AGV intelligent control method and system

Abstract

The application provides an AGV intelligent control method and system, and belongs to the field of intelligent control. An environment grid map is constructed based on the pose of the AGV and sensor data, the g(s) value of a node s is defined as the actual cost from the starting point to the node s, the rhs(s) value of the node s is the first estimated cost from the node s to the target point, and it is determined whether the node s is an inconsistent node. According to the relative position of the inconsistent node and the AGV and the type of the environment change, the update priority of the inconsistent node is calculated. The inconsistent node is added to a priority queue, the g(s) value map is updated, a new driving path from the current position of the AGV to the target point is generated, and the new driving path is executed. The application can intelligently screen and sort the re-planning tasks, concentrate the computing resources on processing the most critical environment changes, improve the decision efficiency and response timeliness in a complex environment, and enhance the safety, smoothness and overall traffic efficiency of the AGV operation.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of intelligent control, and particularly relates to an AGV intelligent control method and system. BACKGROUND

[0002] As the core equipment in modern intelligent logistics and automated production lines, the path planning and intelligent control technology of an automatic guided vehicle (AGV) is the key to determine its running efficiency, stability and safety. In a complex and dynamically changing factory or warehouse environment, the AGV needs to be able to perceive the environmental changes in real time, and quickly and accurately plan or re-plan an optimal or sub-optimal path from the current position to the target point. Although the traditional path planning algorithm, such as the A* algorithm, performs well in a static environment, when there are dynamic obstacles or changes in path information in the environment, it needs to be recalculated globally, which is computationally intensive and has poor real-time performance, and is difficult to meet the needs of efficient operation of the AGV.

[0003] In order to solve the problem of path planning in a dynamic environment, the D*Lite algorithm emerged as the times require. This algorithm is an incremental heuristic search algorithm, which maintains path information through backward search (from the target point to the starting point). When the environment changes, only the node information of the affected area needs to be updated, without the need to recalculate the entire path, greatly improving the efficiency of re-planning. However, the existing AGV control method based on D*Lite still has many limitations in practical application.

[0004] First, in a complex factory or warehouse environment, AGV needs to avoid static and dynamic obstacles flexibly while ensuring transportation efficiency. Classic path planning algorithms such as A* and Dijkstra perform well in static known environments and can search for the optimal path. In actual work scenarios, the environment is dynamically changing, full of other AGVs, workers and temporary stacked goods, which requires the path planning algorithm to have the ability of rapid response and re-planning. Compared with A* and other global re-planning algorithms, when the path is blocked by obstacles in the environment, the D*Lite algorithm does not need to recalculate the entire path, but only updates the generation value of the affected area, thereby improving the efficiency of re-planning. However, the re-planning trigger mechanism is sensitive and lacks priority differentiation, and any minor environmental change may trigger path update, leading to excessive system calculation burden when the environment is frequently disturbed; on the contrary, if the trigger threshold is increased to reduce the calculation frequency, the response to critical obstacles may be delayed. The calculation model of path cost is too simplified, usually only considering distance and static obstacles, and failing to fully integrate the kinematic characteristics of AGV itself, resulting in the planned path may have too large corners, discontinuous curvature and other problems, making AGV unable to execute smoothly and efficiently, and unable to quantify some non-obvious traffic risks such as historical congestion areas, which limits the traffic efficiency and safety of AGV in complex human-machine coexistence scenarios. How to achieve optimal control of AGV in complex dynamic environment is a problem to be solved in the field. SUMMARY

[0005] The purpose of the present application is to provide an AGV intelligent control method and system for solving the problem that the existing AGV is difficult to achieve optimal control in a complex dynamic environment, comprising the following steps:

[0006] An environment grid map is constructed based on the pose of the AGV and sensor data, the g(s) value of a node s is defined as the actual cost from the starting point to the node s, the rhs(s) value of the node s is the first-line estimated cost from the node s to the target point, and the g(s) value and the rhs(s) value of each node s are initialized; when the rhs(s) value of any node s is updated due to environmental changes, if the absolute value of the difference between the updated rhs(s) value and the g(s) value exceeds a threshold value adjusted according to the local environmental complexity and the AGV speed, the node s is determined as a non-consistent node;

[0007] According to the relative position of the non-consistent node and the AGV and the type of environmental change, the update priority of the non-consistent node is calculated; the non-consistent node is added to a priority queue, and the sorting key of the non-consistent node in the queue is calculated or updated according to the g(s) value, the rhs(s) value, the heuristic value and the update priority;

[0008] Iterate to take out the node u with the minimum key from the priority queue for expansion, update the successor node of the node u with the minimum key The value, wherein the transition cost between nodes is calculated according to the AGV kinematics constraint, the dynamic obstacle space-time trajectory prediction result, and the potential passing risk evaluated based on the machine learning model ; based on the updated g(s) value map, a new driving path from the current position of the AGV to the target point is generated and executed.

