Coordinated scheduling planning method and device, computer device and storage medium

Abstract

The application relates to the technical field of intelligent scheduling, and discloses a cooperative scheduling planning method and device, computer equipment and a storage medium, the method comprises the following steps: acquiring planning input data of a current scheduling period, the data containing a robot request set, stability state information of the robot request set and the number of unfinished tasks left over from the last period; selecting a target solving time limit of the current period from a preset time limit according to the stability state and the number of left over tasks; expanding an initial node of a target search model to obtain a candidate node within the time limit, performing multistage pruning on the candidate node in a preset order to obtain an effective candidate node, performing cost weighted evaluation on the effective candidate node through a node evaluation function, updating a global optimal solution, and obtaining a search result until the time limit is reached; and generating a cooperative scheduling scheme according to the search result. The application solves the problem that the existing scheduling method lacks self-adaptive capability for period states and is difficult to balance the solving quality and real-time performance.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling, and more specifically to collaborative scheduling planning methods, devices, computer equipment, and storage media. Background Technology

[0002] With the rapid development of applications such as intelligent warehousing, building delivery, hospital logistics, and on-site transportation, the demand for multi-robot collaborative operations in multi-floor environments is increasing. In these scenarios, the collaborative scheduling of robots and elevators has become a key factor restricting system efficiency. However, existing scheduling methods generally use fixed or single computation time limits, failing to fully consider the dynamic changes in the request set within different scheduling cycles and the impact of legacy tasks from the previous cycle. When requests fluctuate frequently or there are a large number of unfinished tasks, fixed time limits may lead to insufficient searching and low solution quality; while when requests are stable and there are few legacy tasks, excessively long time limits will waste computational resources and affect system response speed. This mechanism, lacking the ability to adapt to scheduling states, struggles to strike a balance between ensuring solution quality and maintaining real-time performance, limiting the engineering practicality of multi-robot, multi-elevator collaborative scheduling systems in dynamic scenarios. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide a collaborative scheduling planning method, apparatus, computer equipment, and storage medium to solve the problem that existing scheduling methods lack the ability to adapt to periodic states and are difficult to balance solution quality and real-time performance.

[0004] In a first aspect, embodiments of the present invention provide a collaborative scheduling planning method, the method comprising: Obtain the planning input data for the current scheduling cycle, wherein the planning input data includes: robot request set, stability state information corresponding to the robot request set, and the number of unfinished tasks left over from the previous scheduling cycle; Based on the stability state information and the number of unfinished tasks, the target solution time limit for the current scheduling period is selected from the preset solution time limits; Within the target solution time limit, the initial nodes of the target search model are expanded to obtain candidate nodes. Multi-level pruning is performed on the candidate nodes in a preset order to obtain effective candidate nodes. Each effective candidate node is evaluated by cost weighting through a node evaluation function. The global optimal solution is updated according to the evaluation results until the target solution time limit is reached to obtain the search result. The target search model is used to search and solve the collaborative scheduling process between the robot and the elevator. The initial nodes and the candidate nodes correspond to collaborative scheduling schemes with different execution progress. Generate a collaborative scheduling scheme corresponding to the set of robot requests based on the search results.

[0005] Furthermore, before expanding the initial nodes of the target search model to obtain candidate nodes, the method further includes: Initialize the set of incomplete robot tasks based on the robot request set; Obtain the elevator time axis initialization parameters and the robot time axis initialization parameters; An elevator timeline is constructed based on the elevator timeline initialization parameters, and a robot timeline is constructed based on the robot timeline initialization parameters. Based on the elevator timeline, the robot timeline, and the set of unfinished robot tasks, an initial node for the target search model is generated, wherein the initial node corresponds to an initial scheduling scheme in which no robot requests have been executed.

[0006] Furthermore, the step of performing multi-level pruning on the candidate nodes according to a preset order to obtain valid candidate nodes includes: Obtain the node cost of the candidate node and the cost of the currently recorded known optimal solution; Perform the first level of pruning: when the node cost of the candidate node is greater than a preset relative multiple of the cost of the known optimal solution, terminate the search in the search queue where the candidate node is located; If the candidate node passes the first-level pruning, then the second-level pruning is performed: when the node cost of the candidate node or its child node is greater than the cost of the known optimal solution, the corresponding branch is skipped; If the candidate node passes the second-level pruning, then the third-level pruning is performed: obtain the extension action corresponding to the candidate node, and skip the corresponding extension when the extension action does not meet the first-step constraint in the preset feasible solution cache library. The preset feasible solution cache library is generated after path pre-calculation based on the elevator reachable area set and the area connectivity relationship. The first-step constraint is that the first step operation of the extension action is located in the preset feasible solution cache library. If the candidate node passes the third-level pruning, then the fourth-level pruning is performed: the access region record corresponding to the candidate node is obtained, and when the access region record contains a repeatedly accessed intermediate region, the corresponding expansion is skipped; The nodes retained after the first-level pruning, the second-level pruning, the third-level pruning, and the fourth-level pruning are considered as the valid candidate nodes.

[0007] Furthermore, after obtaining valid candidate nodes, the method further includes: Obtain the action to be executed corresponding to the valid candidate node; Based on the pre-built elevator timeline and robot timeline, the corresponding target elevator and target robot are determined according to the action to be executed; Obtain the cumulative elevator time on the elevator timeline and the cumulative robot time on the robot timeline for the target elevator; Based on the cumulative time of the elevator, the cumulative time of the robot, and the cost of each stage corresponding to the action to be executed, the single-step completion time of the action to be executed is calculated, and new valid candidate nodes are generated based on the single-step completion time.

[0008] Furthermore, the step of performing a cost-weighted evaluation on each of the effective candidate nodes using a node evaluation function, and updating the global optimal solution based on the evaluation results, includes: Obtain the cumulative cost of the effective candidate nodes, and perform heuristic cost estimation calculation on the remaining requests corresponding to the effective candidate nodes to obtain the heuristic estimated cost; The cumulative cost and the heuristically estimated cost are weighted and summed to obtain the node cost of the effective candidate node; When the node cost of the effective candidate node is better than the cost of the currently recorded known optimal solution, the effective candidate node is updated to the global optimal solution.

[0009] Furthermore, the step of selecting the target solution time limit for the current scheduling period from the preset solution time limits based on the stability state information and the number of unfinished tasks includes: When the stability state information meets the preset stability condition and the number of unfinished tasks is less than the preset number threshold, the first preset solution time limit is selected from the preset solution time limits as the target solution time limit; When the stability state information does not meet the preset stability condition or the number of unfinished tasks is greater than or equal to the preset number threshold, a second preset solution time limit is selected from the preset solution time limits as the target solution time limit, wherein the second preset solution time limit is greater than the first preset solution time limit.

[0010] Furthermore, the step of generating a collaborative scheduling scheme corresponding to the robot request set based on the search results includes: Detect whether the globally optimal solution in the search results covers all robot requests in the robot request set; If the global optimal solution covers all robot requests in the robot request set, then the global optimal solution is output as the cooperative scheduling scheme. Alternatively, if the global optimal solution does not cover all robot requests in the robot request set, then the intermediate candidate node that covers the most robot requests in the robot request set is selected from the candidate nodes visited during the search process, and the local optimal solution corresponding to the intermediate candidate node is output as the cooperative scheduling scheme.

[0011] Furthermore, after obtaining the planning input data for the current scheduling period, the method further includes: Traverse the robot's candidate forward point set and obtain the real-time path cost corresponding to each candidate forward point in the candidate forward point set; The real-time path costs corresponding to each candidate forward point are compared to obtain the comparison results. Based on the comparison results, the target forward point with the best cost within the current scheduling period is selected from the candidate forward point set; The target precursor is integrated into the planning input data of the current scheduling cycle, and subsequent search planning is performed based on the target precursor.

[0012] Furthermore, after generating a collaborative scheduling scheme corresponding to the robot request set based on the search results, the method further includes: When the stability state information of the robot request set meets the preset stability conditions, the current revaluation cost corresponding to the collaborative scheduling scheme and the historical revaluation cost corresponding to the previous historical scheduling scheme are obtained. Calculate the cost deviation between the current revaluation cost and the historical revaluation cost; When the cost deviation value meets the preset cost deviation condition, the previous round of historical scheduling scheme is used as the target collaborative scheduling scheme for the current scheduling cycle.

[0013] Secondly, embodiments of the present invention provide a collaborative scheduling and planning apparatus, the apparatus comprising: The acquisition module is used to acquire the planning input data for the current scheduling cycle, wherein the planning input data includes: robot request set, stability state information corresponding to the robot request set, and the number of unfinished tasks left over from the previous scheduling cycle; The selection module is used to select the target solution time limit for the current scheduling period from the preset solution time limits based on the stability state information and the number of unfinished tasks. The solution module is used to expand the initial nodes of the target search model within the target solution time limit to obtain candidate nodes, perform multi-level pruning on the candidate nodes in a preset order to obtain effective candidate nodes, perform cost-weighted evaluation on each of the effective candidate nodes through a node evaluation function, update the global optimal solution according to the evaluation results, until the target solution time limit is reached to obtain the search result. The target search model is used to search and solve the collaborative scheduling process between the robot and the elevator, and the initial nodes and the candidate nodes correspond to collaborative scheduling schemes with different execution progress. The generation module is used to generate a collaborative scheduling scheme corresponding to the robot request set based on the search results.

[0014] Thirdly, embodiments of the present invention provide a computer device, including: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method described in the first aspect or any corresponding embodiment thereof.

