGV. This set contains all speeds that satisfy both the maximum speed limit and the maximum acceleration / deceleration limit. The allowed speed space is... The difference between this space and all HRVO regions. Within this space, the optimal velocity is searched using numerical optimization methods. The goal of optimization is to minimize the cost function. The first item is the new speed versus the expected speed. The square of the Euclidean distance, The first term is the maximum permissible speed pointing to the task target point. The second term is the speed within a short prediction time domain, assuming the AGV moves at... On the predicted trajectory p(t) formed by uniform speed travel, path integration is performed over the previously constructed environmental risk potential field U. Weights and Adjust based on task priority: For high-priority tasks, increase... Reduce path deviation to ensure efficiency; increase efficiency for low-priority tasks or in high-risk areas. This causes AGVs to tend to choose a steady speed away from potentially dangerous areas, ensuring safety. The searched [items / vehicles] make [the AGVs]... Minimum speed This refers to the execution speed of the next control cycle.

[0044] In an optional embodiment, the weighted cost function is a weighted combination of the expected velocity deviation and the integral of the predicted trajectory risk potential field. Defined as:

[0045] Where v is the candidate velocity, For the desired speed, This represents the maximum speed of the AGV. The predicted position of candidate velocity v at time t Risk potential value on the field, For prediction in the time domain, For efficiency weighting, For safety weights.

[0046] Specifically, when an AGV performs an urgent material delivery task, efficiency weighting will be applied. Set to 0.7, safety weight Let's set the value to 0.3. Assume the desired speed is 1.5 meters per second, and the maximum speed is 2 meters per second. There are two candidate speeds: Speed ​​A is 1.4 meters per second, but the predicted trajectory will glide across the edge of a risky potential field of a material stack, with a calculated average trajectory risk of 0.5; Speed ​​B is 0.8 meters per second, which completely avoids the risky area, with an average trajectory risk of 0.05. For Speed ​​A, the cost mainly comes from the risk term, approximately 0.15, while the speed deviation term is very small. For Speed ​​B, the cost mainly comes from the speed deviation term, approximately 0.7 multiplied by a large deviation value, much greater than 0.15. Therefore, Speed ​​A is chosen to prioritize task timeliness.

[0047] When the AGV switches to performing regular replenishment tasks, the weight will be adjusted to... It equals 0.4. It equals 0.6. Under the same conditions, the cost calculation for speed A becomes 0.3, and the penalty for the risk term increases. However, for speed B, the penalty for the speed deviation term is... The value of speed B is likely to be lower than that of speed A after recalculation. Therefore, in this task mode, choosing the slower but safer speed B reflects a safety-first strategy.

[0048] In an optional embodiment, under the kinematic constraints of the AGV, searching for an optimal speed that minimizes the weighted cost function within the allowed speed space outside the hybrid reciprocal speed barrier includes: A series of candidate velocities are generated within the allowable velocity space discretized with a velocity resolution of 0.1 meters per second and an angular resolution of 5°. For each candidate velocity, extrapolate the trajectory 3 seconds ahead, i.e., predict the time domain. The trajectory is sampled every 0.2 seconds, for a total of 15 sampling points. The integral of the trajectory risk potential field is approximately predicted by calculating the weighted sum of the risk potential field values ​​of the sampling points. Calculate the weighted cost function value for each candidate velocity by combining the expected velocity deviation; The candidate speed that minimizes the weighted cost function value is selected as the optimal speed.

[0049] Specifically, if the AGV's maximum linear speed is 1.5 meters per second, and the maximum turning angular velocity allows for a change in heading of up to ±20° within the next control cycle, then speed values ​​from 0 to 1.5 meters per second are generated in steps of 0.1 meters per second, and heading angles from -20° to +20° are generated in steps of 5°. All combinations of these two sets of values, such as a speed of 0.8 meters per second with a heading deflection of 10°, or a speed of 1.2 meters per second with a heading deflection of -5°, constitute a set containing dozens of candidate speeds.