[0009] Optionally, the threshold value adjusted according to the local environment complexity and the AGV speed includes:

[0010] Taking the node s to be judged as the center, an N×N grid neighborhood is selected, and the number of obstacle grids in the neighborhood is counted , the local environment complexity is calculated;

[0011] The current linear speed of the AGV is obtained ;

[0012] The threshold value T is calculated by the formula , wherein is a basic threshold value, is a complexity weighting coefficient, is a speed weighting coefficient, and N takes a value of 5.

[0013] Optionally, the update priority of the non-consistent node is calculated according to the relative position of the non-consistent node and the AGV and the type of environmental change, including:

[0014] The Euclidean distance between the non-consistent node s and the current position of the AGV is calculated ;

[0015] If the node s is changed from a non-obstacle to an obstacle, then is a high priority value; if the node s is changed from an obstacle to a non-obstacle, then is a low priority value;

[0016] The update priority P is calculated by the formula , wherein is a distance weight coefficient, is a change type weight coefficient, and ε is a small normal number to avoid the denominator being zero when the distance is zero.

[0017] Optionally, the sorting key in the queue is calculated or updated according to the g(s) value, the rhs(s) value, the heuristic value, and the update priority, including:

[0018] The sorting key is a two-dimensional vector K=[k1, k2], and the calculation method is:

[0019] ;

[0020] ;

[0021] wherein, is the heuristic value from the current position of the AGV to node s, and P(s) is the updated priority calculated.

[0022] Optionally, the transfer cost between nodes is calculated according to the kinematics constraint of the AGV, the spatiotemporal trajectory prediction result of the dynamic obstacle, and the potential passing risk evaluated based on the machine learning model , and specifically:

[0023] The transfer cost between nodes is calculated by adding the basic geometric cost, the turning cost, the dynamic obstacle avoidance cost, and the potential risk cost: ;

[0024] wherein, the basic geometric cost is the geometric distance from node u to node ;

[0025] The turning cost is calculated by: obtaining the parent node of node u, calculating the angle θ between the vector and the vector , wherein is a turning penalty coefficient;

[0026] The dynamic obstacle avoidance cost is calculated by: predicting the position of the dynamic obstacle at the estimated time when the AGV reaches node using a Kalman filter, calculating the distance between node and the position at the estimated time, wherein A is a maximum penalty value, is a safety influence range parameter;

[0027] The potential risk cost is calculated by: inputting the feature vector of node into a pre-trained machine learning model to obtain a passing risk score , and then calculating ; wherein the feature vector includes the node a history traversal frequency of the location, a distance to a nearest static obstacle, and a history of people traffic in the area, a risk weight coefficient.

[0028] The application further provides an AGV intelligent control system, comprising the following modules:

[0029] a non-consistent node determination module, configured to construct an environment grid map based on a pose of the AGV and sensor data, define a g(s) value of a node s as an actual cost from a starting point to the node s, a rhs(s) value of the node s as a preceding estimated cost of the node s, and initialize the g(s) value and the rhs(s) value of each node s; when the rhs value of any node s is updated due to an environment change, if an absolute value of a difference between the updated rhs(s) value and the g(s) value exceeds a threshold value adjusted according to a local environment complexity and a speed of the AGV, the node s is determined as a non-consistent node;

[0030] an updating module, configured to calculate an updating priority of the non-consistent node according to a relative position of the non-consistent node and the AGV and a type of the environment change, add the non-consistent node into a priority queue, and calculate or update a sorting key of the non-consistent node in the queue according to the g(s) value, the rhs(s) value, a heuristic value and the updating priority;

[0031] a path determination module, configured to iteratively take out a node u with a minimum key from the priority queue for expansion, update a cost value of a successor node of the node u with the minimum key, and determine a new driving path from a current position of the AGV to a target point based on an updated g(s) value map. a cost value of transition between nodes is obtained according to kinematic constraints of the AGV, a spatiotemporal trajectory prediction result of a dynamic obstacle, and a potential passing risk evaluated based on a machine learning model.

[0032] Optionally, the threshold value adjusted according to the local environment complexity and the speed of the AGV comprises:

[0033] a local environment complexity is calculated by taking a N*N grid neighborhood centered on the node s, counting a number of obstacle grids in the neighborhood , and calculating the local environment complexity ;

[0034] a current linear speed of the AGV is obtained ;

[0035] the threshold value T is calculated by a formula , wherein a basic threshold value is a complexity weighting coefficient is ​is the speed weighting coefficient, and N is 5.

