Methods and systems for task scheduling and conflict avoidance control of multiple service robot clusters

By integrating multi-source big data and improving the ant colony optimization algorithm, the problem of resource contention and conflict among multiple service robot clusters in a building was solved, achieving efficient task allocation and stable operation of the robot cluster, and improving resource utilization and business data support capabilities.

CN123114706APending Publication Date: 2026-11-17SUQIAN WOKE DIGITAL INTELLIGENT TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611595711.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-15
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In smart hotel and smart office building scenarios, multiple service robot clusters frequently encounter conflicts over limited shared resources such as elevators, turnstiles, and access control systems within the building. Existing scheduling algorithms lack conflict prediction in both time and space dimensions, leading to task failures and low resource utilization. Furthermore, the elevator renovation conditions vary across different buildings, resulting in weak adaptability of existing scheduling algorithms and a lack of full-link data statistics.

Method used

By fusion of multi-source big data, a multi-dimensional constraint task allocation algorithm is constructed. Conflict prediction is performed based on time-space binary sequences. Combined with an improved ant colony optimization algorithm and POMDP model, intelligent task allocation and conflict avoidance of robot clusters are realized. Data docking between two types of ladder control modes is supported, and full-link business data statistics are performed.

Benefits of technology

It improves the delivery efficiency and operational stability of robot clusters, reduces the probability of robot congestion and task failure, increases resource utilization, and supports rapid deployment in multi-building environments and reverse optimization of business data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123114706A_ABST
    Figure CN123114706A_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for task scheduling and conflict avoidance control of a multi-service robot cluster, belonging to the field of robot big data scheduling algorithm technology. The method involves initializing and constructing a multi-story building static dataset and configuring elevator control docking mode; periodically collecting multi-source data to obtain a standardized dataset; filtering robots based on multi-dimensional constraints such as fault, load, and battery life to intelligently allocate delivery tasks; predicting spatiotemporal conflicts by converting robot path transformation position-time binary sequences to identify resource contention conflicts such as access channels, elevators, and turnstiles; resolving conflicts by performing path replanning, avoidance waiting, and resource queuing according to the conflict type; completing cross-floor delivery by robots, and allowing users to pick up their items after identity verification; and finally, outputting a business analysis dataset based on the full-link business data. This invention optimizes task allocation, improves cluster operation stability, and is suitable for robot delivery scenarios in smart hotels and office buildings.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of robot big data scheduling algorithm technology, specifically involving a method and system for task scheduling and conflict avoidance control of multiple service robot clusters. Background Technology

[0002] In smart hotel and smart office building scenarios, in order to increase the throughput of delivery services, multiple service robots are often deployed inside the building to form a robot cluster, which together complete the cross-floor delivery of customer-needed materials, takeout, and office supplies.

[0003] However, existing technologies have the following shortcomings: First, elevators, turnstiles, and access control systems in buildings are limited shared resources. When multiple robots work in parallel, conflicts such as competing for passage, elevators, and turnstiles and access control systems applying for passage at the same time are likely to occur. Existing scheduling systems mostly only perform simple path planning and lack big data conflict prediction based on the two-dimensional time-space dimension. After a conflict occurs, the robot stops and causes the task to fail.

[0004] Second, traditional task allocation logic mostly only considers distance factors and lacks multi-dimensional constraints and filtering that integrates robot endurance, cabin load capacity, and fault status. This can easily lead to some robots being overloaded and running out of power, while others remain idle for a long time, resulting in low utilization of cluster resources.

[0005] Third, the conditions for elevator renovation vary from building to building, and the existing scheduling algorithm has weak adaptability to different elevator control modes; there is a lack of data tracking statistics for the entire delivery chain, making it impossible to support the optimization of building material operations based on business data.

