An anti-collision scheduling and emergency obstacle avoidance method for indoor AGV sorting vehicles
Through image edge computing and improved path planning algorithms, the probability of goods falling and the path of small car is adjusted, which solves the problems of conflicts and obstacles of small car in e-commerce logistics sorting, and improves the stability and efficiency of the sorting process.
Patent Information
- Application Number
- CN202210271037.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-18
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-03-18
AI Technical Summary
In the process of e-commerce logistics sorting, it is difficult for the existing technology to effectively avoid trolley operation conflicts and physical obstacles, resulting in insufficiency of sorting and insufficient prediction of cargo drop probability, affecting the stability of the sorting process.
Using an image edge calculation method, the surface image, surface depth image and road condition image of the cart loaded, calculate the probability of falling and dangerous radius of the cart. Combined with the improved A-Star algorithm and Hesitating star algorithm, adjust the cart in real time path, prevent conflicts and avoid obstacles.
It improves the stability and efficiency of the sorting process, ensures that the trolleys can avoid conflicts and obstacles in a timely manner, reduces the risk of goods falling, and improves the overall accuracy and efficiency of logistics sorting.
Smart Images

Figure CN114663856B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of AGV trolleys and relates to an anti-collision scheduling and emergency obstacle avoidance method for an indoor AGV sorting trolley. Background Art
[0002] At present, the development of the e-commerce industry is unstoppable, and logistics is an extremely important part of the industry. The efficiency of logistics sorting has a great impact on the development speed of the industry. However, in actual operation, whether the sorting center adopts manual sorting or intelligent sorting, there will be abnormal situations such as operation conflicts, physical obstacles and even scattered express parcels. This requires a reliable and scientific sorting car scheduling method to make the sorting process of the transfer center more accurate and efficient. Summary of the invention
[0003] In view of the shortcomings of the prior art, the present invention proposes an anti-collision scheduling and emergency obstacle avoidance method for indoor AGV sorting carts. The method is based on the application of image edge computing in image processing, and predicts the probability of cargo falling and the danger radius before the cargo falls, so that the affected cart can avoid obstacles in time, and combines the improved A-Star algorithm to adjust the cart path in real time.
[0004] The present invention provides an indoor AGV sorting vehicle anti-collision scheduling and emergency obstacle avoidance method, comprising the following steps:
[0005] Step 1: First, obtain the surface image, surface depth image and road condition image of the cargo loaded on the trolley, and obtain the contour edge, edge depth, edge intersection and roughness depth of the cargo from the three images;
[0006] Step 2: Obtain the cargo scatter from the edge intersection, and obtain the scatter influence index from the edge depth and cargo scatter;
[0007] Step 3: Calculate the cargo drop index based on the information in the path image between the car and the next hop node and the car's driving conditions; obtain the cargo drop probability from the cargo scatter, convex hull depth and cargo drop impact index;
[0008] Step 4: Use a static road network optimal path solving method based on the Hesitating star algorithm to help the car select the next hop path and initially prevent conflicts;
[0009] Step 5: Use a processing solution decision function Meth to determine the processing method of the trolley according to the movement of the trolley and the characteristics of the cargo falling event.
[0010] The method of the present invention first obtains the surface image, surface depth image and road condition image of the loaded goods of the trolley, and obtains the contour edge, edge depth, edge intersection and rugged depth of the road surface of the goods from the three images; obtains the goods dispersion from the edge intersection, and obtains the dispersion influence index from the edge depth and the goods dispersion; uses the improved A-star algorithm to obtain the next hop node according to the relevant data of the trolley; calculates the goods drop index according to the information in the path image between the trolley and the next hop node and the driving condition of the trolley; obtains the drop probability of the goods from the dispersion of the goods, the convex hull depth and the goods drop influence index; when the drop probability reaches the preset drop threshold, the drop radius is obtained from the dispersion influence index, the convex hull depth and the goods drop influence index; and guides the affected trolley to avoid obstacles by the drop radius and the drop time. The present invention predicts the drop probability and the dangerous radius of the goods before the goods fall, so that the affected trolley can avoid obstacles in time, thereby greatly improving the robustness of the sorting process.
[0011] The further optimized technical solution of the present invention is as follows:
[0012] In step 1, in order to ensure that the car has a correct understanding of the external environment, a weighted defogging image metric evaluation method is used to process the car and its acquired information. The specific operations are as follows:
[0013] Step 1.1, the spherical depth camera deployed above the AGV car is used to obtain the surface image of the cargo loaded on the car, the surface depth image and the road condition image, and then the contour edge, edge depth, edge intersection and rugged depth of the road surface of the cargo on the car are obtained. The surface image specifically refers to the top view image of the cargo; while obtaining the image features, the corresponding data of the car is also obtained through the sensor, and the corresponding data includes the real-time speed, real-time acceleration and driving direction of the car; every 100 frames of image are divided into an interval and recorded as , where n∈{0,1,2,…,100}, and the following formula is used to calculate the image in the interval Structural similarity index, Visual information fidelity and Image color difference matrix,
[0014]
[0015]
[0016] In the formula, Represents the structural similarity parameters of the collected images at different scales The weighted sum of the values, It represents the column vector composed of the Euclidean distance of the color RGB values of each row of squares after the image is divided into several squares. It refers to the ratio of cargo logistics information labels that can be clearly read in the picture. Respectively represent the coordinates of different squares after the image is segmented, , , , Respectively represent the structural similarity index of each square after segmentation, Represents the Euclidean distance of the color RGB values of each row of squares;
[0017] Step 1.2, select , as well as The intersection of the mode intervals is the reference point , use the weighted dehazing method to evaluate the quality of the image at each stage (calculate the weighted quality assessment index),
[0018]
[0019] in, is the weighted quality assessment index of the image at each stage, , and are relative weights, Indicates the fidelity of visual information, that is, the ratio of cargo logistics information labels that can be clearly read in the collected pictures; Represents a column vector consisting of the Euclidean distance of the color RGB values of each row of squares after the image is divided into several squares. Represents the structural similarity parameters of the collected images at different scales The weighted sum of values, represents a unit vector matrix;
[0020] Calculated That is, the weighted quality assessment index, and then select each interval The first 25% of frames in the non-modal region were used as the reference sequence;
[0021] Step 1.3: After determining the reference sequence, first count the edge intersections in the image to get the number of edge intersections. , and then segment the contour edge according to the edge intersection point to obtain the first edge of the unobstructed goods and the second edge of the obstructed goods, and obtain the length of the first edge and the length of the second edge after segmentation, and obtain the obstruction length ratio according to the length of the first edge and the length of the second edge ( = length of first edge / length of second edge); then the exposed area of the goods is obtained from the first edge and the second edge (The exposed area of the cargo = the length of the first edge × the length of the second edge); After obtaining the surface area of the unobstructed cargo, the exposed area is calculated based on the exposed area. and the surface area The exposed area ratio is obtained by the ratio of the exposed area to the surface area of the unobstructed cargo (the surface area of the unobstructed cargo is the top view area of the unobstructed cargo);
[0022] The blocked length ratio, the exposed area ratio and the number of edge intersections are used to prepare for calculating the cargo dispersion.
