Beverage robot path replanning method and system

Abstract

The present application relates to the technical field of path re-planning, in particular to a beverage robot path re-planning method and system, in the present application, the passing time and length of low-risk paths are accumulated by a graph neural network, and a generation value is generated in combination with the crowd density, the unstable paths are screened out to form a candidate set, so that the paths have adaptability to the crowd dynamics and passing environment in the global planning stage, the paths are reordered by an ant colony algorithm, the results are ensured to consider energy consumption and beverage stability, so that the robot can still maintain the efficiency and safety of the path in a multi-constraint environment, in the local search link, the path length, turning radius, channel width and customer gathering point are set as rejection conditions to avoid path failure caused by local narrowness and high-density area, improve local flexibility and environmental response speed, so that the beverage delivery can obtain path stability, energy consumption balance and task timeliness in a complex dynamic environment.

Description

Technical Field

[0001] This invention relates to the field of path replanning technology, and in particular to a method and system for path replanning of beverage robots. Background Technology

[0002] The field of path replanning technology involves a control method for autonomous mobile robots to correct predetermined motion paths based on dynamic environmental information during task execution. The core of this method is to generate new feasible paths in real time through algorithms when the robot encounters obstacles, environmental changes, or execution priority adjustments on a preset trajectory, so as to ensure task continuity and execution efficiency.

[0003] The beverage robot path replanning method is a control method that real-time corrects and replaces the robot's movement path in beverage delivery tasks. Its purpose is to solve the problem of unreachable original paths caused by environmental changes or obstacles during beverage delivery. By regenerating feasible paths, the robot can continue to complete the beverage delivery task, ensuring timely delivery and improving overall service efficiency. This achieves continuous delivery in dynamic environments, avoiding task failure or delays caused by path interruptions or repeated waiting. The beverage delivery robot consists of an autonomous mobile chassis and an Anno Robotics robotic arm module. The robotic arm performs beverage picking, delivery, and placement operations, while the mobile chassis handles path movement and obstacle avoidance. The control systems of both are synchronized via bus communication. When the path replanning algorithm generates a new travel path, the control system simultaneously adjusts the robotic arm's motion posture and work rhythm to ensure that the gripping, lifting, and placement actions of the robotic arm's end effector are dynamically coordinated with the chassis during path changes, thereby preventing beverage tilting, spillage, or gripping failure due to posture mismatch.

[0004] In existing technologies, path correction mechanisms rely more on single obstacle avoidance and dynamic triggering to generate paths, lacking a tiered and quantifiable approach to overall risk. This results in path adjustments often occurring after sudden obstacles or rapid environmental changes, exhibiting a lag. In densely populated areas, this can easily cause robots to frequently stall or wait along the path, increasing the probability of task interruption. Furthermore, existing technologies primarily use path feasibility as a single criterion, neglecting the balance between path length, number of turns, and energy consumption. This can lead to robots choosing passable but time-consuming or energy-intensive paths, increasing the risk of delivery delays and insufficient power. Existing methods often rely on single-point detours, lacking comprehensive consideration of turning radii, channel width, and customer crowding. This results in robots repeatedly invalidating paths in complex areas, increasing the number of times they have to wait and replan, and reducing service efficiency. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method and system for replanning the path of a beverage robot.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a beverage robot path replanning method, comprising the following steps:

[0007] S1: Based on location coordinates, corridor width, pedestrian density, and service area, the nodes are matched with the pallet positions and combined with the channel length and time calculation. The congestion risk and safety gap are compared and assigned values, sorted and mapped, and a path risk classification matrix is ​​generated.

[0008] S2: Based on the path risk classification matrix, a graph neural network is used to accumulate the length and time of low-risk channels and combine them with the flow density to generate a cost value. The preceding items are sorted and selected. The channel width and beverage stability are used to eliminate insufficient options and generate a set of low-risk candidate paths.

[0009] S3: Based on the low-risk candidate path set, the path length and number of turns are accumulated and combined with the estimated time for calculation. The stability of the beverage, the amount of electricity and the priority are compared and the inconsistent items are deleted. The ant colony algorithm is used to re-sort the set and obtain the multi-constraint effective path set.

[0010] S4: Based on the set of effective paths with multiple constraints, search the local area ahead, compare the path length with the maximum value to remove excessively long items, compare the turning radius with the threshold to remove insufficient items, and determine the channel width with the customer gathering point to remove unqualified items, and establish a combination of locally feasible paths.

[0011] S5: Based on the combination of locally feasible paths, determine and delete inconsistent items based on turning radius and safety distance, eliminate insufficient items based on comparison of passage width and beverage stability, delete excessive items based on energy consumption estimation and replenishment point determination, and output the first rank of total agency value to obtain the robot execution path.