[0006] Therefore, there is a need for a control method that enables intelligent task allocation based on multi-source big data for clusters of building service robots, and can identify and resolve conflicts of various shared resources. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method and system for task scheduling and conflict avoidance control of multiple service robot clusters. This method constructs a multi-dimensional constrained task allocation algorithm through multi-source heterogeneous data fusion; performs conflict prediction based on time-space binary sequences and executes conflict resolution strategies in a hierarchical manner; is compatible with two types of ladder control docking modes; and simultaneously completes full-link business data statistics, thereby improving the delivery efficiency and operational stability of the robot cluster.

[0008] To achieve the above objectives, this invention provides a method for task scheduling and conflict avoidance control of a cluster of multiple service robots, comprising the following steps: S1. System initialization, building a static prior knowledge base for the building; S2 receives the building static prior knowledge base from S1, collects the original hardware observation data streams of robots, elevators, and turnstiles, uses fixed-time sliding window stream processing to suppress noise in the observation data, and iteratively updates to obtain the belief state. S3, based on the belief states of S2 and the static prior knowledge base of the building in S1, constructs a tuple model for multi-coupled shared resources in the building; the state space embeds the robot, the queue of tasks to be processed, and the occupancy states of multiple shared resources such as corridors, elevators, and turnstiles; the action space is defined as a set of joint actions, which includes robot-task assignment sub-actions and conflict resolution sub-actions. A reward function is set to maximize the long-term cumulative discount reward as the model optimization objective; S4. Based on the tuple model of S3 and the belief state of S2, the globally optimal joint action set is obtained by solving the problem. During the solution process, the pheromone volatilization degree is adaptively adjusted according to the cluster conflict risk corresponding to the belief state, and the globally optimal joint action set carrying both task allocation instructions and robot conflict avoidance control instructions is output. S5. Deduce the resource occupancy relationship corresponding to the globally optimal joint action set in S4 to construct a resource waiting graph. Quantify and calculate the deadlock risk index based on the resource waiting graph. When the deadlock risk index meets the high-risk deadlock judgment condition, discard the current globally optimal joint action set, solve the new joint action set, complete the deadlock loop resolution, and output the joint action set after deadlock verification. S6. Issue the set of joint actions after deadlock verification in S5 to drive the building delivery robot to perform delivery business; receive the hardware observation message returned by the business execution, and send the returned hardware observation message back to step S2 to realize the closed-loop iterative update of belief state.

[0009] Preferably, in S2, the belief state is the belief state of a partially observable Markov decision process (POMDP), which is updated using a Bayesian iterative update method, combined with a fixed-time sliding window to filter effective historical observations to achieve noise suppression, and the output belief state is subjected to probability normalization verification.

[0010] Preferably, in S3, the tuple model is an improved POMDP quintuple model, and the quintuple includes a state space, an action space, an observation space, a state transition function, and a reward function; the state transition function is configured with constraints, such that only one robot is allowed to occupy a single elevator or a single turnstile at the same time.

[0011] Preferably, in S4, an improved ant colony optimization algorithm is used for solving; the cluster conflict risk is obtained by weighted fusion of risk components corresponding to three types of coupled shared resources: corridors, elevators, and turnstiles. Cluster conflict risk meets the following calculation conditions: ; In the formula, Normalization factor; For the collection of corridor resources; For elevator resource aggregation; For the collection of turnstile access control resources; This is a function representing the risk component of spatiotemporal overlap of passageway paths; A component function for elevator contention conflict risk; This is a component function representing the risk of gate access control preemption conflicts. The risk weighting coefficient for the passageway; This refers to the risk weighting coefficient for elevators. This refers to the risk weighting coefficient for the turnstile access control system.

[0012] Preferably, in S4, the adaptive adjustment of pheromone evaporation satisfies the following update conditions: ; In the formula, The adaptive pheromone evaporation coefficient; The basic volatility coefficient of pheromones; This represents the original value for cluster conflict risk. This is the risk adjustment coefficient; Action path at time t Pheromone concentration; This represents the pheromone increment generated during ant iteration.