[0015] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing computer instructions that cause a computer to perform the method described in the first aspect or any of its corresponding embodiments.

[0016] The method provided in this application has the following beneficial effects: The method provided in this application, by acquiring planning input data including robot request set, stability state information, and the number of unfinished tasks from the previous cycle, can comprehensively perceive the dynamic state of the current scheduling cycle, providing an accurate data foundation for subsequent adaptive decision-making. By selecting a target solution time limit from preset solution time limits based on stability state information and the number of unfinished tasks, the search time budget can be matched with the actual scheduling needs of the current cycle, avoiding resource waste or insufficient search caused by fixed time limits. By expanding the initial nodes of the target search model within the target solution time limit to obtain candidate nodes, the robot-elevator collaboration can be systematically traversed. The state space of the scheduling scheme obtains effective candidate nodes by performing multi-level pruning on candidate nodes in a preset order. This can significantly reduce the number of search branches and reduce computational overhead without losing the optimal solution. The node evaluation function performs cost-weighted evaluation on each effective candidate node and updates the global optimal solution based on the evaluation results until the time limit is reached. This can continuously approach the global optimal scheduling scheme within a finite time and balance the solution accuracy and real-time performance. By generating a collaborative scheduling scheme corresponding to the robot request set based on the search results, the optimization solution can be transformed into a directly executable scheduling output, ensuring the feasibility of the collaborative scheduling scheme in real systems. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1This is a flowchart illustrating a collaborative scheduling and planning method according to some embodiments of the present invention; Figure 2 This is a flowchart illustrating another collaborative scheduling planning method according to some embodiments of the present invention; Figure 3 This is a flowchart illustrating another collaborative scheduling and planning method according to some embodiments of the present invention; Figure 4 This is a flowchart illustrating another collaborative scheduling planning method according to some embodiments of the present invention; Figure 5 This is a structural block diagram of a collaborative scheduling and planning device according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] According to embodiments of the present invention, a collaborative scheduling planning method, apparatus, computer device, and storage medium are provided. It should be noted that the steps shown in the flowcharts in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowcharts, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0021] This embodiment provides a collaborative scheduling planning method. Figure 1 This is a flowchart of a collaborative scheduling planning method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain the planning input data for the current scheduling cycle. The planning input data includes: robot request set, stability state information corresponding to the robot request set, and the number of unfinished tasks left over from the previous scheduling cycle.

[0022] In this embodiment, the current scheduling cycle refers to the continuous scheduling cycle of the current round in the multi-robot multi-elevator collaborative transportation process; the planning input data refers to the complete set of parameters input when the current round of scheduling starts, which is used for subsequent solution processes; the robot request set contains all robot tasks to be scheduled in the current cycle (including the starting area, target area, and task constraints); the stability status information is a status indicator determined based on the consistency between the current robot request set and the request set of the previous round. If the current request is completely consistent with the previous round's request, it is determined to be a stable request; if a new request appears, a request disappears, or the robot's target floor and starting point change, it is determined to be a fluctuating request; the number of unfinished tasks left over from the previous scheduling cycle refers to the total number of robot tasks that have not been covered or executed by the end of the previous round of planning.

[0023] First, all robot tasks to be executed in the current cycle are acquired to form a robot request set. Simultaneously, the current request set is compared with the historical request set from the previous scheduling cycle to generate corresponding stability status information. The number of unresolved requests from the previous planning cycle is then counted as the number of unfinished tasks. Subsequently, the collected robot request set, stability status information, and number of unfinished tasks, along with other necessary environmental parameters such as the elevator reachable area set and area connectivity, are aggregated and packaged into the input basis for this round of planning. This completes the acquisition and integration of all input data required for this round of scheduling planning.

[0024] Step S102: Based on the stability state information and the number of unfinished tasks, select the target solution time limit for the current scheduling cycle from the preset solution time limits.

[0025] In this embodiment, the selection of the target solution time limit for the current scheduling period based on stability state information and the number of unfinished tasks includes: when the stability state information meets the preset stability condition and the number of unfinished tasks is less than a preset number threshold, selecting a first preset solution time limit as the target solution time limit; when the stability state information does not meet the preset stability condition or the number of unfinished tasks is greater than or equal to the preset number threshold, selecting a second preset solution time limit as the target solution time limit, wherein the second preset solution time limit is greater than the first preset solution time limit.

[0026] Specifically, the preset solution time limit refers to multiple available time limit levels pre-configured for the search and solution process, used to allocate differentiated search time budgets under different scenarios; the first preset solution time limit is a shorter time limit, and the second preset solution time limit is a longer time limit, both belonging to different value levels of the preset solution time limit, and the second preset solution time limit is greater than the first preset solution time limit; the preset stability condition is the judgment benchmark corresponding to the stability state information, that is, when the current robot request set is completely consistent with the request set of the previous cycle, it is considered to meet the condition; the preset quantity threshold refers to the pre-set numerical threshold used to judge whether the amount of legacy tasks is at a low level; the target solution time limit refers to the time limit value actually selected from the preset solution time limits after the condition judgment in this step and used for the current scheduling cycle.

[0027] First, obtain the stability status information of the current scheduling cycle and the number of unfinished tasks left over from the previous scheduling cycle; determine whether the stability status information meets the preset stability conditions, and at the same time determine whether the number of unfinished tasks is less than a preset threshold; if the stability status information meets the preset stability conditions and the number of unfinished tasks is less than the preset threshold, then select the first preset solution time limit as the target solution time limit from the preset solution time limits; if the stability status information does not meet the preset stability conditions, or the number of unfinished tasks is greater than or equal to the preset threshold, then select the second preset solution time limit as the target solution time limit from the preset solution time limits, wherein the second preset solution time limit is greater than the first preset solution time limit.

[0028] By selecting a first preset solution time limit when the stability state of the current scheduling cycle meets the preset stability conditions and the number of unfinished tasks is less than a preset threshold, a shorter search time limit is allocated in scenarios with stable requests and light loads, avoiding excessive consumption of computing resources in cycles with low demand for solution changes. By selecting a second preset solution time limit when the stability state does not meet the preset stability conditions or the number of unfinished tasks is greater than or equal to the preset threshold, a longer search time limit is allocated in scenarios with fluctuating requests or heavy loads, providing sufficient computing time for the search process to cope with the solution requirements of complex scenarios. This achieves an adaptive match between the solution time limit and scenario complexity, achieving a technical effect of balancing solution quality and real-time performance.

[0029] Step S103: Within the target solution time limit, the initial nodes of the target search model are expanded to obtain candidate nodes. Multi-level pruning is performed on the candidate nodes in a preset order to obtain effective candidate nodes. Cost-weighted evaluation is performed on each effective candidate node through the node evaluation function. The global optimal solution is updated according to the evaluation results until the target solution time limit is reached and the search results are obtained. The target search model is used to search and solve the collaborative scheduling process between the robot and the elevator. The initial nodes and candidate nodes correspond to collaborative scheduling schemes with different execution progress.

[0030] In this embodiment, the target solution time limit refers to the search time budget selected based on the stability state and the number of legacy tasks in the current scheduling cycle; the target search model is a state-space search model based on dual-time-axis parallel modeling, used to solve the collaborative scheduling process of multiple robots and multiple elevators. Each node maintains both the robot time axis and the elevator time axis, and uses the larger of their ready times as the collaborative synchronization time; the model uses the minimum sum of the total completion times of all robots as the optimization objective for a globally unified solution, and also supports extending the optimization objective to a weighted objective and / or a minimum-maximum completion time objective; while maintaining dual-time-axis parallel modeling, pruning links, timeout degradation, and stability protection unchanged. It is worth mentioning that this unified solution function framework supports flexible replacement of search strategies. It can choose any implementation of heuristic search, best-first search, breadth-first search, depth-first search, greedy search, or a combination thereof to adapt to deployment scenarios with different computing power and latency constraints. The initial node refers to the starting search node corresponding to the one that has not yet executed any robot requests, and its robot timeline and elevator timeline are both in the initial state. The candidate node refers to the new node generated after performing a feasible extension action on an existing node, corresponding to a scheduling scheme for a certain intermediate execution progress. The search result refers to the known optimal solution or intermediate candidate solution output by the search process after the target solution time limit is reached, which is used to generate a collaborative scheduling scheme in the future.

[0031] Starting from the initial node, the expansion operation is performed iteratively within the target solution time limit. Each time, a node to be processed is taken from the search queue, generating all its feasible candidate nodes. Multi-level pruning is then performed on the candidate nodes in a preset order to filter out invalid branches. The remaining valid candidate nodes have their single-step completion time calculated. This calculation is based on the cumulative time of the robot time axis and the elevator time axis, taking the larger value as the collaborative synchronization time, and combining the costs of each stage such as entering, lifting, and exiting the elevator to generate new nodes. Subsequently, the valid candidate nodes are evaluated by a weighted sum of cumulative cost and heuristic estimated cost through a node evaluation function, and the global optimal solution is updated based on the evaluation results. The above expansion, pruning, evaluation, and update process is repeated until the target solution time limit is reached or the search queue is exhausted, at which point the search stops and the currently maintained optimal solution is output as the search result.