[0012] As a further embodiment of the present invention, the path risk classification matrix includes node parameters, channel parameters, and risk level markers; the low-risk candidate path set includes channel cost, channel width, and stability results; the multi-constraint effective path set includes path length value, number of turns, power threshold, and task priority; the locally feasible path combination includes path branch length, turning radius, and channel width; and the robot execution path includes safety distance value, passage width value, and energy consumption path sequence.

[0013] As a further aspect of the present invention, the specific steps for generating the path risk classification matrix are as follows:

[0014] Based on location coordinates, corridor width, pedestrian density, and service area, the node location is compared and the relationship with the tray location is recorded. The channel length is added to the time value to generate a calculation table, and a node channel calculation table is generated.

[0015] Based on the node channel calculation table, the blocking risk value and the safety gap difference are compared and labeled with the level. The channel sequence mapping is used to establish a node-to-channel index and generate a path risk classification matrix.

[0016] As a further aspect of the present invention, the specific steps for generating the low-risk candidate path set are as follows:

[0017] Based on the path risk classification matrix, a graph neural network is used to add the channel length value and time value one by one and superimpose the flow density value into the summation result to generate a continuously accumulated cost value record table. Then, the channel cost value sequence is generated by dividing the group by threshold.

[0018] Based on the channel value sequence, the channel values ​​are rearranged in ascending order of numerical value and a set is formed by taking the first few items. Then, by comparing the channel width with the channels in the set one by one, items with insufficient values ​​are deleted to generate a candidate channel set.

[0019] Based on the candidate channel set, the stability value of the beverage tray is compared with the channel in the set one by one, and channels that do not meet the stability requirements are eliminated. The remaining channels are integrated into a set to generate a low-risk candidate path set.

[0020] As a further aspect of the present invention, the graph neural network performs message aggregation and updating hierarchically. In each layer, the source node vector and edge vector are weighted and merged into the target node to generate an incoming edge aggregation vector, guided by the edge. The node vector and the incoming edge aggregation vector are added and linearly mapped to obtain a new node vector. The residuals are retained across layers to stabilize the values. The edge side is jointly mapped with the updated source node and target node vectors and edge vectors to obtain the blocking probability and output the edge risk weight. Edges above the threshold are marked as disabled and the corresponding connections are cleared in the adjacency relationship. Edges below the threshold are written with risk weights into the edge attributes.

[0021] As a further aspect of the present invention, the specific steps for generating the multi-constraint effective path set are as follows:

[0022] Based on the set of low-risk candidate paths, the path length and the number of turns are counted and added together to obtain the cumulative path value. This value is then superimposed with the estimated time value to form a constraint comparison table, generating a path constraint value table.

[0023] Based on the path constraint value table, the stability of the beverage tray and the power threshold are compared with the path values ​​in the table one by one. Unqualified paths are deleted and the deletion index is recorded. The remaining paths are summarized to generate a set of filtered paths.

[0024] Based on the filtered path set, the ant colony algorithm is used to reorder the task priority values ​​from high to low and readjust the path combinations during the sorting process. Paths that meet the requirements are retained and the integrated set is output to obtain the multi-constraint effective path set.

[0025] As a further aspect of the present invention, the ant colony algorithm initializes individual ants based on the selected path set and randomly selects a starting path within the set. Each ant selects the next path according to the task priority weight and the path feasibility probability and gradually constructs a complete path sequence. The path selection probability is jointly determined by the local pheromone concentration and the heuristic value. After the construction is completed, the fitness is calculated based on the total generation value of the path and pheromone is released on the corresponding path edge. The pheromone is updated and evaporated in each iteration to avoid excessive concentration. The path sequences generated by different ants converge to the high fitness solution in multiple iterations, resulting in a multi-constraint effective path set optimized by priority constraints and global search.

[0026] As a further aspect of the present invention, the specific steps for generating the locally feasible path combination are as follows:

[0027] Based on the set of valid paths with multiple constraints, the paths in the local area ahead are expanded and numbered. After comparing the path length with the maximum limit value, the excess items are deleted, and a local path filtering table is generated.

[0028] Based on the local path filtering table, the turning radius is compared with the threshold to mark unqualified items, and the channel width is compared with the number of customer gathering points to eliminate insufficient items, thus establishing a combination of locally feasible paths.

[0029] As a further aspect of the present invention, the specific steps for generating the robot execution path are as follows:

[0030] Based on the combination of locally feasible paths, the turning radius and safety distance values ​​are calculated one by one and inconsistent items are deleted. The remaining paths are renumbered and reorganized to generate a set of safety verification paths.

[0031] Based on the safety verification path set, the passage width is compared with the stability of the beverage tray to eliminate deficiencies, the energy consumption estimate is compared with the replenishment point conditions to delete excess items, and the total agency value is accumulated and sorted to select the first one to obtain the robot execution path.

[0032] A beverage robot path replanning system, wherein the beverage robot path replanning system is used to execute the above-mentioned beverage robot path replanning method, the system comprising:

[0033] Environmental modeling module: Based on location coordinates, corridor width, pedestrian density, and service area, it matches nodes with tray positions and combines channel length and time values, compares and assigns values ​​for congestion risk and safety gap, maps channel order, and generates a path risk classification matrix.