[0036] Optionally, the updating priority of the inconsistent node is calculated according to the relative position of the inconsistent node and the AGV and the type of environmental change, and the updating priority of the inconsistent node is calculated according to the relative position of the inconsistent node and the AGV and the type of environmental change.

[0037] The Euclidean distance between the inconsistent node s and the current position of the AGV is calculated.

[0038] If the node s is changed from a non-obstacle to an obstacle, then is a high priority value; if the node s is changed from an obstacle to a non-obstacle, then is a low priority value.

[0039] The updating priority P is calculated by the formula , wherein is a distance weight coefficient, is a change type weight coefficient, and ε is a small normal number to avoid the denominator being zero when the distance is zero.

[0040] Optionally, the sorting key in the queue is calculated or updated according to the value of g(s), the value of rhs(s), the heuristic value, and the updating priority.

[0041] The sorting key is a two-dimensional vector K = [k1, k2], and the calculation method is:

[0042]

[0043]

[0044] , wherein is the heuristic value from the current position of the AGV to the node s, and P(s) is the calculated updating priority.

[0045] Optionally, the transfer cost between nodes is calculated according to the kinematic constraints of the AGV, the spatiotemporal trajectory prediction results of dynamic obstacles, and the potential passing risk evaluated based on a machine learning model. , specifically:

[0046] The transfer cost between nodes is calculated by adding the basic geometric cost, the turning cost, the dynamic obstacle avoidance cost, and the potential risk cost.

[0047] , wherein the basic geometric cost is the geometric distance from the node u to the node .

[0048] The turning cost​​​​ By the following way: obtaining the parent node of node u , calculating the vector The angle θ between the vector , , wherein is the turning penalty coefficient;

[0049] The dynamic obstacle avoidance cost By the following way: using a Kalman filter to predict the position of the dynamic obstacle at the estimated time when the AGV reaches node , calculating the distance between node and the position at the estimated time , , wherein A is the maximum penalty value, is the safety influence range parameter;

[0050] The potential risk cost By the following way: inputting the feature vector of node into a pre-trained machine learning model to obtain a passing risk score , and then calculating ; wherein the feature vector includes the historical traversal frequency of node , the distance from the nearest static obstacle, and the historical data of the number of people in the area, is the risk weight coefficient.

[0051] The application sets a re-planning trigger threshold that can reflect the complexity of the local environment and the running speed of the AGV, and determines the priority order of updating nodes in combination with the type of environmental change and the relative position, so that the algorithm can intelligently filter and sort the re-planning tasks, and concentrate computing resources on processing the most critical environmental changes, reducing unnecessary computing overhead, and improving decision-making efficiency and response timeliness in complex environments. In addition, when calculating the path cost, the kinematic constraints of the AGV, the trajectory prediction results of the moving obstacles in the environment, and the potential passing risk evaluated based on the machine learning model are combined, which makes the planned driving path not only easier for the AGV to execute smoothly, but also can avoid future possible collisions and bypass historical congestion areas and other non-explicit risk sources, thereby enhancing the safety, smoothness and overall passing efficiency of the AGV operation. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 is the flowchart of the first embodiment;

[0053] Figure 2 is a schematic diagram of AGV path planning;

[0054] Figure 3This is a schematic diagram of dynamic threshold adjustment;

[0055] Figure 4 This is a diagram illustrating the update priority. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with relevant laws, regulations, and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0057] In the first embodiment, the present invention provides an intelligent control method for AGVs, such as... Figure 1 As shown, it includes the following steps:

[0058] S1. Construct an environmental grid map based on the AGV's pose and sensor data. Define the g(s) value of node s as the actual cost from the starting point to node s, and the rhs(s) value of node s as the estimated cost from node s to the target point. Initialize the g(s) and rhs(s) values ​​of each node s. When environmental changes cause the rhs(s) value of any node s to be updated, if the absolute value of the difference between the updated rhs(s) value and the g(s) value exceeds a threshold that is adjusted according to the local environmental complexity and AGV speed, then node s is determined to be an inconsistent node.

[0059] Environmental point cloud data is acquired using LiDAR, and a two-dimensional environmental grid map is generated using a real-time localization and mapping (SLAM) algorithm such as GMAPping or Cartographer. Each grid cell is marked as occupied, vacant, or unknown. During initialization, the g(s) and rhs(s) values ​​of all nodes s in the map are set to infinity; only the target point... rhs( Set the value to 0 and add it to the priority queue.