[0032] As an example, in high-concurrency load scenarios with large robot request scale, limited elevator resources, and long regional transfer chains, optimization can be carried out in the following progressive stages: The first stage strengthens the feasible domain construction in the pre-computation stage and enhances the filtering strength of the upper bound pruning of transfer counts; the second stage prioritizes the execution of relaxed upper bound termination pruning and standard upper bound skip pruning in online search to quickly shrink the cost upper bound; the third stage performs feasibility pruning and anti-loop deduplication pruning after shrinking the upper bound to increase the feasible branch density; the fourth stage continuously executes dynamic pre-point selection on the scheduler side to reduce conflicts in congested areas. This progressive optimization method can improve the executable allocation coverage under the same computational budget, and still outputs locally executable solutions when a complete solution is not possible, avoiding single-round blocking.

[0033] In this embodiment of the application, multi-level pruning is performed on candidate nodes according to a preset order to obtain valid candidate nodes, including: Step A1: Obtain the node cost of the candidate node and the cost of the known optimal solution currently recorded.

[0034] Specifically, the node cost of a candidate node refers to the cost value obtained by weighting and summing the cumulative cost and heuristic estimated cost of the candidate node through the node evaluation function. It is used to measure the quality of the scheduling scheme corresponding to the node. The cost of the known optimal solution refers to the node cost value corresponding to the optimal scheduling scheme found in the current search process. It is initially set to positive infinity and is continuously updated by better node costs as the search progresses. It serves as the benchmark reference value for subsequent pruning judgments.

[0035] Once a candidate node is generated, its pre-calculated node cost is first read from the candidate node's data structure. At the same time, the known optimal solution cost currently recorded and maintained during the search process is also read. Both are used as input data for subsequent pruning decisions at each level, providing a comparison benchmark for the first-level pruning operation.

[0036] Step A2, perform the first level of pruning: when the node cost of a candidate node is greater than a preset relative multiple of the cost of the known optimal solution, terminate the search of the search queue where the candidate node is located.

[0037] Specifically, the preset relative multiple refers to a pre-set amplification factor greater than 1, used to relax the amplification of the cost of the known optimal solution, and serves as the termination threshold for the first-level pruning; the search queue refers to the set of nodes to be processed organized in a specific data structure (such as a priority queue) during the search process, which are taken out and expanded in sequence after being sorted by node cost. When a search queue is terminated, all nodes in the queue will no longer participate in the subsequent expansion.

[0038] In the first-level pruning, the node cost of a candidate node is compared with the cost of the known optimal solution multiplied by a preset relative multiple. This preset relative multiple, the relative stability threshold coefficient, and the absolute stability threshold can all be configured according to the actual scenario, supporting dynamic adjustment with concurrency levels to adapt to the pruning intensity and scheduling stability requirements under different load conditions. If the node cost of a candidate node is greater than this product value, it indicates that the candidate node and all nodes in its search queue are too far from the current known optimal solution, and expanding the queue is unlikely to yield a better solution. Therefore, the entire search of the candidate node's search queue is terminated, and no further expansion or evaluation operations are performed on any node in that queue. If the node cost of a candidate node does not exceed this product value, the candidate node passes the first-level pruning and continues into subsequent pruning stages.

[0039] Step A3: If the candidate node passes the first-level pruning, then perform the second-level pruning: when the node cost of the candidate node or its child node is greater than the cost of the known optimal solution, skip the corresponding branch.

[0040] Specifically, a child node refers to a new node generated after continuing an extension action from the current candidate node, corresponding to the intermediate scheduling state of completing one more robot task based on the existing execution progress; skipping the corresponding branch means that in the subsequent search and extension process, no further processing is performed on the skipped node, that is, the node is not included in the search queue, nor is any further extension, evaluation or update operation performed on it.

[0041] For a candidate node that passes the first-level pruning, the second-level pruning simultaneously checks the cost of the candidate node itself and all its child nodes. The node cost of the candidate node is compared with the cost of the currently recorded known optimal solution, and the node cost of each of its child nodes is estimated and compared with the cost of the known optimal solution. If the node cost of the candidate node itself is greater than the cost of the known optimal solution, or the node cost of any child node is greater than the cost of the known optimal solution, then the branch of the candidate node itself or its corresponding child node is skipped and it will not be included in the subsequent expansion process. If the node cost of the candidate node and its child nodes is not greater than the cost of the known optimal solution, then the candidate node passes the second-level pruning and continues to enter the subsequent pruning stage.

[0042] Step A4: If the candidate node passes the second-level pruning, then perform the third-level pruning: obtain the expansion action corresponding to the candidate node. When the expansion action does not meet the first-step constraint in the preset feasible solution cache library, skip the corresponding expansion. The preset feasible solution cache library is generated after path pre-calculation based on the elevator reachable area set and the area connectivity relationship. The first-step constraint is that the first-step operation of the expansion action is located in the preset feasible solution cache library.

[0043] Specifically, extended actions refer to the specific scheduling operations executed to complete the next robot request, starting from the scheduling state corresponding to the current candidate node. These operations include a complete sequence of actions such as assigning a robot to a specific elevator or taking the elevator to the target floor. The pre-set feasible solution cache library refers to the feasible solution index table generated before the search begins by enumerating the transfer paths of all region pairs based on the set of elevator reachable regions and the region connectivity. It is used to quickly query the optional transfer schemes from any starting region to any target region. The construction of this cache library can be implemented using any of the following strategies: full pre-computation, incremental update, on-demand lazy loading, or partitioned hierarchical caching. The generation process of this cache library is the path pre-computation stage. In this stage, the upper bound pruning of the number of transfers is performed: after sorting the candidate transfer paths of the same region pair in ascending order of the number of transfers, either the "retain the path scheme with the fewest and second fewest transfers" or the "retain all path schemes that do not exceed the preset upper bound of the number of transfers" can be used to filter them. The remaining schemes with high transfers and more circuitous paths are eliminated in the pre-computation stage to avoid entering the online search space. The first step constraint means that when performing an expansion action from the current candidate node, the first step of the expansion action must be the first step of the feasible scheme recorded for the corresponding region pair in the preset feasible solution cache library. Otherwise, the expansion action is not feasible.

[0044] As an example, there are three possible elevator routes from 1F to 5F: taking elevator L1 directly, taking elevator L2 directly, or taking elevator L3 and then transferring to elevator L4. After enumeration, the routes with the fewest transfers are pruned using the upper bound of the transfer count (P5), and only the routes with the fewest transfers are cached, namely, taking elevator L1 directly and taking elevator L2 directly. When the robot needs to travel from 1F to 5F, it directly reads the two cached direct routes and selects the appropriate elevator to execute.

[0045] For candidate nodes that pass the second-level pruning, the third-level pruning first obtains the extended action to be executed by the candidate node and extracts the first step of the extended action. Then, the first step is compared with the first step constraint from the starting region to the target region in the preset feasible solution cache. If the first step is not in the set of first step solutions recorded in the cache, it means that the extended action cannot lead to a feasible transfer path, and the extension is skipped and not included in the subsequent extension process. If the first step satisfies the first step constraint, the candidate node passes the third-level pruning and continues to enter the subsequent pruning stage. It should be noted that the feasible solutions in the preset feasible solution cache have been filtered by the upper bound pruning of the number of transfers in the pre-calculation stage before the search begins. In the online search stage, the feasible first step operation set can be obtained directly by looking up the table, without having to repeatedly enumerate the transfer path during the search process.

[0046] Step A5: If the candidate node passes the third-level pruning, then perform the fourth-level pruning: obtain the access region record corresponding to the candidate node, and skip the corresponding expansion when the access region record contains a repeatedly accessed intermediate region.

[0047] Specifically, the access area record refers to the list of all areas (including the starting area, intermediate transfer areas, and the target area that have been visited) on the search path from the initial node to the current candidate node. It is used to track the search trajectory to avoid repeated traversal. The intermediate area refers to the transfer floors or transfer stations that the robot passes through on the elevator transfer path other than the starting area and the target area. It is a transitional area that is neither the starting point nor the end point.

[0048] For candidate nodes that pass the third-level pruning, the fourth-level pruning first obtains the access region record corresponding to the candidate node and extracts all visited intermediate regions in the record. Then, it checks whether the new intermediate region corresponding to the current expansion action already exists in the access region record. If an intermediate region is found to have been visited, it indicates that the expansion path has formed a loop, and continuing to expand will lead to redundant search and will not lead to a better solution. Therefore, the expansion operation is skipped directly. If no intermediate region is visited repeatedly, the candidate node passes the fourth-level pruning and is retained as a valid candidate node for subsequent evaluation and updating.

[0049] Step A6: The nodes retained after the first, second, third and fourth pruning stages are selected as valid candidate nodes.

[0050] Specifically, a valid candidate node refers to a candidate node that has not been eliminated by any level of pruning after passing through all the tests of the first, second, third and fourth levels of pruning. This node is qualified to continue to participate in subsequent search expansion and evaluation updates.

[0051] After the candidate nodes have completed all the checks of the first-level pruning, second-level pruning, third-level pruning and fourth-level pruning in sequence, the candidate nodes that have not been terminated or skipped by any of the above-mentioned pruning levels are collected as valid candidate nodes. These valid candidate nodes serve as the credible branches that have passed all pruning checks in this round of expansion, and are used for subsequent node evaluation and global optimal solution update processing.