[0034] Risk calculation module: Based on the path risk classification matrix, a graph neural network is used to aggregate channel features and output the blockage probability. The channel length is added to the time and the cost table is generated after the flow density is accumulated. Threshold comparison is used to divide the interval, sort and truncate the preceding items, and the channel width is compared with the stability of the beverage tray to delete the insufficient items, generating a set of low-risk candidate paths.

[0035] Path filtering module: Based on the low-risk candidate path set, the path length and the number of turns are added together and superimposed with the estimated time to generate an accumulation table. Stability and power threshold are compared one by one to delete unqualified items. Ant colony algorithm is used to iteratively search the remaining path set and rearrange them according to task priority to generate a multi-constraint effective path set.

[0036] Local retrieval module: Based on the set of effective paths with multiple constraints, the local area ahead is expanded, the path length is compared with the maximum value to delete excessively long items, the turning radius is compared with the threshold to delete insufficient items, and the channel width is compared with the customer gathering point to eliminate unqualified items, and a combination of locally feasible paths is established.

[0037] The execution output module: Based on the combination of locally feasible paths, the turning radius and safety distance are compared and the discrepancies are removed, the passage width and beverage tray stability are compared and the deficiencies are eliminated, the energy consumption estimate and replenishment point are compared and the excess items are eliminated, and the remaining paths are sorted and the first one is selected to obtain the robot execution path.

[0038] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0039] In this invention, the travel time and length of low-risk paths are accumulated through a graph neural network, and a cost value is generated by combining it with pedestrian density. After filtering out unstable paths, a candidate set is formed, so that the path has adaptability to pedestrian dynamics and travel environment in the global planning stage.

[0040] In this invention, the ant colony algorithm is used to reorder the results, ensuring that the results take into account both energy consumption and beverage stability. This allows the robot to maintain the efficiency and safety of the path even in a multi-constraint environment. In the local retrieval stage, the path length, turning radius, channel width, and customer gathering point are set as elimination conditions to avoid path failure caused by narrow and high-density areas, thereby improving local flexibility and environmental response speed.

[0041] In this invention, energy consumption estimation and replenishment point distribution are combined in the feasible path combination to ensure the robot's endurance during the delivery process. The optimal execution path is obtained by ranking the total agent value. With the help of a progressive chain of local screening and energy consumption control, beverage delivery can simultaneously achieve path stability, energy consumption balance and task timeliness in a complex and dynamic environment. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the workflow of the present invention;

[0043] Figure 2 This is a system flowchart of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0045] Example 1

[0046] Please see Figure 1 This invention provides a technical solution: a path replanning method for beverage robots, comprising the following steps:

[0047] S1: Based on location coordinates, corridor width, pedestrian density, and service area, the nodes are matched with the pallet positions and combined with the channel length and time calculation. The congestion risk and safety gap are compared and assigned values, sorted and mapped, and a path risk classification matrix is ​​generated.

[0048] S2: Based on the path risk classification matrix, a graph neural network is used to accumulate the length and time of low-risk channels and combine them with the flow density to generate a cost value. The preceding items are sorted and selected. The channel width and beverage stability are used to eliminate insufficient options and generate a set of low-risk candidate paths.

[0049] S3: Based on the low-risk candidate path set, the path length and number of turns are accumulated and combined with the estimated time for calculation. The stability of the beverage, the amount of electricity and the priority are compared and the inconsistent items are deleted. The ant colony algorithm is used to re-sort the set and obtain the multi-constraint effective path set.

[0050] S4: Based on the set of effective paths with multiple constraints, search the local area ahead, compare the path length with the maximum value to remove excessively long items, compare the turning radius with the threshold to remove insufficient items, and determine the channel width with the customer gathering point to filter out unqualified items, and establish a combination of locally feasible paths.

[0051] S5: Based on the combination of locally feasible paths, the turning radius and safety distance are used to determine and delete inconsistent items, the passage width and beverage stability are compared to eliminate insufficient items, the energy consumption is estimated and the replenishment point is determined to delete excessive items, the total value is sorted and the first item is output to obtain the robot execution path.

[0052] The path risk classification matrix includes node parameters, channel parameters, and risk level labels. The low-risk candidate path set includes channel cost, channel width, and stability results. The multi-constraint effective path set includes path length, number of turns, power threshold, and task priority. The locally feasible path combination includes path branch length, turning radius, and channel width. The robot execution path includes safety distance value, passage width value, and energy consumption path sequence.

[0053] The specific steps for generating the path risk classification matrix are as follows:

[0054] Based on location coordinates, corridor width, pedestrian density, and service area, the node location is compared and the relationship with the tray location is recorded. The channel length is added to the time value to generate a calculation table, and a node channel calculation table is generated.