[0060] By continuously comparing the latest sensor data with the grid map, when it is found that the original free grid is occupied by obstacles or the original occupied grid becomes free, the transition cost between the corresponding nodes changes, thereby causing the rhs(s) value of the adjacent nodes to be updated. A dynamic threshold is calculated, for example, the threshold is equal to a basic constant divided by the weighted sum of the local environment complexity and the AGV speed. Among them, the local environment complexity can be quantified as the density of obstacle grids within a certain radius around the AGV, and the AGV speed is the current linear speed of the AGV. When the absolute value of the difference between the g(s) value and the rhs(s) value of the node s is greater than the dynamic threshold, the node s is identified as an inconsistent node.

[0061] In an optional embodiment, the threshold adjusted with the local environment complexity and the AGV speed includes:

[0062] Taking the node s to be determined as the center, an N×N grid neighborhood is selected, and the number of obstacle grids in the neighborhood is counted , the local environment complexity is calculated ;

[0063] The current linear speed of the AGV is obtained ;

[0064] The threshold T is calculated by the formula , wherein is a basic threshold, is a complexity weighting coefficient, is a speed weighting coefficient, and N is 5.

[0065] The dynamic threshold adjustment mechanism enables the sensitivity of the AGV to environmental changes to be adaptive. When the AGV is in a complex environment or is driving at high speed, the path planning needs to respond more quickly to small changes in the cost map to ensure safety. For example, in a 5×5 neighborhood centered on the node to be determined, if 10 obstacle grids are detected, the local environment complexity is 0.4; at this time, if the AGV speed is 1.5 meters per second, the basic threshold is set to 10, the complexity weighting coefficient is 2, and the speed weighting coefficient is 1, then the calculated threshold T is about 3.03. A lower threshold means that even if there is a small difference between the g(s) value and the rhs(s) value, the node update will be triggered, ensuring high vigilance in dangerous or complex situations.

[0066] On the contrary, when the AGV is driving at low speed in an open area, the update frequency can be appropriately reduced to save computing resources. For example, if there are only 2 obstacle grids in the same neighborhood, the complexity is 0.08, and the AGV speed is reduced to 0.5 meters per second, using the same parameters, the calculated threshold T will rise to about 6.02, as Figure 3The higher threshold allows a greater tolerance between the g value and the rhs value without triggering an update, thereby avoiding unnecessary calculations when the environment is stable and the risk is low, improving the overall efficiency of the algorithm.

[0067] S2, according to the relative position of the non-consistent node and the AGV and the type of environmental change, calculates the update priority of the non-consistent node; adds the non-consistent node to the priority queue, and calculates or updates the sorting key of the non-consistent node in the queue according to the g(s) value, the rhs(s) value, the heuristic value, and the update priority;

[0068] The update priority P of the non-consistent node s is calculated by a function that integrates the distance and the type of change. For example, P is equal to a weight coefficient divided by the Euclidean distance between the node s and the current position of the AGV, plus a weight coefficient multiplied by a change type value. Among them, the change type value is preset, such as a newly appeared obstacle, i.e. the grid changes from free to occupied, the value is a higher value 1.0, and the known obstacle disappears, i.e. the grid changes from occupied to free, the value is a lower value 0.2. Insert the non-consistent node s into the priority queue U or update its sorting key in the queue. The key is a two-dimensional or three-dimensional vector used for queue sorting. For example, the first dimension of the key is min(g(s), rhs(s)) plus the heuristic cost h from the current AGV position to the node s, and the heuristic cost h preferably uses Manhattan distance calculation; the second dimension of the key is the update priority P calculated in the previous step, and the smaller the value, the higher the priority in sorting; the third dimension can be set as min(g(s), rhs(s)) as the final decisive standard.

[0069] In an optional embodiment, the calculation of the update priority P of the non-consistent node according to the relative position of the non-consistent node and the AGV and the type of environmental change comprises:

[0070] calculating the Euclidean distance between the non-consistent node s and the current position of the AGV ;

[0071] If the node s is changed from a non-obstacle to an obstacle, then is a high priority value; if the node s is changed from an obstacle to a non-obstacle, then is a low priority value;

[0072] The update priority P is calculated by the formula , wherein, is the distance weight coefficient, is the change type weight coefficient, and ε is a small normal number to avoid the denominator being zero when the distance is zero.

[0073] The update priority calculation method ensures that the system prioritizes environmental changes that have the greatest impact on AGV operation. Update priority is determined by two core factors: the distance between the change point and the AGV, and the nature of the change; the closer the distance, the more direct the potential impact, and therefore, the higher the priority. Simultaneously, newly appearing obstacles pose a greater threat to the current path than disappeared obstacles.