[0052] By obtaining the node cost of candidate nodes and the cost of the known optimal solution and performing the first level of pruning, the search queue of a candidate node is terminated when its node cost exceeds a preset relative multiple of the cost of the known optimal solution. Branches far from the known optimal solution are pruned from the search queue level, compressing the search space. After the candidate node passes the first level of pruning, the second level of pruning is performed. When the node cost of the candidate node itself or any of its child nodes exceeds the cost of the known optimal solution, the corresponding branch is skipped, filtering out expansion paths that are unlikely to produce a better solution branch by branch. After the candidate node passes the second level of pruning, the third level of pruning is performed, utilizing a pre-computed feasible solution cache library to optimize the expansion actions. The first step involves constraint verification. If an expansion action does not meet the first-step constraints of the cache library, the corresponding expansion is skipped to avoid repeated feasibility checks during the online search phase. After candidate nodes undergo third-level pruning, fourth-level pruning is performed. If a repeatedly accessed intermediate region appears in the access region record, the corresponding expansion is skipped to eliminate invalid search paths that form a circular redundancy. By using nodes retained after four levels of pruning as valid candidate nodes, a coarse-to-fine pruning chain is formed, from coarse-grained queue termination to fine-grained expansion filtering. Invalid branches are screened out with progressively increasing pruning precision, ensuring that candidate nodes entering the subsequent evaluation stage are all high-quality branches with potential optimization value.

[0053] In this embodiment of the application, after obtaining valid candidate nodes, the method further includes: Step B1: Obtain the action to be executed corresponding to the valid candidate node.

[0054] Specifically, the action to be executed refers to the specific scheduling operation prepared to be executed in order to advance the scheduling process, starting from the current scheduling state corresponding to the valid candidate node. This operation covers a complete sequence of actions such as assigning a specific robot to a designated elevator and performing an elevator ride to the target floor. It is the basic unit for subsequent acquisition of time axis cumulative time and calculation of single-step completion time.

[0055] After a valid candidate node passes all pruning tests, the corresponding action to be executed is extracted from the valid candidate node. Complete operation information such as robot identity, elevator identity, action type and action sequence involved in the operation is obtained as the input basis for subsequent determination of target elevator and target robot.

[0056] Step B2: Based on the pre-built elevator timeline and robot timeline, determine the corresponding target elevator and target robot according to the action to be executed.

[0057] Specifically, the pre-built elevator timeline refers to the data structure maintained in each node of the target search model to record the cumulative ready time of each elevator, reflecting the available time of each elevator under the current scheduling progress; the pre-built robot timeline refers to the data structure maintained in each node of the target search model to record the cumulative ready time of each robot, reflecting the available time of each robot under the current scheduling progress; the target elevator refers to the specific elevator matched from the set of available elevators to perform the current elevator ride action based on the starting area, target area, and elevator reachable area information in the action to be performed; the target robot refers to the specific robot matched from the current scheduling scenario to perform the current action based on the robot identity information in the action to be performed.

[0058] After obtaining the action to be executed corresponding to the valid candidate node, the robot identity information, starting area, target area and action type are extracted from the action to be executed. Based on the elevator availability status and reachable area information of each elevator recorded in the pre-built elevator timeline, the target elevator that can complete the current elevator ride task and is currently available is matched by combining the starting area and target area of ​​the action to be executed. At the same time, according to the robot identity information in the action to be executed, the corresponding target robot is determined from the pre-built robot timeline, and the dual object locking of the target elevator and the target robot is completed for subsequent acquisition of their respective cumulative time.

[0059] Step B3: Obtain the cumulative elevator time on the elevator timeline and the cumulative robot time on the robot timeline.

[0060] Specifically, the cumulative elevator time refers to the moment when the target elevator can start executing the next task after completing all previously assigned elevator riding tasks in the scheduling state corresponding to the current valid candidate node. This moment is recorded in the elevator timeline. The cumulative robot time refers to the moment when the target robot can start executing the next operation after completing all previously assigned movement and elevator riding tasks in the scheduling state corresponding to the current valid candidate node. This moment is recorded in the robot timeline.

[0061] After determining the target elevator and the target robot, the cumulative ready time of the target elevator is read from the elevator timeline maintained by the current valid candidate nodes as the cumulative elevator time, and the cumulative ready time of the target robot is read from the robot timeline as the robot cumulative time. The obtained cumulative elevator time and cumulative robot time are used as the two basic input data for subsequent single-step completion time calculation.

[0062] Step B4: Calculate the single-step completion time of the action to be executed based on the cumulative elevator time, the cumulative robot time, and the cost of each stage corresponding to the action to be executed, and generate new valid candidate nodes based on the single-step completion time.

[0063] Specifically, the cost of each stage refers to the time consumed by the robot in each sub-stage required to complete the action to be performed, including at least the cost of entering the elevator, the cost of lifting, the cost of exiting the elevator, and the additional cost of the final step. The costs of each stage correspond to: the cost of the robot reaching the previous point is Cr,pre; the cost of the elevator reaching the starting floor is Cl,start; and the costs of entering, lifting, exiting, and the additional cost of the final step are Center, Clift, Ceexit, and Cfinal, respectively. The single-step completion time refers to the time reached by the robot after completing all stages of the action to be performed, starting from the current state. It is calculated by adding the costs of each stage based on the collaborative synchronization time. A new valid candidate node is a search node formed by encapsulating the new scheduling state after the action to be performed is completed, based on the current valid candidate nodes. Both the robot's timeline and the elevator's timeline are updated according to the single-step completion time, serving as a new starting point for subsequent expansion.

[0064] After obtaining the cumulative time of the elevator and the robot, the single-step completion time is calculated based on the stage costs corresponding to the action to be executed. This calculation process can be represented by the following set of cooperative timing calculation formulas:

[0065]

[0066]

[0067]

[0068] Where Tr,pre is the cumulative time before the robot's current step, i.e., the robot's cumulative time; Tl,pre is the cumulative time before the elevator's current step, i.e., the elevator's cumulative time; Tr,ready is the robot's ready time; Tl,ready is the elevator's ready time; Tsync is the cooperative synchronization time; and Tstep is the single-step completion time.

[0069] Based on the above formula, the calculated collaborative synchronization time is summed with the elevator entry cost, elevator rise cost, elevator exit cost, and final step additional cost to obtain the single-step completion time of the action to be executed. Then, the calculated single-step completion time is updated to the target robot timeline and the target elevator timeline respectively, and the set of unfinished robot tasks is updated and encapsulated to form new effective candidate nodes for subsequent search expansion.

[0070] By acquiring the actions to be executed corresponding to valid candidate nodes, the specific operational information required for single-step expansion is locked, providing precise input for subsequent timing calculations. By determining the target elevator and target robot based on the pre-built elevator timeline and robot timeline and the actions to be executed, the elevator and robot objects involved in this action are accurately located in the dual timelines, ensuring the accuracy of timing calculations. By acquiring the cumulative elevator timeline of the target elevator on the elevator timeline and the cumulative robot timeline of the target robot on the robot timeline, the independent ready times of the elevator side and the robot side are read separately, providing basic data for calculating the collaborative synchronization time. By calculating the single-step completion time based on the cumulative elevator timeline, the cumulative robot timeline, and the cost of each stage, the larger value of the two ready times is taken as the collaborative synchronization time, and the costs of entering, ascending, exiting, and the final step are superimposed. At the same time, the waiting costs of the robot waiting for the elevator and the elevator waiting for the robot are characterized. New valid candidate nodes are generated based on the single-step completion time. During the node expansion process, the state of the dual timelines is continuously updated synchronously, so that each node in the search space always reflects the accurate timing state.

[0071] In this embodiment, a cost-weighted evaluation is performed on each valid candidate node using a node evaluation function, and the global optimal solution is updated based on the evaluation results, including: Step C1: Obtain the cumulative cost of the valid candidate nodes, and perform heuristic cost estimation calculation on the remaining requests corresponding to the valid candidate nodes to obtain the heuristic cost estimation.

[0072] Specifically, the cumulative cost refers to the total cost of all completed actions from the initial node to the current valid candidate node, corresponding to g(n) in the node evaluation function; the remaining requests refer to robot requests that have not yet been assigned or completed in the scheduling state corresponding to the current valid candidate node, i.e., the remaining tasks in the robot request set that are still to be scheduled besides the processed requests; the heuristic estimated cost is the estimated cost required to complete all remaining requests starting from the current valid candidate node, corresponding to h(n) in the node evaluation function, used to estimate the approximate remaining cost required to complete all tasks from the current state during the evaluation phase. The node evaluation function can be expressed as follows: , where g(n) is the cumulative cost, h(n) is the heuristic estimation cost, and wg and wh are preset weight coefficients.

[0073] After a valid candidate node is generated, its cumulative cost is first read directly from the valid candidate node as g(n); at the same time, the set of unfinished robot tasks corresponding to the valid candidate node is obtained as the remaining requests. For each task in the remaining requests, the estimated cost required to complete the task from the current state is calculated based on the preset heuristic function, and the estimated costs of each task are summarized to obtain the heuristic estimated cost as h(n); the cumulative cost and the heuristic estimated cost are used as the two basic inputs for subsequent node evaluation calculations.

[0074] Step C2 involves weighted summation of the cumulative cost and the heuristically estimated cost to obtain the node cost of the valid candidate nodes.

[0075] Specifically, weighted summation refers to the operation of multiplying the cumulative cost and the heuristic estimated cost by their respective weight coefficients and then adding them together, i.e., according to the node evaluation function. The calculation is performed, where g(n) is the cumulative cost, h(n) is the heuristic estimation cost, and wg and wh are preset weight coefficients; the node cost refers to the final evaluation value obtained after the above weighted summation operation for the effective candidate nodes, i.e. The calculation results are used to comprehensively evaluate the quality of the node; the smaller the node cost, the better the scheduling scheme corresponding to the candidate node.