[0055] Based on the node channel calculation table, the blocking risk value is compared with the safety gap difference and the level is marked. The channel sequence mapping is used to establish the node to channel index and generate the path risk classification matrix.

[0056] Based on location coordinates, corridor width, pedestrian density, and service area, the Dijkstra algorithm is adopted. During the execution process, the node position and the tray position are compared. Specifically, the input parameters are the node set V and the edge set E. The starting node number is set to s, the distance value of all nodes is initialized to infinity, the distance value of the starting node s is 0, the priority queue is called to select nodes, and a comparison operation is performed on each edge. The current node distance is compared with the cumulative value of the edge length plus the passage time. If the cumulative value is less than the target node record value, the target node distance value is updated and the predecessor node number is recorded. The operation is repeated until all nodes have been traversed, and finally a node channel calculation table is generated.

[0057] Based on the node channel calculation table, the K-means clustering algorithm is used. During the execution process, the blocking risk value and the safety gap difference are compared. Specifically, the input dataset X is the combination of risk value and gap difference for all nodes. The number of clusters k is set to 3, and the cluster center points are initialized as three sets of difference sample points. Iterative calculation is performed. For each data sample, the distance to the three sets of cluster center points is calculated in turn, and the sample is assigned to the nearest cluster center. After one assignment, the cluster center point is recalculated as the average of all samples. The iteration is repeated until all cluster center points are stable and no longer updated. After the iteration is completed, the risk level of each channel is labeled according to the cluster label, and the sequential index from node to channel is established. Finally, a path risk classification matrix is ​​generated.

[0058] The specific steps for generating a low-risk candidate path set are as follows:

[0059] Based on the path risk classification matrix, a graph neural network is used to add the channel length value and time value one by one and superimpose the flow density value into the summation result to generate a continuously accumulated cost value record table. Then, the channel cost value sequence is generated by dividing the group by threshold.

[0060] Based on the channel value sequence, the channel values ​​are rearranged in ascending order of numerical value and a set is formed by taking the first few items. Then, by comparing the channel width with the channels in the set one by one, items with insufficient values ​​are deleted to generate a candidate channel set.

[0061] Based on the candidate channel set, the stability value of the beverage tray is compared with the channel in the set one by one, and channels that do not meet the stability requirements are eliminated. The remaining channels are integrated into a set to generate a low-risk candidate path set.

[0062] Based on the path risk classification matrix, a graph neural network is used. During execution, the channel length and time values ​​are added one by one. The input node feature matrix contains the channel length and time values, and the adjacency matrix describes the channel connection relationship. The weight matrix in the convolutional layer is set to a dimension of 64×64, and the activation function parameter is set to ReLU. During propagation, the sum of the channel length and time values ​​is used as the basic feature, and the pedestrian density value is superimposed on the sum. The number of iteration layers is set to 3, and the output of each layer is normalized. The output layer dimension is fixed at 128. In the last iteration, the cost value of all nodes is recorded, and a threshold division operation is called. The threshold intervals are set as follows: T1 is 0 to 30, T2 is 31 to 60, and T3 is 61 to 100. The cost values ​​are grouped according to the interval division, and finally a continuously accumulated cost value record table is generated, and a channel cost value sequence is generated.

[0063] Based on the channel value sequence, a quicksort algorithm is used. During the execution process, the channel values ​​are rearranged in ascending order. The input sequence length parameter is n, and the middle value is selected as the pivot. When performing comparison operations, channels smaller than the pivot are swapped to the left, and channels larger than the pivot are swapped to the right. This process is repeated until the sequence is completely ordered. After sorting, a number of preceding items are taken, with the number of preceding items set to 10, to form an initial set C. The set C is then compared one by one, and the width value of each channel is compared with a set width threshold Wmin. When the channel width is less than Wmin, the channel is deleted, and finally, a candidate channel set is generated.

[0064] Based on the candidate channel set, a Boolean conditional filtering method is adopted. During the execution process, the stability value of the beverage tray is compared with the channel in the set one by one. The stability threshold Smin is set to 0.8. Each channel in set C is traversed. If the stability value of the channel is less than Smin, the deletion operation is performed. If it is greater than or equal to Smin, it is retained. After the traversal is completed, the remaining channels are integrated into a set, and finally a low-risk candidate path set is generated.

[0065] The graph neural network performs message aggregation and updates hierarchically. In each layer, the source node vector and edge vector are weighted and fed into the target node to generate the incoming edge aggregation vector. The node vector and the incoming edge aggregation vector are added and linearly mapped to obtain the new node vector. The residuals are retained across layers to stabilize the values. The edge side is jointly mapped with the updated source node and target node vectors and edge vectors to obtain the blocking probability and output the edge risk weight. Edges above the threshold are marked as disabled and the corresponding connections are cleared in the adjacency relationship. Edges below the threshold are written to the edge attributes with risk weight.