[0050] Taking a candidate speed of 1.0 m / s with a constant heading as an example, the predicted path of the AGV over the next 3 seconds is calculated. Along this predicted path, starting from 0.2 seconds, a point is sampled every 0.2 seconds until 3.0 seconds, resulting in the position coordinates of 15 sampling points. The risk potential field values ​​of these 15 points on the global risk map are queried and averaged to obtain an approximate risk potential field integral value for the trajectory. This risk potential field value is multiplied by the task weight, and then the cost calculated based on the deviation between the current speed and the desired speed is added to obtain the total cost of the candidate speed. After all candidate speeds undergo the same calculation process, their total cost values ​​are compared, and the one with the lowest cost value, such as 0.9 m / s with a heading deflection of -5°, is sent as a control command to the AGV's underlying motion controller.

[0051] An embodiment of the dynamic planning system for material delivery paths in a textile workshop with multiple AGVs provided by this invention: The multi-AGV textile workshop material delivery path dynamic planning system includes a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned multi-AGV textile workshop material delivery path dynamic planning method is implemented.

[0052] The dynamic planning system for material delivery routes in a textile workshop with multiple AGVs also includes other components well known to those skilled in the art, such as communication interfaces. Their settings and functions are known in the art and will not be described in detail here.

[0053] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained by such a computer-readable medium.

[0054] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A dynamic planning method for material delivery paths in a textile workshop with multiple AGVs, characterized in that, Includes the following steps: S1: Acquire the position, speed, and environmental LiDAR and visual data of each AGV; use a semantic segmentation network to identify and classify obstacles in the environment such as textile equipment, raw material stacks, and workers. By integrating the location information of Lidar point clouds with the semantic category information of obstacles, an asymmetric probabilistic risk potential field is constructed for different categories of obstacles, and the risk potential field gradient of the area corresponding to the staff is ensured to be the highest. S2, For a pair of AGVs with collision risk, calculate an anisotropic velocity barrier with nonlinearly adjusted boundary according to the passage rules of the functional area of ​​the workshop and their respective task priorities; S3. Based on the remaining power and task time margin of each pair of AGVs, calculate an unbalanced reciprocal responsibility allocation coefficient, and use the reciprocal responsibility allocation coefficient to offset the anisotropic speed barrier to generate a hybrid reciprocal speed barrier. S4. Under the kinematic constraints of the AGV, search for an optimal speed that minimizes the weighted cost function in the allowed speed space outside the hybrid reciprocal speed barrier. The weighted cost function is a weighted combination of the expected speed deviation and the integral of the predicted trajectory risk potential field, where the weights are adjusted according to the current task priority of the AGV.

2. The method for dynamic planning of material delivery paths in a textile workshop with multiple AGVs according to claim 1, characterized in that, Semantic segmentation networks are used to identify and classify obstacles in the environment, such as textile equipment, raw material stacks, and workers, including: The 640×480 pixel RGB image captured by the AGV's front camera is input into the pre-trained DeepLabv3+ semantic segmentation model; The DeepLabv3+ semantic segmentation model classifies each pixel in an RGB image and outputs a semantic mask of the same size as the input image, where different pixel values ​​represent textile equipment, raw material stacks, workers, or passable ground areas. Based on the outlines and center points of various obstacle pixel clusters in the semantic mask, the bounding boxes and categories of obstacles are determined.

3. The method for dynamic planning of material delivery paths in a textile workshop with multiple AGVs according to claim 1, characterized in that, Constructing asymmetric probabilistic risk potential fields for different types of obstacles, including: Map the semantic categories of obstacles to the basic risk coefficient; A two-dimensional Gaussian potential field model is constructed with the geometric center of the obstacle as the mean. According to the movement direction of the AGV, the standard deviation of the two-dimensional Gaussian potential field in the AGV's forward direction is set to 1.5 times the obstacle size, and the standard deviation in the vertical direction is set to 1.2 times the obstacle size, forming an asymmetric distribution. Multiplying the basic risk coefficient by a two-dimensional Gaussian potential field yields a probabilistic risk potential field, ensuring that the gradient magnitude of the risk potential field in the area corresponding to the staff is the largest among all obstacles.