[0076] After obtaining the cumulative cost and the heuristic estimated cost, a weighted summation operation is performed according to the node evaluation function: the cumulative cost is multiplied by a preset cumulative cost weight coefficient wg, and the heuristic estimated cost is multiplied by a preset heuristic estimated cost weight coefficient wh. The products of the two are then added together to obtain the result. This is the node cost of the valid candidate node; this node cost serves as the basis for subsequent comparison with the known optimal solution cost and for updating the global optimal solution.

[0077] Step C3: When the node cost of a valid candidate node is better than the cost of the currently recorded known optimal solution, update the valid candidate node to the global optimal solution.

[0078] Specifically, "better than" means that the cost of a valid candidate node is less than the cost of the currently recorded known optimal solution. Since a smaller node cost represents a better scheduling scheme, a smaller node cost means that the scheduling scheme corresponding to the candidate node is better. The global optimal solution refers to the node corresponding to the scheduling scheme with the smallest node cost among all the nodes found in the current search process. As the search progresses, whenever a better node is found, the global optimal solution is updated to that node.

[0079] After calculating the node cost of a valid candidate node, the node cost is compared with the cost of the currently recorded known optimal solution. If the node cost of a valid candidate node is less than the cost of the known optimal solution, it means that the scheduling scheme corresponding to the candidate node is better than all the currently discovered schemes. At this time, the valid candidate node is updated to the new global optimal solution, and its node cost is updated to the cost of the new known optimal solution for subsequent pruning and evaluation. If the node cost of a valid candidate node is not less than the cost of the known optimal solution, the existing global optimal solution remains unchanged.

[0080] Heuristically estimated costs are obtained by acquiring the cumulative costs of valid candidate nodes and performing heuristic cost estimation on the remaining requests. This separates the acquired costs from the future estimated costs, providing an evaluation basis that balances historical consumption and future expectations. The node costs of valid candidate nodes are obtained by weighted summation of the cumulative costs and heuristic estimated costs. Weighting coefficients flexibly adjust the balance between historical and estimated costs, ensuring that node evaluation considers both the actual costs of completed paths and the estimated costs of remaining paths. The global optimal solution is updated when the node cost of a valid candidate node is better than the currently recorded cost of the known optimal solution. This ensures that the global optimal solution continuously converges towards a better direction as the search progresses, guaranteeing that the scheduling scheme output at the end of the search has optimality within the target solution timeframe.

[0081] Step S104: Generate a collaborative scheduling scheme corresponding to the robot request set based on the search results.

[0082] In this embodiment of the application, generating a collaborative scheduling scheme corresponding to the robot request set based on the search results includes: Step D1: Check whether the globally optimal solution in the search results covers all robot requests in the robot request set.

[0083] Specifically, the search result refers to the known optimal solution or intermediate candidate solution output by the search process after the target solution time limit is reached; the global optimal solution refers to the node corresponding to the scheduling scheme with the minimum node cost among all nodes discovered in the current search process; coverage refers to the scheduling scheme that has successfully allocated and solved the elevator ride task for the corresponding robot; the robot request set contains all robot tasks to be scheduled in the current period; all robot requests refer to all tasks in the robot request set; intermediate candidate nodes refer to nodes that have been visited during the search process but have not yet been fully expanded or have not been finally selected as the global optimal solution; and the local optimal solution refers to the executable scheduling scheme corresponding to the intermediate candidate node that only covers a portion of the robot requests.

[0084] After completing the search and obtaining the search results, extract the global optimal solution from the search results; compare each robot request covered by the global optimal solution with all robot requests in the robot request set, and count the number of covered requests; determine whether the number of covered requests is equal to the total number of requests in the robot request set. If the two are equal, it is determined that the global optimal solution covers all robot requests. If the number of covered requests is less than the total number of requests, it is determined that not all requests are covered.

[0085] Step D2: If the globally optimal solution covers all robot requests in the robot request set, then output the globally optimal solution as the collaborative scheduling scheme. Alternatively, if the globally optimal solution does not cover all robot requests in the robot request set, then select the intermediate candidate node that covers the most robot requests in the robot request set from the candidate nodes visited during the search process, and output the local optimal solution corresponding to the intermediate candidate node as the collaborative scheduling scheme.

[0086] Specifically, the collaborative scheduling scheme refers to the complete set of scheduling instructions that are ultimately output for the robots to actually execute, including the allocation relationship between each robot and the elevator, the elevator path and timing arrangement; when the global optimal solution covers all robot requests in the robot request set, it is directly output as the collaborative scheduling scheme; intermediate candidate nodes refer to any node that has been visited during the search process but has not been finally selected as the global optimal solution, which corresponds to an intermediate scheduling state in which tasks have been partially assigned; local optimal solutions refer to the executable scheduling schemes corresponding to intermediate candidate nodes that only cover some robot requests.

[0087] Based on the detection results, branch processing is executed: if the global optimal solution covers all robot requests in the robot request set, then the global optimal solution is directly output as the collaborative scheduling scheme for the current scheduling cycle; if the global optimal solution fails to cover all robot requests, then all candidate nodes visited during the search process are traversed, the number of robot requests covered by each intermediate candidate node is counted, the intermediate candidate node with the largest coverage is selected, and then the local optimal solution corresponding to that intermediate candidate node is output as the collaborative scheduling scheme. It should be noted that in abnormal scenarios where some requests are temporarily unreachable, planning times out, or requests fluctuate briefly, intermediate candidate solutions are continuously maintained during the search; when a preset time limit is reached, a local scheme degradation output is triggered, and no empty result is returned; when the request set is stable, double threshold stability protection is implemented and historical schemes are reused first; when the request set changes significantly, a new round of complete planning is switched to. This anomaly recovery method can achieve continuous executable output under abnormal cycles and reduce execution jitter caused by frequent reassignment.

[0088] By checking whether the global optimal solution in the search results covers all robot requests, the completeness of the solution is determined, providing a basis for subsequent branch output. When the global optimal solution covers all robot requests, the global optimal solution is output as the collaborative scheduling scheme. When the solution is successful, the complete scheme is output directly, ensuring that the execution scheme has global optimality. When the global optimal solution fails to cover all robot requests, the intermediate candidate node with the most covered requests is selected from the visited candidate nodes and its corresponding local optimal solution is output. When timeout or some requests cannot be solved, the locally executable scheme with the most solved robots is still output, avoiding the blocking mode of total failure when there is no global solution, and ensuring the continuous executable output capability of the scheduling system in complex scenarios.

[0089] In this embodiment of the application, before expanding the initial nodes of the target search model to obtain candidate nodes, such as Figure 2 As shown, the method also includes: Step S201: Initialize the set of incomplete robot tasks based on the set of robot requests.

[0090] In this embodiment, the incomplete robot task set refers to the list of robot requests that have not yet been covered or resolved by any scheduling scheme within the current scheduling cycle. Initially, it contains all tasks in the robot request set. As the search progresses, each completed robot request is removed from the set. The robot request set contains all robot tasks to be scheduled within the current cycle (including the starting region, target region, and task constraints), and is the sole data source for initializing the incomplete robot task set.

[0091] Before expanding the initial node, the robot request set for the current scheduling period is first obtained. Then, a task list with the same content as the robot request set is created. This list is used as the initial state of the incomplete robot task set to record all robot requests to be resolved at the initial moment of the search. This incomplete robot task set will be updated step by step in the subsequent search process. Whenever a robot request is successfully assigned to a robot and elevator and the scheduling is completed, the corresponding task is removed from the set until the set is empty, indicating that all requests have been covered.

[0092] Step S202: Obtain elevator time axis initialization parameters and robot time axis initialization parameters.

[0093] In this embodiment, the elevator timeline initialization parameters refer to the initial state data of each elevator required for constructing the elevator timeline, including basic configuration information such as the initial floor of each elevator, the initial ready time (usually 0 or the current scheduling start time), and the reachable area of ​​each elevator. The robot timeline initialization parameters refer to the initial state data of each robot required for constructing the robot timeline, including basic configuration information such as the initial area of ​​each robot, the initial ready time (usually 0 or the current scheduling start time), and the types of tasks that the robot can execute. Together, they constitute the basic state of the dual timelines in the initial search node, providing a starting reference for the subsequent tracking and updating of accumulated timelines.

[0094] Before expanding the initial nodes, the basic configuration information of elevators and robots is first read from the planning input data. Elevator time axis initialization parameters are generated based on the initial floor, initial available time, and reachable area information of each elevator. At the same time, robot time axis initialization parameters are generated based on the initial area, initial available time, and task constraint information of each robot. These two sets of initialization parameters are used as input data for the subsequent construction of elevator and robot time axes to ensure that the time axis in each search node starts recording from a consistent initial state.

[0095] Step S203: Construct the elevator timeline based on the elevator timeline initialization parameters, and construct the robot timeline based on the robot timeline initialization parameters.

[0096] In this embodiment of the application, the elevator time axis refers to the data structure used to record the cumulative ready time of each elevator, and the robot time axis refers to the data structure used to record the cumulative ready time of each robot. Together, they form the basis for parallel modeling of dual time axes in the target search model.

[0097] After obtaining the elevator timeline initialization parameters and the robot timeline initialization parameters, corresponding data structures are constructed using these two sets of parameters. First, based on the initial floor, initial ready time, and reachable area information of each elevator in the elevator timeline initialization parameters, a complete elevator timeline data structure is established to record the cumulative time of each elevator in the current scheduling state. Simultaneously, based on the initial area, initial ready time, and task constraint information of each robot in the robot timeline initialization parameters, a complete robot timeline data structure is established to record the cumulative time of each robot in the current scheduling state. After the two timelines are constructed, they serve as components of the initial nodes of the target search model, providing the basic data structure for reading and updating the cumulative time during the subsequent search process.