[0066] Graph neural networks, according to the formula:

[0067]

[0068] Where: A represents the adjacency matrix of the beverage robot's movement path, and I represents the identity matrix. Indicates by The resulting degree matrix consists of L representing the channel length vector and T representing the travel time vector. Represents a numerical vector of pedestrian density. Indicates the length weighting coefficient. Indicates the time weighting factor. This represents the weighting coefficient for pedestrian density. This represents the risk gain correction factor. This represents the gain correction factor for the time period. Indicates the density persistence correction factor. This represents a trainable weight matrix. Represents a non-linear activation function. H represents the path risk gain matrix, and H represents the time period gain vector. This indicates the duration of pedestrian density.

[0069] Execution process: First, construct the adjacency matrix A of the channel topology and add the identity matrix I, then calculate the degree matrix. And normalization yields the propagation kernel Then, the numerical vectors of channel length L and passage time T are collected and added one by one. The numerical vector of pedestrian density is then superimposed on the sum. Then through weighting coefficients , , The basic cost characteristics are formed, and then the path risk gain matrix is ​​introduced. Element-wise multiplication with length vector L to adjust the length cost of high-risk channels and by coefficients The intensity of the control is controlled by multiplying the time-period gain vector H by the time vector T element-wise to reflect the surge impact during high-risk periods, and then using coefficients. Adjust the weights, then the population density numerical vector Calculate the persistence item The cumulative effect of density on the path segment is represented by coefficients. Subsequently, multiple features are combined into a single input and multiplied by the trainable weight matrix w. Feature transfer in the graph structure is completed under the action of the propagation kernel and then passed through a nonlinear activation function. The cost vector z is obtained, and finally, the cost vector z is divided into groups by setting a threshold to generate the cost sequence of the beverage robot channel for path replanning.

[0070] The specific steps for generating a set of valid paths with multiple constraints are as follows:

[0071] Based on the low-risk candidate path set, the path length and the number of turns are counted and added together to obtain the path cumulative value. Then, the path is superimposed with the expected time value to form a constraint comparison table, generating a path constraint value table.

[0072] Based on the path constraint value table, the stability of the beverage tray and the power threshold are compared with the path values ​​in the table one by one. Unqualified paths are deleted and the deletion index is recorded. The remaining paths are summarized to generate a set of filtered paths.

[0073] Based on the filtered path set, the ant colony algorithm is used to reorder the task priority values ​​from high to low and readjust the path combinations during the sorting process. Paths that meet the requirements are retained and the integrated set is output to obtain the multi-constraint effective path set.

[0074] Based on a low-risk candidate path set, a statistical and cumulative method is used to accumulate the path length point by point during execution. The input parameter is a sequence of path nodes, and the distance between each node is recorded in meters. The total path length is obtained by summing the values ​​one by one. At the same time, the number of turns in the path is counted item by item, with each change of direction counted as 1. After the statistics are completed, the total path length and the total number of turns are added to form the path cumulative value. Then, the path cumulative value is added to the estimated time value, which is input in seconds. The cumulative result is recorded in a table, and finally, a path constraint value table is generated.

[0075] Based on the path constraint value table, a conditional comparison method is used to compare the stability value of the beverage tray, the power threshold, and the path constraint value one by one during the execution process. The stability threshold is set to 0.8 and the power threshold is set to 20. Each path record in the path constraint value table is traversed. When the stability value of the path is less than 0.8 or the power value is less than 20, a deletion operation is performed, and the index position of the path is recorded when it is deleted. After the traversal is completed, the remaining paths are summarized and finally a set of filtered paths is generated.

[0076] Based on the selected path set, an ant colony algorithm is used. During execution, the task priority value is input, with the priority value ranging from 1 to 10. The larger the value, the higher the priority. The initial ant colony size is 50, the initial pheromone concentration is set to 0.1, the volatile factor is set to 0.5, and the number of iterations is set to 100. When each ant selects a path during the iteration, the probability transition rule is invoked, and the paths are sorted and adjusted from high to low according to the priority value. During the sorting process, the path combinations are readjusted. If the path combination meets the input constraints, it is retained; otherwise, the path combination is discarded. After all ant iterations are completed, the retained path set is integrated to finally generate a multi-constraint valid path set.

[0077] The ant colony algorithm initializes individual ants based on a set of selected paths and randomly selects a starting path within the set. Each ant selects the next path according to the task priority weight and the path feasibility probability and gradually constructs a complete path sequence. The path selection probability is determined by the local pheromone concentration and the heuristic value. After the construction is completed, the fitness is calculated based on the total generation value of the path and pheromone is released on the corresponding path edge. The pheromone is updated and evaporated in each iteration to avoid over-concentration. The path sequences generated by different ants converge to the high fitness solution in multiple iterations, and a set of multi-constraint effective paths optimized by priority constraints and global search is obtained.