4. The method for dynamic planning of material delivery paths in a textile workshop with multiple AGVs according to claim 1, characterized in that, Calculate an anisotropic velocity barrier with a nonlinearly adjusted boundary, including: Based on the relative position and size of a pair of AGVs, calculate a standard anisotropic velocity barrier represented as an ellipse; If the functional area where the AGV is located is the main channel, then the length of the major half-axis of the standard anisotropic velocity barrier in the direction of AGV movement is multiplied by an adjustment factor of 1.2, and the length of the minor half-axis in the vertical direction is multiplied by an adjustment factor of 0.

8. If the task priority of the AGV is lower than that of the other AGV, the Sigmoid function is used to perform nonlinear deformation on the standard anisotropic velocity barrier boundary, so that it bulges outward in the velocity direction of the other AGV. The amount of deformation is positively correlated with the difference in task priority.

5. The method for dynamic planning of material delivery paths in a textile workshop with multiple AGVs according to claim 1, characterized in that, Calculate a non-equilibrium reciprocal responsibility allocation coefficient, including: For a pair of AGVs Responsibility allocation factor The calculation formula is: in, for The remaining battery percentage, for Task time margin in seconds; calculate Responsibility allocation factor ; Will By normalizing the responsibility allocation factors, we obtain the reciprocal responsibility allocation coefficient. : Using the reciprocal responsibility allocation coefficient Migrate anisotropic velocity barriers to generate Hybrid reciprocal speed barrier.

6. The method for dynamic planning of material delivery paths in a textile workshop with multiple AGVs according to claim 1, characterized in that, The weighted cost function is a weighted combination of the integral of the expected velocity deviation and the risk potential field of the predicted trajectory. Defined as: Where v is the candidate velocity, For the desired speed, This represents the maximum speed of the AGV. The predicted position of candidate velocity v at time t Risk potential value on the field, To predict the time domain, For efficiency weighting, For safety weights.

7. The method for dynamic planning of material delivery paths in a textile workshop with multiple AGVs according to claim 1, characterized in that, Under the kinematic constraints of the AGV, search for an optimal speed that minimizes the weighted cost function within the allowable speed space outside the hybrid reciprocal speed barrier, including: A series of candidate velocities are generated within the allowable velocity space discretized with a velocity resolution of 0.1 meters per second and an angular resolution of 5°. For each candidate velocity, extrapolate the trajectory 3 seconds ahead, i.e., predict the time domain. The trajectory is sampled every 0.2 seconds, for a total of 15 sampling points. The integral of the trajectory risk potential field is approximately predicted by calculating the weighted sum of the risk potential field values ​​of the sampling points. Calculate the weighted cost function value for each candidate velocity by combining the expected velocity deviation; The candidate speed that minimizes the weighted cost function value is selected as the optimal speed.

8. The method for dynamic planning of material delivery paths in a textile workshop with multiple AGVs according to claim 1, characterized in that, In S1, the semantic segmentation model is the YOLOv8-seg model.

9. The method for dynamic planning of material delivery paths in a textile workshop with multiple AGVs according to claim 1, characterized in that, In S1, the global position of the AGV is obtained through real-time differential GPS or UWB module, and the current speed is obtained through fusion calculation of wheel odometer and inertial measurement unit.

10. A dynamic planning system for material delivery paths in a textile workshop using multiple AGVs, characterized in that, It includes a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the dynamic planning method for material delivery path in a textile workshop with multiple AGVs as described in any one of claims 1-9 is implemented.

Images