[0023] In step 2, in order to improve the sorting efficiency, the emerging multi-layer AGV cart (as shown in FIG3 ) is involved. The scattered degree of the goods is obtained by the blocked length ratio, the exposed area ratio, and the number of edge intersections. The specific operation is as follows:
[0024] Step 2.1, determining the quantity of scattered goods according to the second edge of the obscured goods;
[0025] Step 2.2, determining the layer number of the scattered goods according to the contour edge and the edge depth of the contour edge;
[0026] Step 2.3: Calculate the scatteredness of the goods ,
[0027]
[0028] in, represents the number of edge intersections, Indicates the layer number is The blocking length ratio of the bulk cargo, Indicates the layer number is The exposed area of the bulk cargo, Indicates the number of shelves where scattered goods are stored. Represents the surface area of the cargo on the entire AGV platform;
[0029] Step 2.4: Obtain a dispersion influence index according to the edge depth and the dispersion of the goods, obtain a plurality of layer ratios according to the number of layers of each of the scattered goods and the actual number of layers of the goods, and obtain a dispersion influence index according to the dispersion of the goods and the layer ratio :
[0030]
[0031] In the formula, Indicates the layer number is The scatteredness of scattered goods affects the index, The layer number where the loose goods are located.
[0032] In step 3, the probability of cargo falling is obtained according to the cargo scattering, convex hull depth and cargo falling impact index, and the calculation formula is as follows:
[0033]
[0034] in, is the drop probability of the goods, represents the influence coefficient of the scatteredness of the goods, Indicates the impact coefficient of the cargo drop impact index, Indicates the convex depth of the cargo carried by the trolley. represents the preset convex hull safety threshold, represents the maximum convex hull depth that the car is allowed to pass through, Indicates the impact index of the cargo drop. In this embodiment of the present invention, the influence coefficient of the cargo scatter degree is , the impact coefficient of cargo drop impact index , preset convex hull safety threshold preset convex hull safety threshold .
[0035] In step 3, in order to quantify the risk of cargo falling, a formula for calculating the probability of fire poison falling is proposed. The cargo falling index is calculated based on the information in the path image between the trolley and the next hop node and the driving condition of the trolley. The cargo falling impact index takes into account both straight-line driving and the trolley's universal wheel rotation process. The specific operation is as follows:
[0036] When driving in a straight line, calculate the impact index of the cargo falling :
[0037]
[0038] in, represents the speed standard deviation before the straight-line driving speed change, Indicates the acceleration standard of each in-situ turn of the car, is the quality of the item, is the static friction parameter of the car surface, is the acceleration due to gravity, The maximum acceleration of the vehicle in motion, is the speed of the car every time it makes an emergency stop or turn. is the number of speed change points during the movement of the car;
[0039] When turning in place, calculate the impact index of the cargo drop :
[0040]
[0041] in, Indicates the real-time speed of the car when it is traveling on a curve, Indicates the maximum speed allowed when the car is traveling on a curve. is the steering angle of the car.
[0042] In step 4, in order to finally realize the scheduling and control of the sorting cart, a static road network optimal path solving method called Hesitatingstar algorithm is used. This algorithm improves the commonly used cost function of integrity robot path planning and introduces non-integrity constraints, making it more suitable for obstacle avoidance and abnormal situation handling of non-integrity robots such as AGV carts in complex indoor environments, including:
[0043] Input: Occupancy grid of the environment , Starting position , Target location AGV car specifications (motion model of the self-vehicle) , the probability of cargo falling at each grid point in the vehicle path and the locations of known obstacles ;
[0044] Output: The optimal path for the AGV from a given starting position to the desired target position (path);
[0045] The following steps create the desired output based on the given input:
[0046] (1) By Locate the start point, target, and obstacle locations to provide an occupancy grid map;
[0047] (2) Grid the map at a reasonable scale;
[0048] (3) In the occupancy grid graph, assign a cost function value to each surrounding node around the current node , mainly by and Two polynomial weights (weighting factors is the ratio of the straight-line distance between the current node and the target node to the straight-line distance between the initial node and the target node), where represents the minimum cost to reach the center of each adjacent node of the node, is the minimum step cost from the current node to the target node (the set of paths from the current node to the target node in the figure is ). Usually the cost function value is chosen The lowest adjacent node is used as the successor node:
[0049]
[0050]
[0051]
[0052] In the formula, is an adjustable weight parameter (which can be adjusted as needed), is the number of items in the cart, To calculate the cargo drop impact index when driving in a straight line, To calculate the cargo drop impact index when turning in place, To carry the weight of items on the trolley;
[0053] The algorithm stops until it reaches the target node, and a current optimal path will be obtained;
[0054] (4) Combined with the kinematic constraints of the AGV car itself, the multi-galaxy algorithm is used to follow the extracted waypoints, adjust the car's movement trajectory, find the final optimal path to the target position, and use the simplified car model to calculate the global position: The vehicle kinematic model is as follows:
[0055]
[0056]
[0057]
[0058] in, They represent the moving speed of the car in eight directions: front, back, left, right, left front 45°, right front 45°, left rear 45°, and right rear 45°. represents the required travel length, is the wheelbase, is the steering angle of the ego vehicle, , Both are dynamic correction factors ( is the unit vector of the universal wheel of the car in each direction, is the displacement in each direction), They represent the motion determination vectors of the current position of the car in 8 directions (if there is no speed, it is the 0 vector; if there is speed, it is the unit vector in that direction). represents the motion deflection angle, represents the radius of the centrifugal circle, Indicates the wheelbase of the trolley. represents the deflection angle of this iteration, represents the deflection angle of the previous iteration, Indicates the vehicle head speed in this iteration. Indicates the vehicle head speed in the previous iteration. Indicates the vertical speed of the vehicle head in this iteration. Indicates the speed perpendicular to the vehicle head in the previous round iteration.