[0078] Ant colony algorithm, according to the formula:

[0079]

[0080] in: This represents the probability of the k-th ant moving from node i to node j. This represents the pheromone concentration on the channel from node i to node j. Indicates the channel distance Travel time Resource utilization rate Heuristic functions, This represents the numerical value of the channel distance from node i to node j. This represents the channel time value from node i to node j. This represents the resource usage values ​​from node i to node j. This represents the priority value of the task corresponding to node j. This is an indicator function that indicates whether channels i to j satisfy multiple constraints; it takes the value 1 if feasible and 0 otherwise. This represents the pheromone weighting coefficient. Represents the weight coefficients of the heuristic function. This represents the task priority weighting coefficient. Let represent the set of feasible neighbor nodes of the k-th ant among the nodes;

[0081] Execution process: First, initialize the channel topology and set the pheromone concentration. Take initial values ​​on the edges connecting each node, and then determine the values ​​based on the distance between the channels between nodes. Travel time With resource utilization Computing heuristic functions Next, obtain the task priority value. This is introduced as a node attribute. Based on this, for each ant k, when it is located at node i, the set of feasible neighbors is calculated. For each candidate node j in the set, calculate the selection probability. The probability calculation integrates three types of factors, including the pheromone item. Reflecting the accumulation of historical paths, heuristic items Reflecting the impact of channel characteristics, priority items Reflecting the task sequence requirements, the feasibility indicator function is finally used as a constraint. The algorithm ensures that the path meets multiple constraints and uses a normalized denominator to ensure that the sum of the probabilities of all candidate paths is 1. Then, the algorithm selects paths and updates the pheromone concentration based on the probability distribution. After all ants have completed the iteration, the algorithm selects the set of paths with high probability and that meet the task priority constraints and outputs the integrated set. Finally, the algorithm obtains the set of effective paths for the beverage robot with multiple constraints for replanning.

[0082] The specific steps for generating a combination of locally feasible paths are as follows:

[0083] Based on the set of effective paths with multiple constraints, the paths in the local area ahead are expanded and numbered. After comparing the path length with the maximum limit value, the excess items are deleted, and a local path filtering table is generated.

[0084] Based on the local path screening table, the turning radius is compared with the threshold to mark unqualified items, and the channel width is compared with the number of customer gathering points to eliminate insufficient items, thus establishing a combination of locally feasible paths.

[0085] Based on a set of valid paths with multiple constraints, a depth-first traversal method is adopted. During the execution process, the path in the preceding local area is expanded. The input parameter is a set of path nodes. Each node is recursively traversed and assigned a number. The numbering rule is to increment sequentially from the starting node. The path length is recorded during the traversal. The path length is accumulated segment by segment in meters. A comparison operation is called to compare the length of each path with the maximum limit value Lmax. When the path length is greater than Lmax, a deletion operation is performed and the index position is recorded. After the deletion is completed, the remaining path numbers and lengths are reorganized and output to finally generate a local path filtering table.

[0086] Based on the local path filtering table, a conditional comparison method is used. During the execution process, the turning radius value is compared with the threshold Rmin one by one. When the turning radius is less than Rmin, it is marked as an unqualified item. Then, the channel width value is compared with the number of customer gathering points. The minimum channel width Wmin and the maximum number of customer gathering points Pmax are input. Each path in the filtering table is traversed. When the channel width is less than Wmin or the number of customer gathering points is greater than Pmax, a deletion operation is performed. After deletion, the remaining paths are integrated to finally establish a combination of locally feasible paths.

[0087] The specific steps for generating the robot's execution path are as follows:

[0088] Based on the combination of locally feasible paths, the turning radius and safety distance values ​​are calculated one by one and inconsistent items are deleted. The remaining paths are renumbered and reorganized to generate a set of safety verification paths.

[0089] Based on the safety verification path set, the passage width is compared with the stability of the beverage tray to eliminate deficiencies, the energy consumption is estimated and the replenishment point conditions are compared to delete excess items, and the total agency value is accumulated and sorted to select the first one to obtain the robot execution path;

[0090] Based on the combination of locally feasible paths, a numerical comparison method is adopted. During the execution process, the turning radius and safety distance are calculated one by one. The input parameters include path number, turning radius R, and corresponding safety distance S. When traversing the path set, the values ​​of R and S are read one by one. When the comparison operation is called, the deletion operation is performed when R is less than the set threshold Rmin or S is less than the set threshold Smin. The deleted path number is recorded in the index table. After the traversal is completed, the remaining paths are renumbered and reorganized. The numbering rule is to start from 1 and increment sequentially. Finally, a safety verification path set is generated.