[0074] For example, assuming distance weighting coefficient The weighting coefficient for the type of change is 100. The value is 50. When a new obstacle appears within 2 meters of the AGV, it is assigned a high change type value, such as 1.0. Its update priority P is calculated to be approximately 99.75. Conversely, if an old obstacle disappears at a distance of 10 meters, it is assigned a low change type value, such as 0.1, and its priority P is approximately 14.99. Figure 4 As shown. In this way, newly added obstacles in the vicinity are dealt with first, because they are directly related to obstacle avoidance safety, while the path optimization opportunities represented by obstacles that disappear in the distance are given secondary importance.

[0075] In an optional embodiment, calculating or updating the sorting key in the queue based on the g(s) value, rhs(s) value, heuristic value, and update priority includes:

[0076] The sorting key is a two-dimensional vector K=[k1,k2], and the calculation method is as follows:

[0077] ;

[0078] ;

[0079] in, To start from the current position of the AGV The heuristic value to node s, P(s) is the calculated update priority.

[0080] Two-dimensional sorting keys are used to optimize the processing order of nodes in a priority queue, making path replanning both efficient and timely. The primary key k1 functions similarly to... The F value in the algorithm combines the known minimum cost from the starting point to the current node and the heuristic estimated cost from the current node to the target point, aiming to guide the search towards the optimal path. When the k1 values of multiple nodes are the same, the secondary key k2 plays a decisive role by introducing a negative update priority P, taking the urgency of environmental changes as the second sorting standard. For example, there are two nodes sA and sB to be processed, both with k1 values of 15, causing a sorting conflict. If node sA is an inconsistent node generated by the disappearance of a distant obstacle, its update priority P is lower, such as 20, and its k2 value is -20. If node sB is generated by the addition of a new obstacle nearby, its update priority P is extremely high, such as 100, and its k2 value is -100. During sorting, since -100 is less than -20, node sB will be taken out of the queue and processed first, ensuring that in the case of similar path cost values, the areas that have the greatest impact on AGV safety can be updated first. In an alternative embodiment, in order to give priority a greater role, the update priority P is integrated into the calculation of k1. Alternatively, or .

[0081] S3, iteratively take out the node u with the smallest key from the priority queue for expansion, update the values of the successor nodes of the node u with the smallest key , wherein the transition cost between nodes is calculated based on AGV kinematics constraints, dynamic obstacle space-time trajectory prediction results, and potential passing risk evaluation based on machine learning models ; based on the updated g(s) value map, generate and execute a new driving path from the current position of the AGV to the target point.

[0082] Take out the node u with the smallest key from the priority queue U, and update the rhs value of all successor nodes of u according to the formula , and the transition cost is a composite cost whose value is obtained by weighted summation of multiple parts. The first part is the kinematics cost calculated based on the Reeds-Shepp curve model, which is used to punish large-angle turning that does not meet the minimum turning radius of the AGV. The second part is the dynamic obstacle collision risk predicted based on the long short-term memory network (LSTM) model. If a dynamic obstacle is predicted to occupy node at a certain time in the future, the cost will increase significantly, Figure 2 ​The diagram illustrates dynamic obstacles under path planning. The third part assesses the potential risk cost based on a pre-trained Gradient Boosting Decision Tree (GBDT) model, which assigns higher risk costs to nodes in congested or accident-prone areas based on historical traffic data. In one embodiment, a new driving path is generated using an updated g-value map. In yet another alternative embodiment, once the replanning process, i.e., the expansion iteration of the priority queue, has stabilized, the path starts from the AGV's current node. Initially, a greedy strategy is used to find the next path point, that is, among all neighboring nodes, the one that makes the expression... The neighbor node with the smallest value As the next node on the path, this process is repeated until the target point is reached, thus forming a path composed of discrete nodes. Curve fitting algorithms such as B-spline interpolation are used to smooth this discrete path, generating a trajectory with continuous curvature, which is then executed by the AGV's underlying motion controller. Optionally, the path extraction is performed after the replanning process—that is, after expanding the nodes and updating the g-value.

[0083] In an optional embodiment, the transfer cost between nodes is calculated based on the AGV kinematic constraints, the spatiotemporal trajectory prediction results of dynamic obstacles, and the potential passage risks assessed based on a machine learning model. Specifically:

[0084] The transfer cost between nodes The calculation is performed by adding the basic geometric cost, steering cost, dynamic obstacle avoidance cost, and potential risk cost: ;

[0085] Wherein, the basic geometric cost For node u to node geometric distance;

[0086] The turning cost Calculate by obtaining the parent node of node u. Calculate vector with vector The included angle θ between them ,in This is the steering penalty coefficient;

[0087] The dynamic obstacle avoidance cost The calculation is performed as follows: A Kalman filter is used to predict dynamic obstacles as the AGV arrives. The position at the estimated time, calculate the node Distance from the position at the estimated time , Where A is the maximum penalty value. For parameters related to the safety impact range;

[0088] the potential risk cost by inputting the feature vector of the node into a pre-trained machine learning model to obtain a passing risk score , and then calculating; wherein the feature vector includes the historical traversal frequency of the node , the distance from the nearest static obstacle, and the historical data of the people flow in the area, is the risk weight coefficient.