[0059] In step 4, in view of the common car collision phenomenon in daily sorting work, a dynamic adjustment mechanism to avoid car travel conflicts is proposed based on the optimal path solving method of the static road network of the Hesitating star algorithm, as follows:
[0060] Step 4.1: Set the priority for each car operation in the environment. First, calculate the priority reference value of each car. ,
[0061] ,
[0062] in, The value is always between, The upper limit of the vehicle's load-bearing capacity. is the weight of the cargo carried by the trolley, is the remaining percentage of the specified pickup time. is the ratio of the current remaining distance of the car to the distance of receiving the pickup instruction, is the working status of the car, where a truck is 0 and an empty car is 1. For each car involved in the risk of conflict, Gather all the cars in the venue;
[0063] Finally, sort the cars according to the priority reference value. The car with the higher priority has the higher priority to keep the original path.
[0064] Step 4.2: Calculate the conflict index between the cars that appear at the same time in each 2*2 unit grid ,
[0065]
[0066]
[0067] in, is the grid conflict coefficient, They are the maximum distances from the top-down center of each car that appears simultaneously in each 2*2 unit grid to the horizontal and vertical boundaries of the grid. is the unit grid area, For each car involved in the risk of conflict, is the set of cars involved in the risk of conflict, The time that different cars coexist in each 2*2 unit grid according to the original path planning. is the length of the car body, is the relative velocity vector between two cars in each 2*2 unit grid, The distance between the two vehicles' endpoints is closer. The distance between the two vehicle endpoints is far;
[0068] Step 4.3: Determine the severity of the vehicle operation. If , it means that the car conflict in this area is serious. Except for the car with the highest priority in the area, all other cars will re-plan their routes and choose the best grid among the 8 adjacent grids except the current 2*2 unit grid as the next hop;
[0069] if , it means that the area is crowded with cars. Except for the cars with the first 1 / 2 priority (rounded down), all cars in the area will re-plan their routes and choose the best grid among the 8 adjacent grids except the current 2*2 unit grid as the next hop;
[0070] if , then there is no need to adjust the path selection of the car in the area. is the number of vehicles in the adjacent 8 grids.
[0071] In step 4, in order to optimize the dynamic mediation mechanism to avoid the vehicle travel conflict, a control method for optimizing the vehicle avoidance process is proposed, and the specific operation is as follows:
[0072] (1) Assuming that the car chooses to avoid, it accelerates The driving time for uniform deceleration is , we can get the distance traveled by the avoiding vehicle A during deceleration for:
[0073]
[0074] in, It is the reaction time from when the car receives the signal to when it makes the corresponding deceleration operation, and the machine reaction time from when the car makes the deceleration operation to when it starts to decelerate. represents the relationship between the speed of the avoiding vehicle A and time, is the integral symbol (this integral is integrated with respect to t), is the speed of the vehicle before it received the warning information, is the vehicle’s acceleration at this moment;
[0075] (2) Through reaction time The speed of the vehicle before receiving the warning information in the reaction time to avoid the vehicle can be obtained Distance traveled while moving at a constant speed for:
[0076]
[0077] Obviously, the total displacement distance of the car in the avoidance process is and ;
[0078] (3) Assume that the vertical swing distance of the trolley is ,So
[0079]
[0080] in, is the linear displacement of the trolley;
[0081] (4) To ensure that the car does not touch other cars or obstacles during the avoidance process, the vertical swing distance of the car is , it is easy to get: .
[0082] In step 4, if a parcel is dropped, an abnormal situation handling method is required to solve the problem of dropped parcels, and the operations are as follows:
[0083] (1) Use the car's built-in bird's-eye view depth camera to determine the location of the dropped express and partition the area. , that is, to mark obstacles;
[0084] (2) Using a processing decision function To determine how to handle dropped shipments:
[0085]
[0086] In the formula, is the number of grids occupied by the dropped cargo, is the distance between the car and the drop point, is the distance between the car and the destination;
[0087] when Less than When the number of cars near the dropped package is small or the car is closer to the dropped location, the car must change its target to the dropped location and retrieve the package under the constraints of the dynamic mediation mechanism to avoid car travel conflicts;
[0088] when Greater than When the number of cars near the dropped parcel is large or the car is closer to the target node, the car will choose to deliver the remaining parcels to the target node and then determine whether it can retrieve the dropped parcel.
[0089] when exist and When the value is between 0 and 1 (including 0 and 1), the situation of the dropped package is still unclear. The car can pause in place and wait for the function value to change, and then operate after leaving the interval. BRIEF DESCRIPTION OF THE DRAWINGS
[0090] Figure 1 The figure is a flow chart of the obstacle avoidance method of the AGV sorting vehicle in the present invention.
[0091] Figure 2 It is a schematic diagram of the multi-layer pallet AGV in the present invention.