[0013] Preferably, in S3, the reward function is composed of a weighted average of the task's positive reward, waiting cost, conflict penalty, and deadlock risk penalty. The reward function satisfies the following calculation conditions: ; In the formula, For single-step returns; Positive rewards for task completion; The cost of the robot's waiting time; Penalties for shared resource conflicts; This is a penalty item for potential deadlock risk; The non-negative weighting coefficient for the positive return term; The non-negative weighting coefficient for the waiting time cost term; The non-negative weight coefficient for the shared resource conflict penalty term; This is the non-negative weighting coefficient for the potential deadlock risk penalty term.

[0014] Preferably, in S5, the deadlock risk indicator is the ratio of the number of resource waiting graph loops to the total number of robots participating in scheduling; when the deadlock risk indicator is greater than or equal to the preset deadlock risk judgment threshold, it is judged as a high-risk deadlock. The deadlock risk metric meets the following calculation conditions: ; In the formula, As a deadlock risk indicator; Identify the total number of loops in the resource wait graph; This represents the total number of robots in the cluster.

[0015] This invention also provides a multi-service robot cluster task scheduling and conflict avoidance control system, including: The initialization module is used to execute S1 and build a static prior knowledge base for buildings; The update module, connected to the initialization module, is used to execute S2, receive the building static prior knowledge base, collect the original hardware observation data streams of robots, elevators, and turnstile access control, use fixed-time sliding window stream processing to suppress noise in the observation data, and iteratively update to obtain the belief state; The tuple model construction module, connected to the update module, is used to execute S3. Based on the belief state and the static prior knowledge base of the building, it constructs a tuple model for multiple coupled shared resources in the building. The state space embeds the robot, the queue of tasks to be processed, and the occupancy states of multiple shared resources such as corridors, elevators, and turnstiles. The action space is defined as a set of joint actions, which includes robot-task assignment sub-actions and conflict resolution sub-actions. A reward function is set to maximize the long-term cumulative discount reward as the model optimization objective. The joint action solving module, connected to the tuple model construction module, is used to execute step S4, and solve for the globally optimal joint action set based on the tuple model and belief state; during the solving process, the pheromone volatilization degree is adaptively adjusted according to the cluster conflict risk corresponding to the belief state, and the globally optimal joint action set carrying both task allocation instructions and robot conflict avoidance control instructions is output. The deadlock risk verification and resolution module is connected to the joint action solving module. It is used to execute step S5, deduce the resource occupancy relationship corresponding to the globally optimal joint action set to construct a resource waiting graph, and quantitatively calculate the deadlock risk index based on the resource waiting graph. When the deadlock risk index meets the high-risk deadlock judgment condition, the current globally optimal joint action set is discarded, a new joint action set is solved again, the deadlock loop is resolved, and the joint action set after deadlock verification is output. The business-driven and closed-loop iteration module is connected to the deadlock risk verification and resolution module. It is used to execute step S6, issue a set of joint actions after deadlock verification, drive the building delivery robot cluster to perform delivery business; receive hardware observation messages returned by business execution, and send the returned hardware observation messages back to the observation filtering and belief update module to realize closed-loop iterative update of belief state.

[0016] Therefore, the present invention employs the above-mentioned multi-service robot cluster task scheduling and conflict avoidance control method. Compared with the prior art, the technical solution of the present invention has the following beneficial effects: (1) This invention adopts a multi-source big data fusion method, integrates multi-dimensional constraints such as robot endurance, cabin load, distance, and building facility status to construct a task allocation algorithm, so as to avoid robot overload or idleness; and relies on batch floor map processing algorithm to quickly complete multi-floor environment initialization, reduce cluster project deployment time, and improve the overall delivery throughput of multi-robot clusters. (2) The present invention adopts a time-space two-dimensional conflict prediction algorithm to identify multiple shared resource contention conflicts in building scenarios, such as passages, elevators, and turnstiles; through a hierarchical resolution decision algorithm, it distinguishes between scenario execution path replanning, priority avoidance, and resource timing queuing, and is equipped with timeout rollback logic to effectively reduce the probability of cluster robot blockage and task failure, and improve the stability of cluster operation. (3) This invention realizes the collection and statistics of business data of the entire delivery chain, outputs a standardized business analysis dataset, and can assist in the optimization of building material allocation; at the same time, the algorithm supports data docking and adaptation of two elevator control modes, with low dependence on building renovation; it realizes contactless delivery on the business side and improves the service experience of building users.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating an embodiment of the multi-service robot cluster task scheduling and conflict avoidance control method of the present invention; Figure 2 This is a flowchart illustrating the improved ant colony solution and deadlock verification process in an embodiment of the multi-service robot cluster task scheduling and conflict avoidance control method 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. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used in the present invention should have the ordinary meaning understood by those skilled in the art.