[0089] The composite cost function aims to generate a path that is not only the shortest, but also smoother, safer and smarter in actual operation; it decomposes the cost of a movement into four aspects for comprehensive evaluation, making the AGV's decision-making closer to the needs of complex environments.

[0090] Taking a specific calculation example, assuming that the AGV plans to move from node u to the adjacent node, the geometric distance between the two points is 1 unit, if this movement constitutes a 90-degree sharp turn compared to the path from the parent node to u, and the turning penalty coefficient is 5, then a turning cost equal to 5 will be generated to punish the non-smooth movement. At the same time, if a Kalman filter predicts that a moving pedestrian will be at a distance of only 1.1 meters from the adjacent node when the AGV arrives at the adjacent node, and the maximum penalty value A is 10 and the safety influence range parameter σ is 1, then the dynamic obstacle avoidance cost will be calculated to be about 5.46, and the high cost will prompt the AGV to choose a path away from the pedestrian.

[0091] In addition, the method also introduces risk assessment based on historical data. If the target node is located near a doorway that is often passed by people and is very close to the wall, the pre-trained machine learning model outputs a risk score as high as 0.8 according to its historical traversal frequency, distance from obstacles, etc. If the risk weight coefficient is 3, the potential risk cost is 2.4. Adding all the costs, the total transition cost from u to the target node is 13.86. The comprehensive cost value enables the path planner to generate an optimized path that effectively avoids various known and potential risks.

[0092] In a second embodiment, the present application provides an AGV intelligent control system, comprising the following modules:

[0093] ​The non-consistent node determining module is configured to construct an environment grid map based on a pose of the AGV and sensor data, define a g(s) value of a node s as an actual cost from a starting point to the node s, a rhs(s) value of the node s as a first-estimate cost of the node s to a target point, and initialize the g(s) value and the rhs(s) value of each node s; when the rhs(s) value of any node s is updated due to an environment change, if an absolute value of a difference between the updated rhs(s) value and the g(s) value exceeds a threshold value adjusted according to a local environment complexity and a speed of the AGV, the node s is determined as a non-consistent node;

[0094] The updating module is configured to calculate an updating priority of the non-consistent node according to a relative position of the non-consistent node and the AGV and a type of the environment change, add the non-consistent node into a priority queue, and calculate or update a sorting key of the non-consistent node in the queue according to the g(s) value, the rhs(s) value, a heuristic value, and the updating priority.

[0095] The path determining module is configured to iteratively take out a node u with a minimum key from the priority queue for expansion, update the g(s) value of a successor node of the node u, and update the rhs(s) value of the node u. The transition cost between the nodes is calculated according to kinematic constraints of the AGV, a spatiotemporal trajectory prediction result of a dynamic obstacle, and a potential passing risk evaluated based on a machine learning model. Based on the updated g(s) value map, a new driving path from a current position of the AGV to the target point is generated and executed.

[0096] In an optional embodiment, the threshold value adjusted according to the local environment complexity and the speed of the AGV includes:

[0097] A N*N grid neighborhood is selected with the node s as a center, a number of obstacle grids in the neighborhood is counted, and a local environment complexity is calculated. A current linear speed of the AGV is obtained.

[0098] The threshold value T is calculated by a formula T = Tbase * (1 + Ccomplex * complexity) * (1 + Cspeed * speed), where Tbase is a basic threshold value, Ccomplex is a complexity weighting coefficient, Cspeed is a speed weighting coefficient, and N is 5.

[0099]

[0100] In an optional embodiment, the calculation of the updating priority of the non-consistent node according to the relative position of the non-consistent node and the AGV and the type of the environment change includes: ​​​​​​​​

[0101] calculating the Euclidean distance between the non-consistent node s and the current position of the AGV ;

[0102] if the node s is changed from a non-obstacle to an obstacle, then is a high priority value; if the node s is changed from an obstacle to a non-obstacle, then is a low priority value;

[0103] the updated priority P is calculated by the formula , wherein is a distance weight coefficient, is a change type weight coefficient, and ε is a small normal number to avoid the denominator being zero when the distance is zero.

[0104] In an optional embodiment, the sorting key in the queue is calculated or updated according to the value of g(s), the value of rhs(s), the heuristic value, and the updated priority, which includes:

[0105] the sorting key is a two-dimensional vector K = [k1, k2], and the calculation method is:

[0106] ;

[0107] ;

[0108] wherein, is the heuristic value from the current position of the AGV to the node s, and P(s) is the updated priority calculated.