[0092] Figure 3a It is the flow chart of Hesitating star algorithm in the present invention; Figure 3b The present invention is a flowchart of the method for solving the optimal path of a static road network using the Hesitating Star algorithm.
[0093] Figure 4 is an example diagram of the Hesitating star algorithm path adjustment process. Figure 4a The occupancy grid map in the Hesitating star algorithm of the present invention; Figure 4b It is a gridded occupancy grid map in the Hesitating star algorithm of the present invention; Figure 4c A schematic diagram of the current optimal path in the Hesitating star algorithm of the present invention; Figure 4d It is a schematic diagram of the final optimal path to reach the target location in the Hesitating Star algorithm of the present invention.
[0094] Figure 5 This is a flow chart of the small car charging avoidance mechanism of the present invention. DETAILED DESCRIPTION
[0095] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings: This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and specific operation process are given, but the protection authority of the present invention is not limited to the following embodiments.
[0096] Example 1
[0097] This embodiment proposes an indoor AGV sorting vehicle anti-collision scheduling and emergency obstacle avoidance method, such as Figure 1 and Figure 2 As shown, the following steps are included:
[0098] Step 1: First, obtain the surface image, surface depth image and road condition image of the cargo loaded on the trolley, and obtain the contour edge, edge depth, edge intersection and roughness depth of the cargo from the three images;
[0099] Step 2: Obtain the cargo scatter from the edge intersection, and obtain the scatter influence index from the edge depth and cargo scatter;
[0100] Step 3: Calculate the cargo drop index based on the information in the path image between the car and the next hop node and the car's driving conditions; obtain the cargo drop probability from the cargo scatter, convex hull depth and cargo drop impact index;
[0101] Step 4: Use a static road network optimal path solving method based on the Hesitating star algorithm to help the car select the next hop path and initially prevent conflicts. Specifically, when the drop probability reaches the preset drop threshold, the drop radius is obtained by the zero-scattering influence index, the convex hull depth, and the cargo drop influence index; the drop radius and drop time are used to guide the affected car to avoid obstacles.
[0102] Step 5: Use a processing solution decision function Meth to determine the processing method of the trolley according to the characteristics of the trolley movement and the cargo drop event. At the same time, use the improved A-star algorithm to obtain the next hop node based on the trolley related data.
[0103] In order to ensure that the car has a correct understanding of the external environment, a weighted defogging image metric evaluation method is used to process the car and its acquired information. The specific operations are as follows:
[0104] Step 1.1, use the spherical depth camera deployed above the AGV to obtain the surface image of the cargo loaded on the AGV, the surface depth image and the road condition image, and then obtain the contour edge, edge depth, edge intersection and rugged depth of the cargo on the AGV. The surface image specifically refers to the top view image of the cargo. While obtaining the image features, the corresponding data of the AGV is also obtained through the sensor, including the real-time speed, real-time acceleration and driving direction of the AGV. Divide each 100-frame image into an interval and record it as , where n∈{0,1,2,…,100}, and calculate the image in the interval Structural similarity index, Visual information fidelity and Image color difference matrix,
[0105]
[0106]
[0107] In the formula, Represents the structural similarity parameters of the collected images at different scales The weighted sum of the values, It represents the column vector composed of the Euclidean distance of the color RGB values of each row of squares after the image is divided into several squares. It refers to the ratio of cargo logistics information labels that can be clearly read in the picture. Respectively represent the coordinates of different squares after the image is segmented, , , , Respectively represent the structural similarity index of each square after segmentation, Represents the Euclidean distance of the color RGB values of each row of squares.
[0108] Step 1.2, select , as well as The intersection of the mode intervals is the reference point , use the weighted dehazing method to evaluate the quality of the image at each stage (calculate the weighted quality assessment index):
[0109]
[0110] in, is the weighted quality assessment index of the image at each stage, , and are relative weights, Indicates the fidelity of visual information, that is, the ratio of cargo logistics information labels that can be clearly read in the collected pictures; Represents a column vector consisting of the Euclidean distance of the color RGB values of each row of squares after the image is divided into several squares. Represents the structural similarity parameters of the collected images at different scales The weighted sum of values, represents a unit vector matrix.
[0111] Calculated That is, the weighted quality assessment index, and then select each interval The first 25% of the frames in the non-modal region are used as the reference sequence.
[0112] Step 1.3: After determining the reference sequence, first count the edge intersections in the image to get the number of edge intersections. ;
[0113] The contour edge is segmented according to the edge intersection point to obtain the first edge of the unobstructed cargo and the second edge of the obstructed cargo, and the length of the first edge and the length of the second edge after segmentation are obtained. The obstruction length ratio is obtained according to the length of the first edge and the length of the second edge. ( = length of first edge / length of second edge);
[0114] The exposed area of the goods is obtained from the first edge and the second edge (The exposed area of the cargo = the length of the first edge × the length of the second edge); After obtaining the surface area of the unobstructed cargo, the exposed area is calculated based on the exposed area. and the surface area The exposed area ratio is obtained by the ratio of the exposed area to the surface area of the unobstructed cargo (the surface area of the unobstructed cargo is the top view area of the unobstructed cargo);
[0115] The blocked length ratio, the exposed area ratio and the number of edge intersections are used to prepare for calculating the cargo dispersion.
[0116] In order to improve the sorting efficiency, the emerging multi-layer AGV cart (as shown in Figure 3) is involved. The scattered degree of the goods is obtained by the blocked length ratio, the exposed area ratio, and the number of edge intersections. The specific operation is as follows:
[0117] Step 2.1: Determine the quantity of scattered goods according to the second edge of the obscured goods.
[0118] Step 2.2: Determine the layer number of the scattered goods according to the contour edge and the edge depth of the contour edge.
[0119] Step 2.3: Calculate the scatteredness of the goods :
[0120]
[0121] in, represents the number of edge intersections, Indicates the layer number is The blocking length ratio of the bulk cargo, Indicates the layer number is The exposed area of the bulk cargo, Indicates the number of shelves where scattered goods are stored. Represents the surface area of the cargo on the entire AGV platform.