[0020] Example 1 like Figure 1 As shown, this embodiment provides a method for task scheduling and conflict avoidance control of a cluster of multiple service robots, including the following steps: S1. System initialization: Construct static prior knowledge of the building, configure the hyperparameters of the algorithms corresponding to Partially Observable Markov Decision Process (POMDP), sliding window, improved ant colony algorithm, and deadlock detection.

[0021] Complete the building static topology map and import robot hardware constraint parameters; complete the registration and access of all robots in the cluster; configure algorithm hyperparameters including sliding window duration W and POMDP discount factor. Ant colony size M, maximum number of iterations pheromone initial value pheromone basic volatility coefficient Heuristic factors Deadlock risk threshold Risk weighting coefficients for three types of shared resources .

[0022] The system allows for optional configuration of either cloud-based or hardware-based elevator control interface mode, establishing communication links between the dispatch server and elevators, turnstiles, and access control systems.

[0023] Output all initialization configuration parameters and the building static prior knowledge base, and pass them to steps S2 and S3 as the basic static constraints for belief state iteration and POMDP model construction.

[0024] S2 receives the initialization configuration parameters and building static prior knowledge base output by S1, receives the raw observation data streams from robots, elevators, and turnstiles, uses fixed-time sliding window stream processing, updates the belief state distribution of POMDP, and suppresses IoT message packet loss and observation noise.

[0025] The sliding window stream processing and POMDP belief update sub-item execution includes the following sub-steps: S201. Maintain a time sliding window with a fixed duration of W and continuously receive real-time raw observation data streams. ; Observing the new moment Pushing into the window removes expired historical observations outside the window's time boundaries, retaining only valid observation sequences within the window. .

[0026] S202, Read the belief distribution of the previous moment. The joint action performed in the previous step Obtain the observation probability State transition probability Perform iterative update calculations of belief states: ; In the formula, The environment is in state at time t. The probability of belief; This is a normalization constant; For IoT message packet loss and device observation noise; The probability of environmental state transition; This represents the distribution of beliefs at the previous moment.

[0027] S203. Complete the normalization check to ensure that the sum of the belief probabilities corresponding to all states is equal to 1, and output the complete belief state at the current moment. .

[0028] Sliding window filtering filters expired and invalid packets, reducing state transition interference caused by network jitter; outputs noise-suppressed POMDP belief states. The data is then passed to steps S3 and S4 for POMDP model inference and cluster conflict risk calculation.

[0029] S3, Receive the current belief state output by S2. The static prior knowledge base of buildings output by S1 is used to construct an improved POMDP quintuple model for multi-coupled shared resources in buildings: ; Among them, the state space Embedded robot running status, pending task queue status, and multiple shared resource occupancy markers for corridors, elevators, and turnstiles; Action space Defined as a set of joint actions, which includes both robot-task assignment sub-actions and conflict resolution sub-actions; Observation space The corresponding set of noisy observation messages reported by the hardware; T represents the state transition function. This constraint allows only one robot to occupy a single elevator / single turnstile at any given time. Construct a multi-dimensional weighted return function, expressed as: ; In the formula, For single-step returns; Positive rewards for task completion; The cost of the robot's waiting time; Penalties for shared resource conflicts; This is a penalty item for potential deadlock risk; The non-negative weighting coefficient for the positive return term; The non-negative weighting coefficient for the waiting time cost term; The non-negative weight coefficient for the shared resource conflict penalty term; This is the non-negative weighting coefficient for the potential deadlock risk penalty term.