[0109] In an optional embodiment, the transfer cost between nodes is calculated according to the kinematic constraints of the AGV, the spatiotemporal trajectory prediction results of dynamic obstacles, and the potential passing risk evaluated based on a machine learning model , specifically:

[0110] the transfer cost between nodes is calculated by adding the basic geometric cost, the turning cost, the dynamic obstacle avoidance cost, and the potential risk cost: ;

[0111] wherein, the basic geometric cost is the geometric distance from the node u to the node ;

[0112] the turning cost is calculated by: obtaining the parent node of the node u, calculating the included angle θ between the vector and the vector ; wherein is a turning penalty coefficient;

[0113] the dynamic obstacle avoidance cost is calculated by predicting the position of the dynamic obstacle at the estimated time when the AGV reaches the node using a Kalman filter, calculating the distance between the node and the position at the estimated time , wherein A is a maximum penalty value, is a safety influence range parameter;

[0114] the potential risk cost is calculated by inputting the feature vector of the node into a pre-trained machine learning model to obtain a passing risk score , and then calculating ; wherein the feature vector includes the historical traversal frequency of the node , the distance from the nearest static obstacle, and the historical data of the passenger flow in the area, is a risk weight coefficient.

[0115] From the above description of the embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software and the necessary general hardware platform. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in the various embodiments or some parts of the embodiments.

[0116] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiments, since it is basically similar to the method embodiments, it is described more simply, and the relevant parts can be referred to the part of the method embodiments. The above-described system and system embodiments are only illustrative, and the units described as separate components can be or can not be physically separated, and the components displayed as units can be or can not be physical units, i.e., they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0117] The above provides a method for providing commodity object information and an electronic device. The principles and implementation manners of the present application are described by using specific examples in the present application. The above description of the embodiments is only used to help understand the method of the present application and its core idea. Meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manners and application ranges will be changed. In summary, the content of the present application should not be understood as a limitation.

Claims

1. An AGV intelligent control method, characterized in that, The method comprises the following steps: constructing an environment grid map based on the pose of the AGV and sensor data, and initializing the g(s) value and the rhs(s) value of each node s; when the rhs value of any node s is updated due to environmental changes, if the absolute value of the difference between the rhs(s) value and the g(s) value exceeds a threshold value adjusted according to the local environmental complexity and the AGV speed, the node s is determined to be a non-consistent node; calculating the update priority of the non-consistent node according to the relative position of the non-consistent node and the AGV and the type of environmental changes; adding the non-consistent node to a priority queue, and calculating or updating the sorting key of the non-consistent node in the queue according to the g(s) value, the rhs(s) value, the heuristic value and the update priority; Iteratively retrieve the node u with the smallest key from the priority queue and expand it, then update the successor node of node u with the smallest key. of Value, where the transfer cost between nodes The calculation is based on the AGV's kinematic constraints, the spatiotemporal trajectory prediction results of dynamic obstacles, and the potential passage risks assessed based on the machine learning model; based on the updated g-value map, a new travel path from the AGV's current position to the target point is generated and executed; the method of calculating the update priority of the non-consistent node according to the relative position of the non-consistent node and the AGV and the type of environmental changes comprises: calculating the Euclidean distance between the non-consistent node s and the current position of the AGV ; If node s changes from non-obstacle to obstacle, then is a high priority value; if node s changes from obstacle to non-obstacle, then is a low priority value; The update priority P is calculated by the formula wherein is a distance weight coefficient, is a change type weight coefficient, and ε is a small positive number to avoid division by zero when the distance is zero. the method of calculating or updating the sorting key of the non-consistent node in the queue according to the g(s) value, the rhs(s) value, the heuristic value and the update priority comprises: the sorting key is a two-dimensional vector K=[k1, k2], and the calculation method is: ； ； wherein, is the heuristic value from the current position of the AGV to the node s, and P(s) is the calculated updated priority. The transfer cost between the nodes The calculation of the transfer cost between the nodes is based on the kinematic constraints of the AGV, the spatiotemporal trajectory prediction results of dynamic obstacles, and the potential traffic risks evaluated based on the machine learning model, and is specifically as follows: the inter-node transfer cost The inter-node transfer cost is calculated by adding the base geometry cost, the turning cost, the dynamic obstacle avoidance cost, and the potential risk cost: ; where the base geometry cost is the geometric distance from node u to node v; The turning cost Calculate by obtaining the parent node of node u. Calculate vector with vector The included angle θ between them ,in This is the steering penalty coefficient; The dynamic obstacle avoidance cost The calculation is performed as follows: A Kalman filter is used to predict dynamic obstacles as the AGV arrives. The position at the estimated time, calculate the node Distance from predicted location , Where A is the maximum penalty value. For parameters related to the safety impact range; The potential risk costs Calculated in the following way: Target node The feature vectors are input into a pre-trained machine learning model to obtain a passage risk score. And then through The calculation yields the following result; wherein the feature vector includes nodes. Historical traversal frequency, distance to the nearest static obstacle, and historical pedestrian traffic data for the area. This is the risk weighting coefficient.