[0122] Step 2.4, obtaining a dispersion influence index according to the edge depth and the dispersion of the goods, including:
[0123] According to the number of layers of each of the scattered goods and the actual number of layers of the goods, a plurality of layer ratios are obtained, and the scattered degree impact index is obtained from the scattered degree of the goods and the layer ratio. :
[0124]
[0125] In the formula, Indicates the layer number is The scatteredness of scattered goods affects the index, The layer number where the loose goods are located.
[0126] The probability of cargo falling is obtained based on cargo scatter, convex hull depth and cargo falling impact index. The calculation formula is as follows:
[0127]
[0128] in, is the drop probability of the goods, represents the influence coefficient of the scatteredness of the goods, Indicates the impact coefficient of the cargo drop impact index, Indicates the convex depth of the cargo carried by the trolley. represents the preset convex hull safety threshold, represents the maximum convex hull depth that the car is allowed to pass through, Indicates the impact index of the cargo drop. In this embodiment of the present invention, the influence coefficient of the cargo scatter degree is , the impact coefficient of cargo drop impact index , preset convex hull safety threshold preset convex hull safety threshold .
[0129] In order to quantify the risk of cargo falling, a formula for calculating the probability of fire poison falling is proposed. The cargo falling index is calculated based on the information in the path image between the trolley and the next hop node and the driving conditions of the trolley. The cargo falling impact index takes into account both straight-line driving and the trolley's universal wheel rotation process. The specific operation is as follows:
[0130] When driving in a straight line, calculate the impact index of the cargo falling :
[0131]
[0132] in, represents the speed standard deviation before the straight-line driving speed change, Indicates the acceleration standard of each in-situ turn of the car, is the quality of the item, is the static friction parameter of the car surface, is the acceleration due to gravity, The maximum acceleration of the vehicle in motion, is the speed of the car every time it makes an emergency stop or turn. It is the number of speed change points during the movement of the car.
[0133] When turning in place, calculate the impact index of the cargo drop :
[0134]
[0135] in, Indicates the real-time speed of the car when it is traveling on a curve, Indicates the maximum speed allowed when the car is traveling on a curve. is the steering angle of the car.
[0136] In order to ultimately achieve the scheduling and control of the sorting cart, a static road network optimal path solving method called Hesitating Star algorithm is used. This algorithm improves the commonly used cost function of integrity robot path planning and introduces non-integrity constraints, making it more suitable for obstacle avoidance and abnormal situation handling of non-integrity robots such as AGV carts in complex indoor environments, including:
[0137] Input: Occupancy grid of the environment , Starting position , Target location , AGV car specifications (motion model of the self-vehicle), probability of cargo falling at each grid point in the car path and the locations of known obstacles .
[0138] Output: The optimal path for the AGV from a given starting position to the desired target position (path).
[0139] like Figure 3a and 3b As shown, the following steps create the desired output based on the given input:
[0140] (1) By The starting point, target and obstacle positions are located on the grid to provide an occupancy grid map, such as Figure 4a shown.
[0141] (2) The map is gridded at a reasonable scale. In this embodiment, the map is divided into 6*6 grids, such as Figure 4b shown.
[0142] (3) In the occupancy grid graph, assign a cost function value to each surrounding node around the current node , mainly by and Two polynomial weights (weighting factors is the ratio of the straight-line distance between the current node and the target node to the straight-line distance between the initial node and the target node), where Indicates the minimum cost to reach the center of each adjacent node (8 directions) of the node, is the minimum step cost from the current node to the target node (the set of paths from the current node to the target node in the figure is ). Usually the cost function value is chosen The lowest adjacent node is used as the successor node:
[0143]
[0144]
[0145]
[0146] In the formula, is an adjustable weight parameter (which can be adjusted as needed), is the number of items in the cart, To calculate the cargo drop impact index when driving in a straight line, To calculate the impact index of cargo falling when turning in place, To carry the weight of items on the trolley;
[0147] The algorithm stops when it reaches the target node, and a current optimal path is obtained, such as Figure 4c shown.
[0148] (4) Combined with the kinematic constraints of the AGV, the multi-galaxy algorithm is used to follow the extracted waypoints, adjust the vehicle's movement trajectory, and find the final optimal path to the target location, such as Figure 4d As shown. Using the simplified car model, the global position is calculated as The vehicle kinematic model is as follows:
[0149]
[0150]
[0151]
[0152] in, They represent the moving speed of the car in eight directions: front, back, left, right, left front 45°, right front 45°, left rear 45°, and right rear 45°. represents the required travel length, is the wheelbase, is the steering angle of the ego vehicle, , Both are dynamic correction factors ( is the unit vector of the universal wheel of the car in each direction, is the displacement in each direction), They represent the motion determination vectors of the current position of the car in eight directions: front, back, left, right, 45° left front, 45° right front, 45° left back, and 45° right back (if there is no speed, it is a 0 vector; if there is a speed, it is a unit vector in that direction). represents the motion deflection angle, represents the radius of the centrifugal circle, Indicates the wheelbase of the trolley. represents the deflection angle of this iteration, represents the deflection angle of the previous iteration, Indicates the vehicle head speed in this iteration. Indicates the vehicle head speed in the previous iteration. Indicates the vertical speed of the vehicle head in this iteration. Indicates the speed perpendicular to the vehicle head in the previous round iteration.
[0153] Aiming at the common car collision phenomenon in daily sorting work, a dynamic adjustment mechanism to avoid car travel conflicts is proposed based on the optimal path solution method of static road network of Hesitating Star algorithm. Figure 5 As shown, the details are as follows:
[0154] Step 4.1: Set the priority for each car operation in the environment. First, calculate the priority reference value of each car. ,
[0155] ,
[0156] in, The value is always between, The upper limit of the vehicle's load-bearing capacity. is the weight of the cargo carried by the trolley, is the remaining percentage of the specified pickup time. is the ratio of the current remaining distance of the car to the distance of receiving the pickup instruction, is the working status of the car, where a truck is 0 and an empty car is 1. For each car involved in the risk of conflict, Gather all the cars in the venue;
[0157] Finally, sort the cars according to the priority reference value. The car with the higher priority has the higher priority to keep the original path.