[0030] The model optimization objective is to maximize the long-term cumulative discounted return. ; In the formula, This is the discount factor for POMDP; The strategy to be solved.

[0031] Output the complete and improved POMDP quintuple model and pass it to step S4 as the solution object for the ant colony algorithm.

[0032] S4, Improved POMDP quintuple model receiving output from S3, Belief state output from S2 The algorithm hyperparameters output by S1 are used to solve the problem by executing the adaptive pheromone weight improved ant colony optimization algorithm, and the global optimal joint action set is output.

[0033] like Figure 2 As shown, the ant colony iterative solution process involves the following sub-steps: S401, Based on Belief State Calculate the raw value of cluster conflict risk The risk components of three coupled resources—corridors, elevators, and turnstile access control—are weighted and integrated: ; In the formula, Normalization factor; For the collection of corridor resources; For elevator resource aggregation; For the collection of turnstile access control resources; This is a function representing the risk component of spatiotemporal overlap of passageway paths; A component function for elevator contention conflict risk; This is a component function representing the risk of gate access control preemption conflicts. The risk weighting coefficient for the passageway; This refers to the risk weighting coefficient for elevators. This refers to the risk weighting coefficient for the turnstile access control system.

[0034] S402. Initialize the ant population and pheromone matrix. Each ant corresponds to a set of candidate joint action sequences. The ants construct candidate joint actions based on heuristic information and the current pheromone matrix. Each set of candidate joint actions includes task assignment sub-actions and conflict resolution sub-actions.

[0035] S403, Perform adaptive pheromone evaporation update: ; In the formula, The adaptive pheromone evaporation coefficient; The basic volatility coefficient of pheromones; This represents the original value for cluster conflict risk. This is the risk adjustment coefficient; Action path at time t Pheromone concentration; The pheromone increment generated by ant iteration; the higher the risk of cluster conflict, the greater the pheromone evaporation coefficient, which accelerates the dissipation of pheromones in suboptimal action paths and avoids premature convergence.

[0036] S404, The iteration has reached the maximum number of iterations. The candidate solution with the largest cumulative reward in POMDP is selected to obtain the globally optimal joint action set; the joint action set includes both task allocation instructions and corresponding robot conflict avoidance control instructions.

[0037] Output the globally optimal set of joint actions and pass it to step S5 for deadlock risk verification.

[0038] S5, Receive the globally optimal set of joint actions output by S4, and the belief state output by S2. Deadlock risk threshold output by S1 Based on the resource waiting graph, deadlock risk indicators are quantitatively calculated to complete deadlock risk assessment and closed-loop resolution.

[0039] Deadlock risk identification and closed-loop rollback execution includes the following sub-steps: S501. Based on the globally optimal set of joint actions, deduce the expected resource occupancy relationship and construct a resource waiting graph. The graph nodes include cluster robot nodes and elevator and turnstile access control shared resource nodes. Directed edges represent the waiting dependency relationship where a robot is waiting for a certain shared resource and that resource is occupied by another robot.

[0040] S502. Traverse the resource waiting graph, detect all closed loops in the graph, and count the total number of loops. Read the total number of robots participating in the cluster scheduling. Calculate deadlock risk indicators : ; In the formula, As a deadlock risk indicator; Identify the total number of loops in the resource wait graph; This represents the total number of robots in the cluster.

[0041] S503, Deadlock Risk Branch Detection: like If the current global optimal set of joint actions is discarded, the trigger marker is sent back to S4, and the improved ant colony algorithm is called again to solve for a new set of joint actions. like If no high-risk deadlock is detected, the current set of joint actions is retained as the set of actions available after verification.

[0042] The set of joint actions after deadlock verification is output and passed to step S6 to drive the building delivery robot to perform its business.