[0098] Step S204: Based on the elevator timeline, the robot timeline, and the set of unfinished robot tasks, generate the initial node of the target search model, where the initial node corresponds to the initial scheduling scheme for which no robot requests have been executed.

[0099] In this embodiment, the initial scheduling scheme refers to the initial state of scheduling before any robot requests have been executed, i.e., a blank scheduling scheme in which all robots are located in their respective starting areas, all elevators are at their initial floors and no elevator riding tasks have been assigned; the target search model refers to a state space search model constructed based on parallel modeling of dual time axes, in which each node maintains both the robot time axis and the elevator time axis, and uses the maximum value of the ready times of the two as the collaborative synchronization time. The initial node is the starting node in the target search model, corresponding to the initial scheduling scheme.

[0100] After constructing and initializing the elevator timeline, robot timeline, and set of unfinished robot tasks, the three are encapsulated together as the initial node of the target search model: the elevator timeline in this initial node records the initial floor and initial ready time of each elevator, the robot timeline records the initial area and initial ready time of each robot, and the set of unfinished robot tasks contains all tasks in the set of robot requests; this initial node serves as the starting state of the search, corresponding to the initial scheduling scheme in which no robot requests have been executed, and all subsequent expansions and searches start from this initial node.

[0101] By initializing the set of incomplete robot tasks based on the set of robot requests, a complete task list is established at the start of the search, providing a unified benchmark for tracking the completion status of subsequent tasks. By obtaining the initialization parameters of the elevator time axis and the robot time axis and constructing the elevator time axis and robot time axis respectively, a dual-time axis parallel data structure is established at the start of the search, providing a unified timing reference benchmark for calculating the completion time of subsequent single steps. By generating initial nodes based on the elevator time axis, robot time axis and the set of incomplete robot tasks, the initial state of the scheduling scheme is fully encapsulated at the start of the search, ensuring that the initial node and all subsequent candidate nodes maintain a unified data structure, guaranteeing the consistency and comparability of the states of each node during the search process.

[0102] In this embodiment of the application, after obtaining the planning input data for the current scheduling period, as follows: Figure 3 As shown, the method also includes: Step S301: Traverse the robot's candidate forward point set and obtain the real-time path cost corresponding to each candidate forward point in the candidate forward point set.

[0103] In this embodiment, the candidate forward point set refers to a set of one or more selectable connection points (i.e., the waiting positions the robot reaches before entering the elevator process) pre-configured for each robot. Each candidate forward point in this set is located outside the elevator (i.e., a non-SM point). The robot travels from its current location to this forward point before entering the elevator to complete the elevator connection. The elevator interior stopping point is called an SM point, and the external candidate connection point is called an LM point. In a path structure where SM points and LM points intertwine (e.g., SM1...), the connection point is defined as follows: LM2 LM3 ...), when the robot travels from its current path to the SM point, the second-to-last non-SM point on its path is selected as the next forward point; when multiple candidate forward points exist, the one with the best cost is selected as the target forward point after comparing path costs. This forward point selection mechanism is dynamically calculated by the scheduler during the planning phase, unlike the connection scheme determined by fixed detection points on the robot's side. Real-time path cost refers to the estimated time or comprehensive cost required for the robot to move from its current position to a candidate forward point along a practically feasible path. This cost changes dynamically with the robot's current position and the environmental traffic conditions, reflecting the impact of different forward point selections on scheduling efficiency.

[0104] After obtaining the planning input data for the current scheduling cycle, for each robot that needs to be scheduled within the current scheduling cycle, obtain its corresponding set of candidate forward points; traverse each candidate forward point in the set, and calculate the real-time path cost required to reach each candidate forward point based on the actual path distance between the robot's current location and each candidate forward point and the environmental traffic status information; record each candidate forward point and its corresponding real-time path cost one by one as the basis for subsequent cost comparison and selection.

[0105] Step S302: Compare the real-time path costs corresponding to each candidate forward point to obtain the comparison results.

[0106] In this embodiment of the application, cost comparison refers to the process of comparing the real-time path costs corresponding to each candidate pre-point. The superiority or inferiority of different candidate pre-points is distinguished by comparing the numerical values ​​of each cost. The comparison result refers to the comparison conclusion output after the cost comparison operation. The result is presented in the form of ranking or difference comparison of each candidate pre-point and its corresponding real-time path cost, which serves as the direct basis for subsequent screening of the target pre-point with the optimal cost.

[0107] After obtaining the real-time path cost corresponding to each candidate forward point in the candidate forward point set, the real-time path costs of all candidate forward points are aggregated together; then, the values ​​of these real-time path costs are compared with each other, arranged in ascending order of value, or the relative size relationship between each cost is compared one by one, thus forming a comparison result containing the ranking of the costs of each candidate forward point and the information on the differences between them, so that subsequent steps can use this to select the target forward point with the best cost in the current scheduling cycle.

[0108] Step S303: Based on the comparison results, select the target pre-position with the best cost in the current scheduling cycle from the candidate pre-position set.

[0109] In this embodiment, the optimal cost refers to the lowest real-time path cost among all candidates in the candidate forward point set, i.e., the lowest estimated time or comprehensive cost for the robot to move from its current position to the forward point among all candidate solutions; the target forward point refers to the candidate forward point with the lowest real-time path cost selected from the candidate forward point set after cost comparison, and this forward point will be included in the input data of this round of scheduling planning.

[0110] After obtaining the cost comparison results, based on the ranking or comparison information of the real-time path costs of each candidate forward point in the comparison results, the candidate forward point with the smallest real-time path cost is selected from the candidate forward point set; the selected forward point is determined as the target forward point of the robot in the current scheduling cycle, and is used as the optimal connection position selected by the robot before entering the elevator connection process, and is integrated into the planning input data of the current scheduling cycle.

[0111] Step S304: Integrate the target precursor into the planning input data of the current scheduling cycle, and execute subsequent search planning based on the target precursor.

[0112] In this embodiment, the target preceding point refers to the candidate connection point with the minimum real-time path cost from the robot's current position to that point within the current scheduling cycle. After the robot completes the movement to the preceding point, it will enter the elevator connection process. The planning input data refers to the complete set of parameters input when the current scheduling cycle starts, including the robot request set, stability status information, the number of unfinished tasks left over from the previous cycle, the set of elevator reachable areas and the connectivity relationship between areas, path cost evaluation information, etc. All subsequent searches and planning are based on this input data. Integration refers to writing the selected target preceding point as an additional attribute of each robot request into the planning input data, making it a constraint condition for the robot to move from the starting area to the preceding point and then perform the elevator action in the subsequent search process. Subsequent search planning refers to the complete search process based on the target search model, including expansion, pruning, and evaluation, that is, the entire process of searching for the global optimal solution or locally executable solution within the target solution time limit starting from the initial node.

[0113] After determining the target pre-point, this target pre-point is written into the planning input data of the current scheduling cycle as the pre-boarding connection position information for the corresponding robot, making it a fixed pre-constraint for the robot's current scheduling task. Subsequently, in the subsequent search planning, each robot, starting from its initial area, first goes to the target pre-point to complete the elevator connection, and then performs subsequent actions such as entering the elevator, riding the elevator, and exiting the elevator. After integration, the subsequent iterative search planning is started based on the updated planning input data: within the target solution time limit, the initial nodes of the target search model are expanded, pruned, evaluated, and the global optimal solution is searched. In this subsequent search, each robot, according to its assigned request, starts from the initial area and first goes to the target pre-point to complete the connection preparation from the LM point to the SM point, and then performs the elevator entry, elevator ride, exit, and final actions, realizing complete elevator collaborative scheduling.

[0114] By traversing the candidate foreground point set to obtain the real-time path cost corresponding to each candidate foreground point, dynamic path cost information of each foreground point is collected during the planning stage, providing a real-time data foundation for foreground point selection. By comparing the real-time path costs of each candidate foreground point, the comparison results are obtained, and the costs of each foreground point are ranked and compared using a unified dimension to eliminate subjective selection bias. Based on the comparison results, the target foreground point with the best cost is selected from the candidate foreground point set, and the elevator connection position is selected with the minimization of real-time path cost as the guide, reducing the robot's movement time before reaching the elevator. By integrating the target foreground point into the planning input data and performing subsequent search planning based on the target foreground point, the scheduler completes the dynamic optimization selection of foreground points during the planning stage. Unlike the fixed detection point scheme on the robot side, the elevator connection position can be adaptively optimized according to changes in traffic conditions, improving the quality of elevator connection.

[0115] In this embodiment of the application, after generating a collaborative scheduling scheme corresponding to the robot request set based on the search results, as follows: Figure 4 As shown, the method also includes: Step S401: When the stability state information of the robot request set meets the preset stability conditions, obtain the current re-evaluation cost corresponding to the collaborative scheduling scheme and the historical re-evaluation cost corresponding to the previous historical scheduling scheme.

[0116] In this embodiment, the current re-evaluation cost refers to the total completion cost of the optimal collaborative scheduling scheme (i.e., the global optimal solution or the local optimal solution) obtained in the current scheduling cycle after re-evaluation according to the unified cost model, corresponding to Cnew in the stability protection criterion; the historical re-evaluation cost refers to the total completion cost of the previous round's historical scheduling scheme after re-evaluation according to the unified cost model, corresponding to Clast in the stability protection criterion; the preset stability condition refers to the condition that is satisfied when the current robot request set is completely consistent with the request set of the previous cycle, and this condition is the same as the condition used to determine whether to select the first preset solution time limit; the previous round's historical scheduling scheme refers to the target collaborative scheduling scheme that was finally output and actually executed in the previous scheduling cycle.