[0158] Step 4.2: Calculate the conflict index between the cars that appear at the same time in each 2*2 unit grid ,
[0159]
[0160]
[0161] in, is the grid conflict coefficient, They are the maximum distances from the top-down center of each car that appears simultaneously in each 2*2 unit grid to the horizontal and vertical boundaries of the grid. is the unit grid area, For each car involved in the risk of conflict, is the set of cars involved in the risk of conflict, The time that different cars coexist in each 2*2 unit grid according to the original path planning. is the length of the car body, is the relative velocity vector between two cars in each 2*2 unit grid, The distance between the two vehicles' endpoints is closer. The distance between the two vehicle endpoints is far;
[0162] Step 4.3: Determine the severity of the vehicle operation. If , it means that the car conflict in this area is serious. Except for the car with the highest priority in the area, all other cars will re-plan their routes and choose the best grid among the 8 adjacent grids except the current 2*2 unit grid as the next hop;
[0163] if , it means that the area is crowded with cars. Except for the cars with the first 1 / 2 priority (rounded down), all cars in the area will re-plan their routes and choose the best grid among the 8 adjacent grids except the current 2*2 unit grid as the next hop;
[0164] if , then there is no need to adjust the path selection of the car in the area.
[0165] is the number of vehicles in the adjacent 8 grids.
[0166] In order to optimize the dynamic mediation mechanism to avoid car travel conflicts, a control method to optimize the car avoidance process is proposed. The specific operations are as follows:
[0167] (1) Assuming that the car chooses to avoid, it accelerates The driving time for uniform deceleration is , we can get the distance traveled by the avoiding vehicle A during deceleration for:
[0168]
[0169] in, It is the reaction time from when the car receives the signal to when it makes the corresponding deceleration operation, and the machine reaction time from when the car makes the deceleration operation to when it starts to decelerate. represents the relationship between the speed of the avoiding vehicle A and time, is the integral symbol (this integral is integrated with respect to t), is the speed of the vehicle before it received the warning information, is the vehicle’s acceleration at this moment;
[0170] (2) Through reaction time The speed of the vehicle before receiving the warning information in the reaction time to avoid the vehicle can be obtained Distance traveled while moving at a constant speed for:
[0171]
[0172] Obviously, the total displacement distance of the car in the avoidance process is and ;
[0173] (3) Assume that the vertical swing distance of the trolley is ,So
[0174]
[0175] in, is the linear displacement of the trolley;
[0176] (4) To ensure that the car does not touch other cars or obstacles during the avoidance process, the vertical swing distance of the car is , it is easy to get: .
[0177] In order to deal with problems such as dropped express mail, the present invention proposes an AGV sorting vehicle obstacle avoidance method based on image edge computing. If express mail drops, an abnormal situation handling method is required to solve the problem of dropped express mail, and the operations are as follows:
[0178] (1) Use the car's built-in bird's-eye view depth camera to determine the location of the dropped express and partition the area. , that is, to mark obstacles;
[0179] (2) Using a processing decision function To determine how to handle dropped shipments:
[0180]
[0181] In the formula, is the number of grids occupied by the dropped cargo, is the distance between the car and the drop point, is the distance between the car and the destination;
[0182] when Less than When the number of cars near the dropped package is small or the car is closer to the dropped location, the car must change its target to the dropped location and retrieve the package under the constraints of the dynamic mediation mechanism to avoid car travel conflicts.
[0183] when Greater than When the number of cars near the dropped parcel is large or the car is closer to the target node, the car will choose to deliver the remaining parcels to the target node and then determine whether it can retrieve the dropped parcel.
[0184] when exist and When the value is between 0 and 1 (including 0 and 1), the situation of the dropped package is still unclear. The car can pause in place and wait for the function value to change, and then operate after leaving the interval.
[0185] The above description is only a specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technology can understand and think of any changes or substitutions within the technical scope disclosed by the present invention, which should be included in the scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. An indoor AGV sorting vehicle anti-collision scheduling and emergency obstacle avoidance method, characterized in that: The following steps are involved: Step 1: First, obtain the surface image, surface depth image and road condition image of the cargo loaded on the trolley, and obtain the contour edge, edge depth, edge intersection and roughness depth of the cargo from the three images; Step 2: Obtain the cargo scatter from the edge intersection, and obtain the scatter influence index from the edge depth and cargo scatter; Step 3: Calculate the cargo drop index based on the information in the path image between the car and the next hop node and the car's driving conditions; obtain the cargo drop probability from the cargo scatter, convex hull depth and cargo drop impact index; Step 4: Use the static road network optimal path solving method to help the car select the next hop path and initially prevent conflicts; The optimal path solution method for static road networks using the Hesitating Star algorithm, obstacle avoidance and abnormal situation handling of AGV vehicles in complex indoor environments, including: Input: occupancy grid map M of the environment, starting position qS, target position qG, AGV car specifications Veh, probability of cargo falling at each grid point in the car path p fall and the location of known obstacles OM; Output: The optimal path for the AGV from a given starting position to the desired target position; The following steps create the desired output based on the given input: (1) By fall ,OM) to locate the starting point, target and obstacle positions to provide an occupancy grid map; (2) Grid the map at a reasonable scale; (3) In the occupancy grid graph, a cost function value F(n) is assigned to each surrounding node around the current node, which is mainly obtained by weighting two polynomials g(n) and h(n), where h(n) represents the minimum cost to reach the center of each adjacent node of the node, and g(n) is the minimum step cost from the current node to the target node; the adjacent node with the lowest cost function value F(n) is usually selected as the successor node: F(n)=u·g(n)+(1-u)·h(n) g(n)=min{d i } In the formula, u is an adjustable weight parameter, item_num is the number of items in the cart, and I d To calculate the cargo drop impact index when driving in a straight line, I t To calculate the impact index of cargo falling when turning in place, item_weigh is the weight of the item carried by the trolley; The algorithm stops until it reaches the target node, and a current optimal path will be obtained; (4) Using the simplified car model, the vehicle kinematic model with a global position of (X, Y, θ) is calculated as follows: Among them, inf, bac, lef, rig, il, ir, bl, br represent the moving speed of the car in the eight directions of front, back, left, right, left front 45°, right front 45°, left rear 45° and right rear 45° respectively, L represents the required travel length, W b is the wheelbase, α is the steering angle of the ego vehicle, τ, are all dynamic correction factors. dpm1, dpm2, …, dpm8 represent the motion determination vectors of the current position of the car in eight directions, β represents the motion deflection angle, R represents the radius of the centrifugal circle, and W b Indicates the wheelbase of the trolley, θ n represents the deflection angle of this iteration, θ n-1 represents the deflection angle of the previous iteration, X n Indicates the vehicle head speed in this iteration, X n-1 Indicates the vehicle head speed in the previous iteration, Y n Indicates the vertical speed of the vehicle head in this iteration, Y n-1 Indicates the vertical speed of the vehicle head in the previous round iteration; Step 5: Use a processing solution decision function to determine the processing method of the trolley based on the movement of the trolley and the characteristics of the cargo falling event.