2. The method of claim 1, wherein, the threshold value adjusted according to the local environmental complexity and the AGV speed comprises: With the node s to be judged as the center, an N×N grid neighborhood is selected, and the number of obstacle grids in the neighborhood is counted , the local environment complexity is calculated ; Obtain the current linear velocity of the AGV ; The threshold value T is calculated by the formula wherein Tbase is a base threshold value, Cw is a complexity weighting factor, Vw is a velocity weighting factor, and N has a value of 5.

3. An AGV intelligent control system, characterized in that, The method comprises the following steps: a non-consistent node determination module, configured to construct an environment grid map based on the pose of the AGV and sensor data, and initialize the g(s) value and the rhs(s) value of each node s; when the rhs value of any node s is updated due to environmental changes, if the absolute value of the difference between the rhs(s) value and the g(s) value exceeds a threshold value adjusted according to the local environmental complexity and the AGV speed, the node s is determined to be a non-consistent node; an update module, configured to calculate the update priority of the non-consistent node according to the relative position of the non-consistent node and the AGV and the type of environmental changes; add the non-consistent node to a priority queue, and calculate or update the sorting key of the non-consistent node in the queue according to the g(s) value, the rhs(s) value, the heuristic value and the update priority; The path determination module iteratively retrieves the node u with the smallest key from the priority queue for expansion and updates the successor node of node u with the smallest key. of Value, where the transfer cost between nodes The calculation is based on the AGV's kinematic constraints, the spatiotemporal trajectory prediction results of dynamic obstacles, and the potential passage risks assessed based on the machine learning model; based on the updated g-value map, a new travel path from the AGV's current position to the target point is generated and executed; the method of calculating the update priority of the non-consistent node according to the relative position of the non-consistent node and the AGV and the type of environmental changes comprises: calculating the Euclidean distance between the non-consistent node s and the current position of the AGV ; If node s changes from non-obstacle to obstacle, then is a high priority value; if node s changes from obstacle to non-obstacle, then is a low priority value; The update priority P is calculated by the formula wherein is a distance weight coefficient, is a change type weight coefficient, and ε is a small normal number to avoid the denominator being zero when the distance is zero. the method of calculating or updating the sorting key of the non-consistent node in the queue according to the g(s) value, the rhs(s) value, the heuristic value and the update priority comprises: the sorting key is a two-dimensional vector K=[k1, k2], and the calculation method is: ； ； wherein, is the heuristic value from the current position of the AGV to the node s, and P(s) is the calculated updated priority; The transfer cost between the nodes The calculation of the transfer cost between the nodes is based on the kinematic constraints of the AGV, the spatiotemporal trajectory prediction results of dynamic obstacles, and the potential traffic risks evaluated based on the machine learning model, and is specifically as follows: the inter-node transfer cost by adding the base geometry cost, the turning cost, the dynamic obstacle avoidance cost, and the potential risk cost: ; where the base geometry cost is the geometric distance from node u to node v; The turning cost Calculate by obtaining the parent node of node u. Calculate vector with vector The included angle θ between them ,in This is the steering penalty coefficient; The dynamic obstacle avoidance cost The calculation is performed as follows: A Kalman filter is used to predict dynamic obstacles as the AGV arrives. The position at the estimated time, calculate the node Distance from predicted location , Where A is the maximum penalty value. For parameters related to the safety impact range; The potential risk costs Calculated in the following way: Target node The feature vectors are input into a pre-trained machine learning model to obtain a passage risk score. And then through The calculation yields the following result; wherein the feature vector includes nodes. Historical traversal frequency, distance to the nearest static obstacle, and historical pedestrian traffic data for the area. This is the risk weighting coefficient.

4. The system of claim 3, wherein, the threshold value adjusted according to the local environmental complexity and the AGV speed comprises: With the node s to be judged as the center, an N×N grid neighborhood is selected, and the number of obstacle grids in the neighborhood is counted , the local environment complexity is calculated ; Obtain the current linear velocity of the AGV ; The threshold value T is calculated by the formula wherein Tbase is a base threshold value, Cw is a complexity weighting factor, Vw is a velocity weighting factor, and N has a value of 5.

Images