[0043] S6 receives the set of joint actions after deadlock verification output by S5, outputs action instructions to drive the building delivery robot to perform delivery services, and receives hardware observation messages returned by the service execution.

[0044] The verified set of joint actions is sent to the delivery robots in each building of the cluster; the robots parse the sub-actions and execute the corresponding delivery navigation and conflict avoidance behaviors such as avoidance, queuing, and path replanning. After completing cross-floor navigation, elevator access, and turnstile passage, the robot can execute the following logic upon reaching the target location: notify the user, verify identity, and open the compartment to retrieve the item. The robot's local hardware sensors and IoT devices continuously report new observation messages, and the scheduling server receives the hardware observation messages returned by the business.

[0045] Output the service feedback hardware observation message and send it back to step S2 for the next round of iteration to update the POMDP belief state.

[0046] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for task scheduling and conflict avoidance control of a cluster of multiple service robots, characterized in that, Includes the following steps: S1. System initialization, building a static prior knowledge base for the building; S2 receives the building static prior knowledge base from S1, collects the original hardware observation data streams of robots, elevators, and turnstiles, uses fixed-time sliding window stream processing to suppress noise in the observation data, and iteratively updates to obtain the belief state. S3, based on the belief states of S2 and the static prior knowledge base of the building of S1, constructs a tuple model for multi-coupled shared resources in the building; the state space embeds the robot, the queue of tasks to be processed, and the occupancy states of multiple shared resources such as corridors, elevators, and turnstiles. The action space is defined as a set of joint actions, which includes both robot-task assignment sub-actions and conflict resolution sub-actions. A reward function is set to maximize the long-term cumulative discount reward as the model optimization objective. S4, based on the tuple model of S3 and the belief state of S2, solve to obtain the globally optimal set of joint actions; During the solution process, the pheromone volatilization level is adaptively adjusted according to the cluster conflict risk corresponding to the belief state, and the output is a globally optimal set of joint actions that simultaneously carries task allocation instructions and robot conflict avoidance control instructions. S5. Deduce the resource occupancy relationship corresponding to the globally optimal joint action set in S4 to construct a resource waiting graph. Quantify and calculate the deadlock risk index based on the resource waiting graph. When the deadlock risk index meets the high-risk deadlock judgment condition, discard the current globally optimal joint action set, solve the new joint action set, complete the deadlock loop resolution, and output the joint action set after deadlock verification. S6. Issue the set of joint actions after deadlock verification in S5 to drive the building delivery robot to perform delivery business; receive the hardware observation message returned by the business execution, and send the returned hardware observation message back to step S2 to realize the closed-loop iterative update of belief state.

2. The method for task scheduling and conflict avoidance control of a multi-service robot cluster according to claim 1, characterized in that, In S2, the belief state is the belief state of a partially observable Markov decision process (POMDP). It adopts a Bayesian iterative update method, combined with a fixed-time sliding window to filter effective historical observations to achieve noise suppression, and performs probability normalization verification on the output belief state.

3. The method for task scheduling and conflict avoidance control of a multi-service robot cluster according to claim 2, characterized in that, In S3, the tuple model is an improved POMDP quintuple model. The quintuple includes a state space, an action space, an observation space, a state transition function, and a reward function. The state transition function is configured with constraints, such that only one robot is allowed to occupy a single elevator or a single turnstile at any given time.

4. The method for task scheduling and conflict avoidance control of a multi-service robot cluster according to claim 3, characterized in that, In S4, an improved ant colony optimization algorithm is used for solving the problem; the cluster conflict risk is obtained by weighted fusion of risk components corresponding to three types of coupled shared resources: corridors, elevators, and turnstiles. Cluster conflict risk meets the following calculation conditions: ; In the formula, Normalization factor; For the collection of corridor resources; For elevator resource aggregation; This is a collection of turnstile access control resources; This is a function representing the risk component of spatiotemporal overlap of passageway paths; A component function for elevator contention conflict risk; This is a component function for the risk of gate access control preemption conflicts; The risk weighting coefficient for the passageway; The risk weighting coefficient for elevators; This refers to the risk weighting coefficient for the turnstile access control system.