[0117] After generating the collaborative scheduling scheme for the current period based on the search results, the stability status information of the robot request set in the current scheduling period is first obtained, and it is determined whether the stability status information meets the preset stability condition. If the condition is met (i.e., the current request set is completely consistent with the request set of the previous round), the current re-estimated cost corresponding to the output scheme of the current scheduling period is obtained, and the historical re-estimated cost corresponding to the historical scheduling scheme of the previous round is also obtained. The two cost values ​​are used as the basic input data for subsequent cost deviation value calculation. If the stability status information does not meet the preset stability condition, this step and subsequent stability protection operations are not executed.

[0118] Step S402: Calculate the cost deviation between the current revaluation cost and the historical revaluation cost.

[0119] In the embodiments of this application, the cost deviation value refers to the difference between the current re-evaluated cost and the historical re-evaluated cost. It is used to quantify the degree of deviation of the new scheme obtained in the current scheduling cycle from the historical scheme in the previous round in terms of total completion cost, and is the quantitative basis for judging the stability threshold.

[0120] After obtaining the current revaluation cost and the historical revaluation cost, the current revaluation cost is subtracted from the historical revaluation cost, and the difference between the two is calculated as the cost deviation value. If the current revaluation cost is less than the historical revaluation cost, the cost deviation value is negative, indicating that the new solution is better than the historical solution. If the current revaluation cost is greater than the historical revaluation cost, the cost deviation value is positive, indicating that the new solution is worse than the historical solution. If the two are equal, the cost deviation value is zero.

[0121] Step S403: When the cost deviation value meets the preset cost deviation condition, the previous round of historical scheduling scheme is used as the target collaborative scheduling scheme for the current scheduling cycle.

[0122] In this embodiment, the preset cost deviation condition refers to a pre-set dual threshold criterion used to determine whether to reuse a historical scheme, including two dimensions: a relative threshold and an absolute threshold. Both must be satisfied simultaneously to trigger the reuse of the historical scheme. This criterion can be expressed as a stability protection criterion:

[0123] Where Clast is the cost of revaluing the previous round's solution (i.e., the historical revaluation cost), Cnew is the cost of the optimal solution in this round (i.e., the current revaluation cost), α is the preset relative threshold, and Δ is the preset absolute threshold.

[0124] As an example, the preset relative threshold coefficient α can be 0.9 (that is, the cost of the new solution does not exceed 90% of the cost of the original solution), and the preset absolute threshold Δ can be 20 (that is, the cost of the new solution does not exceed the difference between the cost of the original solution and 20). The judgment condition is: the cost of the new solution ≤ 0.9 × the cost of the original solution and the cost of the new solution ≤ the cost of the original solution - 20. When both conditions are met, the reuse of the historical solution is triggered.

[0125] The relative threshold refers to the upper limit of the ratio of the current revaluation cost to the historical revaluation cost, that is, the requirement that Cnew must meet the following conditions. (equivalent to) (Reuse conditions are triggered when the condition is met); the absolute threshold refers to the lower limit of the absolute value of the difference between the current revaluation cost and the historical revaluation cost, i.e., the requirement... (equivalent to) (Reuse conditions are triggered when the scheme is established); the target collaborative scheduling scheme refers to the scheduling scheme that is finally determined as the actual output of the current scheduling cycle after stability protection judgment. This scheme may be the collaborative scheduling scheme obtained in this round of search (global optimal solution or local optimal solution), or it may be a reused historical scheduling scheme from the previous round.

[0126] After calculating the cost deviation value, first check whether the cost deviation value simultaneously meets two conditions: Condition 1 is that the current revaluation cost is less than or equal to the product of the historical revaluation cost and the preset relative threshold coefficient (i.e., Condition two is that the current revaluation cost is less than or equal to the historical revaluation cost minus a preset absolute threshold (i.e., The above two conditions are equivalent to the stability protection criterion Clast≤min(α). Cnew,Cnew+Δ), that is, the revaluation cost of the previous scheme does not exceed (Cnew,Cnew+Δ). and The smaller of the two conditions triggers the reuse of historical solutions. If both conditions are met, it means that the improvement of the new solution is significant enough (both in terms of relative optimization and in terms of absolute value exceeding the threshold), and the previous round's historical scheduling solution is used as the target collaborative scheduling solution output to maintain scheduling stability. If either condition is not met, it means that the improvement of the new solution is limited or inferior to the historical solution, and the collaborative scheduling solution obtained in this round of search is used as the target collaborative scheduling solution output.

[0127] By obtaining the current re-estimation cost of the collaborative scheduling scheme and the historical re-estimation cost of the previous historical scheduling scheme when the robot request set meets the preset stability conditions, a unified cost model is used to fairly quantify and compare the costs of the new and old schemes in a stable request scenario, providing an objective economic basis for stability decisions. By calculating the cost deviation between the current re-estimation cost and the historical re-estimation cost, the improvement of the new scheme relative to the historical scheme is quantified by the cost difference, providing a quantitative basis for whether to switch schemes. By using the previous historical scheduling scheme as the target collaborative scheduling scheme for the current scheduling cycle when the cost deviation meets the preset cost deviation conditions, the scheme is switched only when the improvement crosses both the relative threshold and the absolute threshold, filtering out reassignment behaviors with limited benefits but causing frequent path switching, achieving a balance between scheduling stability and scheme excellence, and reducing robot execution jitter and on-site disturbances.

[0128] This embodiment also provides a collaborative scheduling and planning device for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0129] This embodiment provides a collaborative scheduling and planning device, such as... Figure 5 As shown, it includes: The acquisition module 51 is used to acquire the planning input data of the current scheduling cycle. The planning input data includes: robot request set, stability state information corresponding to the robot request set, and the number of unfinished tasks left over from the previous scheduling cycle. Module 52 is selected to select the target solution time limit for the current scheduling period from the preset solution time limit based on stability state information and the number of unfinished tasks; The solution module 53 is used to expand the initial nodes of the target search model within the target solution time limit to obtain candidate nodes, perform multi-level pruning on the candidate nodes in a preset order to obtain effective candidate nodes, perform cost-weighted evaluation on each effective candidate node through the node evaluation function, update the global optimal solution according to the evaluation results, until the target solution time limit is reached and the search result is obtained. The target search model is used to search and solve the collaborative scheduling process between the robot and the elevator. The initial nodes and candidate nodes correspond to collaborative scheduling schemes with different execution progress. The generation module 54 is used to generate a collaborative scheduling scheme corresponding to the robot request set based on the search results.

[0130] In this embodiment of the application, the apparatus further includes: a construction module, configured to initialize an incomplete robot task set based on a robot request set; obtain elevator time axis initialization parameters and robot time axis initialization parameters; construct an elevator time axis based on the elevator time axis initialization parameters and construct a robot time axis based on the robot time axis initialization parameters; and generate an initial node of the target search model based on the elevator time axis, the robot time axis, and the incomplete robot task set, wherein the initial node corresponds to an initial scheduling scheme in which no robot request has been executed.

[0131] In this embodiment, the solving module 53 is specifically used to obtain the node cost of the candidate node and the cost of the currently recorded known optimal solution; perform first-level pruning: when the node cost of the candidate node is greater than a preset relative multiple of the cost of the known optimal solution, the search of the search queue where the candidate node is located is terminated; if the candidate node passes the first-level pruning, then perform second-level pruning: when the node cost of the candidate node or its child node is greater than the cost of the known optimal solution, the corresponding branch is skipped; if the candidate node passes the second-level pruning, then perform third-level pruning: obtain the expansion action corresponding to the candidate node, and when the expansion action is not... If the first-step constraint in the preset feasible solution cache library is satisfied, the corresponding expansion is skipped. The preset feasible solution cache library is generated after path pre-calculation based on the elevator reachable area set and the region connectivity relationship. The first-step constraint is the first-step operation of the expansion action, which is located in the preset feasible solution cache library. If the candidate node passes the third-level pruning, the fourth-level pruning is performed: the access area record corresponding to the candidate node is obtained. When the access area record contains a repeatedly accessed intermediate area, the corresponding expansion is skipped. The nodes retained after the first-level pruning, second-level pruning, third-level pruning, and fourth-level pruning are taken as valid candidate nodes.

[0132] In this embodiment of the application, the device further includes: a calculation module, used to obtain the action to be executed corresponding to the valid candidate node; based on the pre-constructed elevator timeline and robot timeline, determine the corresponding target elevator and target robot according to the action to be executed; obtain the cumulative elevator time on the elevator timeline and the cumulative robot time on the robot timeline; calculate the single-step completion time of the action to be executed according to the cumulative elevator timeline, the cumulative robot timeline and the stage cost corresponding to the action to be executed, and generate a new valid candidate node based on the single-step completion time.

[0133] In this embodiment of the application, the solution module 53 is specifically used to obtain the cumulative cost of the effective candidate node, and to perform heuristic cost estimation calculation on the remaining requests corresponding to the effective candidate node to obtain the heuristic estimated cost; to perform a weighted summation of the cumulative cost and the heuristic estimated cost to obtain the node cost of the effective candidate node; when the node cost of the effective candidate node is better than the cost of the currently recorded known optimal solution, the effective candidate node is updated to the global optimal solution.