2. According to claim 1, an indoor AGV sorting vehicle anti-collision scheduling and emergency obstacle avoidance method is characterized in that: In step 1, the weighted defogging image metric evaluation method is used to process the car and its acquired information. The specific operations are as follows: Step 1.1: Use the spherical depth camera deployed above the AGV to obtain the surface image, surface depth image and road condition image of the cargo loaded on the AGV, and then obtain the contour edge, edge depth, edge intersection and rugged depth of the cargo on the AGV. While obtaining the image features, the real-time speed, real-time acceleration and driving direction of the AGV are also obtained. Divide every 100 frames of images into an interval denoted as δ n , where n∈{0,1,2,…,100}, and the MSSIM structural similarity index, VIF visual information fidelity and MSiCID image color difference matrix of each image in the interval are calculated by the following formula: Where, MSSIM(x, y) represents the weighted sum of the structural similarity parameter SSIM values of each collected image at different scales, MSiCID(x, y) represents the column vector composed of the Euclidean distance of the color RGB values of each row of squares after the image is divided into several squares, VIF refers to the ratio of the cargo logistics information identification that can be clearly read in the image, Clo n Represents the Euclidean distance of the color RGB values of each row of squares; Step 1.2: Select the intersection of SSIM, VIF and iCID mode intervals as the reference point y, and use the weighted dehazing method to evaluate the quality of the image at each stage. WM-DIE(x,y) =ω1|1-VIF(x,y)|·[ω2(MSiCID(x k ,y) n )+ω3(E-MSSIM(x,y))] Among them, WM-DIE(x, y) is the weighted quality evaluation index of the image at each stage, ω1, ω2 and ω3 are relative weights, VIF(x, y) represents the ratio of the cargo logistics information identification that can be clearly read in the collected images; MSiCID(x k , y) represents the column vector composed of the Euclidean distance of the color RGB values of each row of squares after the image is divided into several squares, MSSIM (x, y) represents the weighted sum of the structural similarity parameter SSIM values of each collected image at different scales, and E represents the unit vector matrix; Then select each interval δ n The first 25% of frames in the non-modal region were used as the reference sequence; Step 1.3: First, count the edge intersections in the image to obtain the number of edge intersections N, and then segment the contour edge according to the edge intersections to obtain the first edge of the unobstructed cargo and the second edge of the obstructed cargo, and obtain the length of the first edge and the length of the second edge after segmentation, and obtain the obstruction length ratio L according to the length of the first edge and the length of the second edge. k ; Then the exposed area S of the goods is obtained from the first edge and the second edge k After obtaining the surface area of the unobstructed cargo, according to the exposed area S k The ratio of the exposed area to the surface area S is obtained; The blocked length ratio, the exposed area ratio and the number of edge intersections are used to prepare for calculating the cargo dispersion.
3. According to claim 2, an indoor AGV sorting vehicle anti-collision scheduling and emergency obstacle avoidance method is characterized in that: In step 2, the cargo scatter is obtained by the blocking length ratio, the exposed area ratio, and the number of edge intersections. The specific operation is as follows: Step 2.1, determining the quantity of scattered goods according to the second edge of the obscured goods; Step 2.2, determining the layer number of the scattered goods according to the contour edge and the edge depth of the contour edge; Step 2.3: Calculate the cargo dispersion ε k , Wherein, N represents the number of edge intersections, L k represents the blocking length ratio of the loose goods with layer number k, S k represents the exposed area of the scattered goods at layer number k, l represents the number of shelves with scattered goods, and S represents the surface area of the goods on the entire AGV platform; Step 2.4, obtaining a dispersion influence index according to the edge depth and the dispersion of the goods, In the formula, ε′ k It represents the scattered degree impact index of the scattered goods with layer number k, where k is the layer number where the scattered goods are located.
4. According to claim 3, an indoor AGV sorting vehicle anti-collision scheduling and emergency obstacle avoidance method is characterized in that: In step 3, the probability of cargo falling is obtained according to the cargo scattering, convex hull depth and cargo falling impact index, and the calculation formula is as follows: Among them, p fall is the probability of cargo falling, β is the influence coefficient of cargo dispersion, γ is the influence coefficient of cargo falling influence index, D is the depth of convex hull of cargo carried by the trolley, M1 is the preset convex hull safety threshold, and D max represents the maximum convex hull depth that the trolley is allowed to pass through, I represents the impact index of cargo falling, ε′ k Indicates the fragmentation of the goods.