5. The method for task scheduling and conflict avoidance control of a multi-service robot cluster according to claim 4, characterized in that, In S4, the adaptive adjustment of pheromone evaporation satisfies the following update condition: ; In the formula, The adaptive pheromone evaporation coefficient; The basic volatility coefficient of pheromones; This represents the original value for cluster conflict risk. This is the risk adjustment coefficient; Action path at time t Pheromone concentration; This represents the pheromone increment generated during ant iteration.

6. The method for task scheduling and conflict avoidance control of a cluster of multiple service robots according to claim 5, characterized in that, In S3, the reward function is a weighted sum of the task's positive reward, waiting cost, conflict penalty, and deadlock risk penalty. The reward function satisfies the following calculation conditions: ; In the formula, For single-step returns; Positive rewards for task completion; The cost of the robot's waiting time; Penalties for shared resource conflicts; This is a penalty item for potential deadlock risk; The non-negative weighting coefficient for the positive return term; The non-negative weighting coefficient for the waiting time cost term; The non-negative weight coefficient for the shared resource conflict penalty term; This is the non-negative weighting coefficient for the potential deadlock risk penalty term.

7. The method for task scheduling and conflict avoidance control of a cluster of multiple service robots according to claim 6, characterized in that, In S5, the deadlock risk indicator is the ratio of the number of resource waiting graph loops to the total number of robots participating in scheduling; when the deadlock risk indicator is greater than or equal to the preset deadlock risk judgment threshold, it is judged as a high-risk deadlock. The deadlock risk metric meets the following calculation conditions: ; In the formula, As a deadlock risk indicator; Identify the total number of loops in the resource wait graph; This represents the total number of robots in the cluster.

8. A multi-service robot cluster task scheduling and conflict avoidance control system, applied to the multi-service robot cluster task scheduling and conflict avoidance control method according to any one of claims 1 to 7, characterized in that, include: The initialization module is used to execute S1 and build a static prior knowledge base for buildings; The update module, connected to the initialization module, is used to execute S2, receive the building static prior knowledge base, collect the original hardware observation data streams of robots, elevators, and turnstile access control, use fixed-time sliding window stream processing to suppress noise in the observation data, and iteratively update to obtain the belief state; The tuple model construction module, connected to the update module, is used to execute S3, constructing a tuple model for multiple coupled shared resources in the building based on belief states and the building's static prior knowledge base; the state space embeds the robot, the queue of tasks to be processed, and the occupancy states of multiple shared resources such as corridors, elevators, and turnstiles. The action space is defined as a set of joint actions, which includes both robot-task assignment sub-actions and conflict resolution sub-actions. A reward function is set to maximize the long-term cumulative discount reward as the model optimization objective. The joint action solving module is connected to the tuple model construction module and is used to execute step S4 to solve for the globally optimal set of joint actions based on the tuple model and belief state. During the solution process, the pheromone volatilization level is adaptively adjusted according to the cluster conflict risk corresponding to the belief state, and the output is a globally optimal set of joint actions that simultaneously carries task allocation instructions and robot conflict avoidance control instructions. The deadlock risk verification and resolution module is connected to the joint action solving module. It is used to execute step S5, deduce the resource occupancy relationship corresponding to the globally optimal joint action set to construct a resource waiting graph, and quantitatively calculate the deadlock risk index based on the resource waiting graph. When the deadlock risk index meets the high-risk deadlock judgment condition, the current globally optimal joint action set is discarded, a new joint action set is solved again, the deadlock loop is resolved, and the joint action set after deadlock verification is output. The business-driven and closed-loop iteration module is connected to the deadlock risk verification and resolution module. It is used to execute step S6, issue a set of joint actions after deadlock verification, drive the building delivery robot cluster to perform delivery business; receive hardware observation messages returned by business execution, and send the returned hardware observation messages back to the observation filtering and belief update module to realize closed-loop iterative update of belief state.