[0091] Based on the safety verification path set, a conditional filtering and cumulative sorting method is adopted. During the execution process, the passage width value and the beverage tray stability value are compared one by one. The passage width threshold Wmin and the tray stability threshold Stmin are input. If the width of a certain path is less than Wmin or the tray stability is less than Stmin, a rejection operation is performed. At the same time, the energy consumption value is estimated. The energy consumption calculation parameters include path length L, current I, and voltage U. The energy consumption value E is calculated according to the value of L×I×U. Then, the energy consumption value is compared with the replenishment point conditions one by one. When the energy consumption is greater than the replenishment capacity of the replenishment point, a deletion operation is performed. In the remaining path set, the cumulative sorting operation is called. The total cost value is obtained by accumulating the path cost value. The total cost value is sorted from largest to smallest, and the first path in the sorting result is selected to finally obtain the robot execution path.

[0092] A beverage robot path replanning system is used to execute the aforementioned beverage robot path replanning method. The system includes:

[0093] Environmental modeling module: Based on location coordinates, corridor width, pedestrian density, and service area, it matches nodes with tray positions and combines channel length and time values, compares and assigns values ​​for congestion risk and safety gap, maps channel order, and generates a path risk classification matrix.

[0094] Risk calculation module: Based on the path risk classification matrix, a graph neural network is used to aggregate channel features and output the blockage probability. The channel length is added to the time and the cost table is generated after the flow density is accumulated. Threshold comparison is used to divide the interval, sort and truncate the preceding items, and the channel width is compared with the stability of the beverage tray to delete the insufficient items and generate a set of low-risk candidate paths.

[0095] Path selection module: Based on a low-risk candidate path set, the path length and number of turns are added together and superimposed with the estimated time to generate a cumulative table. Stability and power threshold are compared one by one to delete unqualified items. The ant colony algorithm is used to iteratively search the remaining path set and rearrange them according to task priority to generate a multi-constraint effective path set.

[0096] Local retrieval module: Based on the set of effective paths with multiple constraints, the local area in front is expanded, the path length is compared with the maximum value to delete excessively long items, the turning radius is compared with the threshold to delete insufficient items, and the channel width is compared with the customer gathering point to eliminate unqualified items, and a combination of locally feasible paths is established.

[0097] The execution output module: Based on the combination of locally feasible paths, the turning radius and safety distance are compared and the inconsistencies are removed. The passage width and the stability of the beverage tray are compared and the deficiencies are removed. The energy consumption is estimated and the replenishment point is compared and the excess items are removed. The remaining paths are sorted and the first one is selected to obtain the robot execution path.

[0098] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for re-planning a route for a beverage robot, characterized in that, The method comprises the following steps: S1: based on the position coordinates, the corridor width, the crowd density, and the service area, the node is matched with the tray position, the risk of blockage is compared with the safety gap, the order is mapped, and the path risk grading matrix is generated; S2: based on the path risk grading matrix, the graph neural network is used to accumulate the length and time of the low-risk path, combine the crowd density to generate the value of the generation, select the front sequence, and determine the width of the path and the stability of the beverage to remove the insufficient items, generate a low-risk candidate path set; S3: based on the low-risk candidate path set, the path length and the number of turns are accumulated, combined with the expected time calculation, the beverage stability, the power and the priority are compared to delete the inconsistent items, the ant colony algorithm is used, the set is reordered to obtain a multi-constrained effective path set; S4: based on the multi-constrained effective path set, the local area in front is expanded to search, the path length is compared with the maximum value to remove the too long items, the turning radius is compared with the threshold to remove the insufficient items, the path width is compared with the customer gathering point to remove the unqualified items, and the local feasible path combination is established; S5: based on the local feasible path combination, the turning radius and the safety distance are determined to delete the inconsistent items, the path width and the beverage stability are compared to remove the insufficient items, the energy consumption estimation and the supply point are determined to delete the items exceeding the limit, the total generation value is sorted to output the first, and the robot execution path is obtained.

2. The drink bot path re-planning method of claim 1, wherein, The path risk grading matrix includes node parameters, channel parameters, and risk level markers, the low-risk candidate path set includes channel generation value, channel width, and stability result, the multi-constrained effective path set includes path length value, number of turns, power threshold, and task priority, the local feasible path combination includes path branch length, turning radius, and path width, and the robot execution path includes safety distance value, path width value, and energy consumption path sequence.

3. The drink bot path re-planning method of claim 1, wherein, The specific steps for generating the path risk grading matrix are: Based on the position coordinates, the corridor width, the crowd density, and the service area, the node position is compared with the tray position, the relationship is recorded, the channel length and time values are added to generate a calculation table, and the node channel calculation table is generated; Based on the node channel calculation table, the blockage risk value is compared with the safety gap difference value and marked with the level, the channel order is mapped to establish the node to channel index, and the path risk grading matrix is generated.

4. The drink bot path re-planning method of claim 1, wherein, The specific steps for generating the low-risk candidate path set are: Based on the path risk grading matrix, the graph neural network is used to add the channel length value and the time value one by one and superimpose the crowd density value in the addition result to generate a record table of continuous addition, and the channel generation value sequence is generated by threshold division grouping; Based on the channel generation value sequence, the channel generation values are rearranged from small to large in value and the first several items are formed into a set, and the channel width is compared with the channels in the set to delete the insufficient items one by one, and the candidate channel set is generated; Based on the candidate channel set, the beverage tray stability value is compared with the channels in the set one by one to delete the channels that do not meet the stability requirements, and the remaining channels are integrated into a set to generate a low-risk candidate path set.