[0134] In this embodiment of the application, module 52 is specifically used to select a first preset solution time limit as the target solution time limit when the stability state information meets the preset stability conditions and the number of unfinished tasks is less than a preset number threshold; and to select a second preset solution time limit as the target solution time limit when the stability state information does not meet the preset stability conditions or the number of unfinished tasks is greater than or equal to the preset number threshold, wherein the second preset solution time limit is greater than the first preset solution time limit.

[0135] In this embodiment of the application, the generation module 54 is specifically used to detect whether the global optimal solution in the search results covers all robot requests in the robot request set; if the global optimal solution covers all robot requests in the robot request set, the global optimal solution is output as a collaborative scheduling scheme; or, if the global optimal solution does not cover all robot requests in the robot request set, the intermediate candidate node that covers the most robot requests in the robot request set is selected from the candidate nodes visited during the search process, and the local optimal solution corresponding to the intermediate candidate node is output as a collaborative scheduling scheme.

[0136] In this embodiment, the device further includes: a traversal module, used to traverse the candidate foreground point set of the robot, obtain the real-time path cost corresponding to each candidate foreground point in the candidate foreground point set; compare the real-time path costs corresponding to each candidate foreground point to obtain the comparison result; select the target foreground point with the best cost in the current scheduling cycle from the candidate foreground point set according to the comparison result; integrate the target foreground point into the planning input data of the current scheduling cycle, and perform subsequent search planning based on the target foreground point.

[0137] In this embodiment of the application, the device further includes: a revaluation module, used to obtain the current revaluation cost corresponding to the collaborative scheduling scheme and the historical revaluation cost corresponding to the previous historical scheduling scheme when the stability state information of the robot request set meets the preset stability conditions; calculate the cost deviation value between the current revaluation cost and the historical revaluation cost; and when the cost deviation value meets the preset cost deviation condition, use the previous historical scheduling scheme as the target collaborative scheduling scheme for the current scheduling cycle.

[0138] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 6 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system).

[0139] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0140] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0141] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device as shown by a landing page for an app. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0142] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0143] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0144] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0145] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A collaborative scheduling planning method, characterized in that, The method includes: Obtain the planning input data for the current scheduling cycle, wherein the planning input data includes: robot request set, stability state information corresponding to the robot request set, and the number of unfinished tasks left over from the previous scheduling cycle; Based on the stability state information and the number of unfinished tasks, the target solution time limit for the current scheduling period is selected from the preset solution time limits; Within the target solution time limit, the initial nodes of the target search model are expanded to obtain candidate nodes. Multi-level pruning is performed on the candidate nodes in a preset order to obtain effective candidate nodes. Each effective candidate node is evaluated by cost weighting through a node evaluation function. The global optimal solution is updated according to the evaluation results until the target solution time limit is reached to obtain the search result. The target search model is used to search and solve the collaborative scheduling process between the robot and the elevator. The initial nodes and the candidate nodes correspond to collaborative scheduling schemes with different execution progress. Generate a collaborative scheduling scheme corresponding to the set of robot requests based on the search results.

2. The method according to claim 1, characterized in that, Before expanding the initial nodes of the target search model to obtain candidate nodes, the method further includes: Initialize the set of incomplete robot tasks based on the robot request set; Obtain the elevator time axis initialization parameters and the robot time axis initialization parameters; An elevator timeline is constructed based on the elevator timeline initialization parameters, and a robot timeline is constructed based on the robot timeline initialization parameters. Based on the elevator timeline, the robot timeline, and the set of unfinished robot tasks, an initial node for the target search model is generated, wherein the initial node corresponds to an initial scheduling scheme in which no robot requests have been executed.

3. The method according to claim 1, characterized in that, The step of performing multi-level pruning on the candidate nodes according to a preset order to obtain valid candidate nodes includes: Obtain the node cost of the candidate node and the cost of the currently recorded known optimal solution; Perform the first level of pruning: when the node cost of the candidate node is greater than a preset relative multiple of the cost of the known optimal solution, terminate the search in the search queue where the candidate node is located; If the candidate node passes the first-level pruning, then the second-level pruning is performed: when the node cost of the candidate node or its child node is greater than the cost of the known optimal solution, the corresponding branch is skipped; If the candidate node passes the second-level pruning, then the third-level pruning is performed: obtain the extension action corresponding to the candidate node, and skip the corresponding extension when the extension action does not meet the first-step constraint in the preset feasible solution cache library. The preset feasible solution cache library is generated after path pre-calculation based on the elevator reachable area set and the area connectivity relationship. The first-step constraint is that the first step operation of the extension action is located in the preset feasible solution cache library. If the candidate node passes the third-level pruning, then the fourth-level pruning is performed: the access region record corresponding to the candidate node is obtained, and when the access region record contains a repeatedly accessed intermediate region, the corresponding expansion is skipped; The nodes retained after the first-level pruning, the second-level pruning, the third-level pruning, and the fourth-level pruning are considered as the valid candidate nodes.

4. The method according to claim 1, characterized in that, After obtaining valid candidate nodes, the method further includes: Obtain the action to be executed corresponding to the valid candidate node; Based on the pre-built elevator timeline and robot timeline, the corresponding target elevator and target robot are determined according to the action to be executed; Obtain the cumulative elevator time on the elevator timeline and the cumulative robot time on the robot timeline for the target elevator; Based on the cumulative time of the elevator, the cumulative time of the robot, and the cost of each stage corresponding to the action to be executed, the single-step completion time of the action to be executed is calculated, and new valid candidate nodes are generated based on the single-step completion time.

5. The method according to claim 4, characterized in that, The step of performing a cost-weighted evaluation of each of the effective candidate nodes using a node evaluation function, and updating the global optimal solution based on the evaluation results, includes: Obtain the cumulative cost of the effective candidate nodes, and perform heuristic cost estimation calculation on the remaining requests corresponding to the effective candidate nodes to obtain the heuristic estimated cost; The cumulative cost and the heuristically estimated cost are weighted and summed to obtain the node cost of the effective candidate node; When the node cost of the effective candidate node is better than the cost of the currently recorded known optimal solution, the effective candidate node is updated to the global optimal solution.

6. The method according to claim 1, characterized in that, The step of selecting the target solution time limit for the current scheduling period from the preset solution time limits based on the stability state information and the number of unfinished tasks includes: When the stability state information meets the preset stability condition and the number of unfinished tasks is less than the preset number threshold, the first preset solution time limit is selected from the preset solution time limits as the target solution time limit; When the stability state information does not meet the preset stability condition or the number of unfinished tasks is greater than or equal to the preset number threshold, a second preset solution time limit is selected from the preset solution time limits as the target solution time limit, wherein the second preset solution time limit is greater than the first preset solution time limit.

7. The method according to claim 1, characterized in that, The step of generating a collaborative scheduling scheme corresponding to the robot request set based on the search results includes: Detect whether the globally optimal solution in the search results covers all robot requests in the robot request set; If the global optimal solution covers all robot requests in the robot request set, then the global optimal solution is output as the cooperative scheduling scheme. Alternatively, if the global optimal solution does not cover all robot requests in the robot request set, then the intermediate candidate node that covers the most robot requests in the robot request set is selected from the candidate nodes visited during the search process, and the local optimal solution corresponding to the intermediate candidate node is output as the cooperative scheduling scheme.

8. The method according to claim 1, characterized in that, After obtaining the planning input data for the current scheduling period, the method further includes: Traverse the robot's candidate forward point set and obtain the real-time path cost corresponding to each candidate forward point in the candidate forward point set; The real-time path costs corresponding to each candidate forward point are compared to obtain the comparison results. Based on the comparison results, the target forward point with the best cost within the current scheduling period is selected from the candidate forward point set; The target precursor is integrated into the planning input data of the current scheduling cycle, and subsequent search planning is performed based on the target precursor.

9. The method according to claim 1, characterized in that, After generating a collaborative scheduling scheme corresponding to the robot request set based on the search results, the method further includes: When the stability state information of the robot request set meets the preset stability conditions, the current revaluation cost corresponding to the collaborative scheduling scheme and the historical revaluation cost corresponding to the previous historical scheduling scheme are obtained. Calculate the cost deviation between the current revaluation cost and the historical revaluation cost; When the cost deviation value meets the preset cost deviation condition, the previous round of historical scheduling scheme is used as the target collaborative scheduling scheme for the current scheduling cycle.

10. A collaborative scheduling and planning device, characterized in that, The device includes: The acquisition module is used to acquire the planning input data for the current scheduling cycle, wherein the planning input data includes: robot request set, stability state information corresponding to the robot request set, and the number of unfinished tasks left over from the previous scheduling cycle; The selection module is used to select the target solution time limit of the current scheduling period from the preset solution time limits based on the stability state information and the number of unfinished tasks. The solution module is used to expand the initial nodes of the target search model within the target solution time limit to obtain candidate nodes, perform multi-level pruning on the candidate nodes in a preset order to obtain effective candidate nodes, perform cost-weighted evaluation on each of the effective candidate nodes through a node evaluation function, update the global optimal solution according to the evaluation results, until the target solution time limit is reached to obtain the search result. The target search model is used to search and solve the collaborative scheduling process between the robot and the elevator, and the initial nodes and the candidate nodes correspond to collaborative scheduling schemes with different execution progress. The generation module is used to generate a collaborative scheduling scheme corresponding to the robot request set based on the search results.

11. A computer device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method of any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the method of any one of claims 1 to 9.

Images