5. According to claim 4, an indoor AGV sorting vehicle anti-collision scheduling and emergency obstacle avoidance method is characterized in that: In step 3, the cargo drop index is calculated based on the information in the path image between the car and the next hop node and the driving condition of the car. The specific operation is as follows: When driving in a straight line, calculate the cargo drop impact index I d : Among them, σ v represents the standard deviation of the speed before the straight-line driving speed change, σ a Indicates the acceleration standard of each in-situ turn of the car, m c is the mass of the object, μ is the static friction parameter of the surface of the car, g is the acceleration due to gravity, and a diff The maximum acceleration of the vehicle in motion, v diff is the speed of the car each time it stops and turns suddenly, and num is the number of speed change points during the car's movement; When turning in place, calculate the cargo drop impact index I t : Wherein, v represents the real-time speed of the car when it is traveling on a curve, v max It represents the maximum speed allowed when the car is traveling on a curve, and Δθ is the steering angle of the car.
6. According to claim 5, an indoor AGV sorting vehicle anti-collision scheduling and emergency obstacle avoidance method is characterized in that: In step 4, based on the optimal path solving method for static road networks of the Hesitating star algorithm, a dynamic adjustment mechanism for avoiding vehicle travel conflicts is proposed, as follows: Step 4.1: Set the priority for each car operation in the environment. First, calculate the priority reference value Pri of each car. j , Among them, weight cap is the upper limit of the vehicle's load-bearing capacity, item weigh is the weight of the cargo carried by the trolley, t order is the remaining proportion of the specified pickup time, d gap is the ratio of the current remaining distance of the car to the distance of the received pickup instruction. sta is the working state of the car, where 0 is for a truck and 1 is for an empty car, j is for each car involved in the conflict risk, and C is the set of all cars in the venue; Finally, sort the cars according to the priority reference value. The car with the higher priority has the higher priority to keep the original path. Step 4.2: Calculate the conflict index p between the cars that appear at the same time in each 2*2 unit grid conflict , Among them, μ loc is the grid conflict coefficient, d xj d yj are the maximum distances from the top view center of each car that appears simultaneously in each 2*2 unit grid to the horizontal and vertical boundaries of the grid where it is located, S grid is the unit grid area, j is each car involved in the conflict risk, C conflict is the set of cars involved in the conflict risk, Δt visible L is the time that different cars coexist in each 2*2 unit grid according to the original path planning. car is the length of the car body, v jnjm is the relative velocity vector between two cars in each 2*2 unit grid, d close The distance between the two vehicle endpoints is closer, d remote The distance between the two vehicle endpoints is far; Step 4.3: Determine the severity of the vehicle operation. If This indicates that there is a serious conflict between vehicles in the area. All vehicles except the vehicle with the highest priority in the area will re-plan their routes and choose the best grid among the 8 adjacent grids excluding the current 2*2 unit grid as the next hop. if This means that the area is crowded with cars. Except for the top 1 / 2 cars in the area, all cars will re-plan their routes and choose the best grid among the 8 adjacent grids excluding the current 2*2 unit grid as the next hop. if There is no need to adjust the path selection of the car in the area.
7. According to claim 6, an indoor AGV sorting vehicle anti-collision scheduling and emergency obstacle avoidance method is characterized in that: In step 4, a control method for optimizing the vehicle avoidance process is proposed based on a dynamic mediation mechanism for avoiding vehicle travel conflicts. The specific operations are as follows: (1) Assuming that after the car chooses to avoid, the time it takes to travel with a uniform deceleration at an acceleration of a1 is t1, the distance x1 traveled during the deceleration of the avoiding vehicle A is: Among them, Δt is the reaction time from when the car receives the signal to when it makes the corresponding deceleration operation and the machine reaction time from when the car makes the deceleration operation to when it starts to decelerate itself, v A(t) represents the relationship between the speed of the avoiding vehicle A and time, v A is the speed of the vehicle before it receives the warning information, and a1 is the acceleration of the vehicle at this moment; (2) The reaction time Δt can be used to obtain the vehicle speed v before the vehicle receives the warning information during the reaction time. A The distance traveled when moving at a uniform speed is x2: Obviously, the total displacement distance of the car in the avoidance process is x c And x c =x1+x2; (3) Let the vertical swing distance of the trolley be |Δx int |, then Among them, R dir is the linear displacement of the trolley; (4) To ensure that the car does not touch other cars or obstacles during the avoidance process, the vertical swing distance of the car Easy to get:
8. According to claim 7, an indoor AGV sorting vehicle anti-collision scheduling and emergency obstacle avoidance method is characterized in that: In step 4, if a parcel is dropped, an abnormal situation handling method is required to solve the problem of dropped parcels, and the operations are as follows: (1) Use the car's built-in bird's-eye view depth camera to determine the partition where the dropped express is located, and set the block OM(n)≠0, which is used as an obstacle mark; (2) Use a processing solution decision function Meth to determine the method for handling dropped express mail: In the formula, num ocu is the number of grids occupied by the dropped goods, d fts is the distance between the trolley and the drop point, d gts is the distance between the car and the destination; When Meth is less than 0, it means that there are fewer cars near the dropped package or the car is closer to the dropped location. The car must change its target to the dropped location and retrieve the package under the constraints of the dynamic mediation mechanism to avoid car travel conflicts. When Meth is greater than 1, it means that there are more cars near the dropped parcel or the car is closer to the target node. The car chooses to deliver the remaining parcels to the target node and then determines whether it can retrieve the dropped parcel. When Meth is between 0 and 1, the situation of dropped express is still unclear. The car can be paused on the spot, waiting for the function value to change, and then operate after leaving the interval.
Citation Information
Patent Citations
A warehouse shortest path optimization method based on simultaneous picking of goods by multiple people
CN109447317A
AGV obstacle avoidance method based on cloud computing and image processing
CN113283401A
Multi-AGV scheduling method and system based on artificial intelligence and visual perception
CN113342004A
AGV cargo falling detection method and system based on artificial intelligence
CN113382203A