5. The drink bot path re-planning method of claim 4, wherein, The graph neural network performs message aggregation and update by layers, in each layer, the source node vector and the edge vector are weighted and merged into the target node to generate an edge aggregation vector, the node vector and the edge aggregation vector are added and linearly mapped to obtain a new node vector, a residual is reserved across layers to stabilize the numerical value, the updated source node and target node vectors and the edge vector are jointly mapped to obtain a blocking probability and output an edge risk weight, edges above the threshold are marked as disabled and the corresponding connection is removed in the adjacency relationship, and edges below the threshold are written into the edge attribute with the risk weight.

6. The drink bot path re-planning method of claim 1, wherein, The specific steps for generating the multi-constraint effective path set are: Based on the low-risk candidate path set, the path length and the number of turns are counted and added to obtain a path cumulative value, which is then added to the estimated time value one by one to form a constraint comparison table, and a path constraint value table is generated; Based on the path constraint value table, the beverage tray stability, the power threshold, and the path values in the table are compared one by one, the unqualified paths are deleted and the deletion index is recorded, and the remaining paths are uniformly summarized to generate a filtered path set; Based on the filtered path set, an ant colony algorithm is used to reorder the task priority values from high to low and adjust the path combination in the sorting, retain the paths that meet the requirements, and output the integrated set to obtain the multi-constraint effective path set.

7. The drink bot path re-planning method of claim 6, wherein, The ant colony algorithm initializes the ant individuals based on the filtered path set and randomly selects a starting path in the set, each ant selects the next path according to the task priority weight and path feasibility probability and gradually constructs a complete path sequence, the path selection probability is determined by the local pheromone concentration and the heuristic value, after the construction is completed, the fitness is calculated according to the total generation value of the path, and the pheromone is released on the corresponding path edge, the pheromone is updated and evaporated with iteration to avoid excessive concentration, and the path sequences generated by different ants converge to high fitness solutions in multiple iterations, obtaining the multi-constraint effective path set optimized by priority constraint and global search.

8. The drink bot path re-planning method of claim 1, wherein, The specific steps for generating the local feasible path combination are: Based on the multi-constraint effective path set, the paths in the local area are expanded and numbered, and the paths exceeding the maximum limit value are deleted after comparing the path length with the maximum limit value, to generate a local path screening table; Based on the local path screening table, the unqualified items are marked by comparing the turning radius with the threshold value, and the insufficient items are removed by comparing the channel width with the number of customer gathering points, to establish a local feasible path combination.

9. The drink bot path re-planning method of claim 1, wherein, The specific steps for generating the robot execution path are: Based on the local feasible path combination, the turning radius and the safety distance value are calculated and the inconsistent items are deleted, the remaining paths are renumbered and arranged to generate a safety check path set; Based on the safety check path set, the insufficient items are removed by comparing the passage width with the beverage tray stability, the over-limit items are deleted by comparing the energy consumption estimation with the supply point conditions, the total generation value is added and sorted to select the first one, to obtain the robot execution path.

10. A drink bot path replanning system, characterized in that, The beverage robot path re-planning method according to any one of claims 1-9, the system comprises: An environment modeling module: based on the position coordinates, the corridor width, the people flow density, and the service area, the node is matched with the tray position, combined with the channel length and time value, the blocking risk and safety gap comparison assignment, the channel order mapping, and the path risk grading matrix is generated; A risk calculation module: based on the path risk grading matrix, the channel features are aggregated by using a graph neural network and the blocking probability is output, the channel length and time are added and the generation value table is generated after the people flow density is accumulated, the threshold value is compared to divide the interval, the order is intercepted to obtain the previous item, the channel width and the beverage tray stability are compared to delete the insufficient items, and a low-risk candidate path set is generated; A path screening module: based on the low-risk candidate path set, the path length and the number of turns are added and the expected time is superimposed to generate a cumulative table, the stability and the power threshold are compared one by one to delete unqualified items, an ant colony algorithm is used to iteratively search in the remaining path set and the tasks are rearranged according to the priority, and a multi-constrained effective path set is generated; A local search module: based on the multi-constrained effective path set, the local area in front is expanded, the path length is compared with the maximum value to delete the too long items, the turning radius is compared with the threshold to delete the insufficient items, the channel width is compared with the customer gathering point to delete the unqualified items, and the local feasible path combination is established; An execution output module: based on the local feasible path combination, the turning radius is compared with the safety distance to delete the inconsistent items, the passing width is compared with the beverage tray stability to delete the insufficient items, the energy consumption estimation is compared with the supply point to delete the over-limit items, the remaining paths are sorted and selected as the first, and the robot execution path is obtained.

Images