Distributed scheduling method and device for large-scale autonomous mobile robot cluster

Abstract

The invention provides a distributed scheduling method and device for a large-scale autonomous mobile robot cluster. The method comprises the following steps: receiving a scheduling related message sent by a first mobile robot; the scheduling related message comprises an identifier of the first mobile robot and is used for indicating the first mobile robot to request a scheduling decision and / or arrive at a target position; based on the scheduling related message, scheduling the first mobile robot; wherein the first mobile robot is any one mobile robot in an autonomous mobile robot cluster. Therefore, the action scheduling execution efficiency of the large-scale autonomous mobile robot cluster is optimized, and the efficient collaborative operation of the large-scale cluster is ensured.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a distributed scheduling method and device for a large-scale autonomous mobile robot cluster. Background Art

[0002] Mobile robot swarms are already being used in a variety of fields, including express delivery and logistics, domestic services, autonomous driving, and the military. Relying on the number of robots and their coordination, swarm systems can overcome the limitations of individual robots and significantly improve system efficiency. Multi-robot systems are a future trend. In life service sectors like warehousing and logistics, the increase in the number of service robots and their improved collaboration will significantly boost social productivity and improve people's living standards. In the autonomous driving sector, as the number of self-driving cars on the road increases, the future will inevitably face the problem of planning and coordinating large-scale autonomous vehicles. In the military and defense sector, as more and more autonomous robots are deployed on the battlefield, their collaborative capabilities will directly determine victory or defeat. In short, multi-robot systems and their collaborative capabilities are key areas for future research, and mobile robot swarm systems have broad application value.

[0003] In multi-robot swarm systems, the ability to plan collision-free paths to different target locations is one of the most intuitive, fundamental, and crucial capabilities of multi-robot collaboration. However, traditional centralized planning methods approach this problem as an optimization or search problem, resulting in poor scalability and requiring significant computational resources and time for large robot swarms. Summary of the Invention

[0004] The present invention provides a distributed scheduling method and device for large-scale autonomous mobile robot clusters, which is used to solve the defect in the prior art that large-scale robot clusters require a large amount of computing resources and computing time for path planning, and realize efficient and adaptive large-scale mobile robot cluster management.

[0005] The present invention provides a distributed scheduling method for a large-scale autonomous mobile robot cluster, comprising the following steps: receiving a scheduling-related message sent by a first mobile robot; the scheduling-related message includes an identifier of the first mobile robot and is used to instruct the first mobile robot to request a scheduling decision and / or arrive at a target location; Scheduling the first mobile robot based on the scheduling-related message; The first mobile robot is any mobile robot in an autonomous mobile robot cluster.

[0006] In some embodiments, scheduling the first mobile robot based on the scheduling-related message includes: In a case where the scheduling-related message is used to instruct the first mobile robot to request a scheduling decision, determining surrounding environment information of the first mobile robot based on the scheduling-related message; the surrounding environment information of the first mobile robot includes information of all nodes within a preset range centered on the first mobile robot; determining a scheduling action for the first mobile robot to a next target location based on the surrounding environment information, and instructing the first mobile robot to execute the scheduling action; In a case where the scheduling-related message is used to instruct the first mobile robot to arrive at a target location, updating a discretized map and information about a shortest path for the first mobile robot to reach a destination based on the scheduling-related message; The discretized map is a four-connected grid graph obtained by discretizing the actual autonomous mobile robot cluster scheduling scenario, and is composed of multiple nodes and multiple edges.

[0007] In some embodiments, updating the discretized map and the optimal path information for the first mobile robot to reach the destination based on the scheduling-related message includes: updating the discretized map based on the identifier of the first mobile robot and the current discretized map; Based on the node where the first mobile robot is currently located, the optimal path information for the first mobile robot to reach the destination is updated.

[0008] In some embodiments, determining the surrounding environment information of the first mobile robot based on the scheduling-related message includes: Determining all nodes within a preset range centered on the first mobile robot based on the identifier of the first mobile robot and the current discretized map; Based on a matrix obtained from relevant information of all nodes within the preset range, the surrounding environment information of the first mobile robot is determined.

[0009] In some embodiments, determining a scheduling action for the first mobile robot to move to a next target location based on the surrounding environment information, and instructing the first mobile robot to perform the scheduling action, includes: Inputting the surrounding environment information into a trained decision model to determine the next target position of the first mobile robot and a scheduling action of the first mobile robot for the next target position; Determining whether the scheduling action is legal, and updating an illegal scheduling action to a legal scheduling action; the legal scheduling action is a scheduling action that does not cause a collision between the autonomous mobile robot cluster; Based on the identifier of the first mobile robot, the first mobile robot is instructed to perform a legal scheduling action, and the state of the node corresponding to the next target position in the discretized map is updated to be occupied by the first mobile robot.

[0010] In some embodiments, determining whether the scheduling action is legal and updating an illegal scheduling action to a legal scheduling action includes: Determining the state of the node corresponding to the next target position based on the discretized map; If the node corresponding to the next target position is in a state of an obstacle, a map boundary, or occupied by any mobile robot other than the first mobile robot, the scheduling action is determined to be illegal; if the node corresponding to the next target position is in a state of vacant, the scheduling action is determined to be legal; In the case that the scheduling action is illegal, the scheduling action is modified to a stop action.

[0011] The present invention also provides a distributed scheduling device for a large-scale autonomous mobile robot cluster, comprising the following modules: a receiving module, configured to receive a scheduling-related message sent by a first mobile robot; the scheduling-related message includes an identifier of the first mobile robot and is configured to instruct the first mobile robot to request a scheduling decision and / or to arrive at a target location; a scheduling module, configured to schedule the first mobile robot based on the scheduling-related message; The first mobile robot is any mobile robot in an autonomous mobile robot cluster.

[0012] The present invention also provides a distributed scheduling system for a large-scale autonomous mobile robot cluster, comprising: Autonomous mobile robot swarms, and a distributed scheduling device for large-scale autonomous mobile robot swarms as described above; Wherein, any mobile robot in the autonomous mobile robot cluster is used to send scheduling-related messages to the distributed scheduling device and accept scheduling by the distributed scheduling device.

[0013] The present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the distributed scheduling method for a large-scale autonomous mobile robot cluster as described above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the distributed scheduling method for a large-scale autonomous mobile robot cluster as described in any one of the above.

[0015] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-described distributed scheduling methods for large-scale autonomous mobile robot clusters.

[0016] The distributed scheduling method and device for large-scale autonomous mobile robot swarms provided by this invention utilizes a method where each mobile robot autonomously sends a scheduling-related message containing a unique identifier, instructing it to request a scheduling decision and / or reach a target location. Scheduling decisions are dynamically allocated based on the received messages, and the identities of individual mobile robots are identified and scheduled accordingly, achieving coordination between global tasks and local execution. This optimizes the efficiency of action scheduling and execution in large-scale autonomous mobile robot swarms, ensuring efficient collaborative operation of the swarm. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 It is a flow chart of the distributed scheduling method for a large-scale autonomous mobile robot cluster provided by the present invention.

[0019] Figure 2 It is a flow chart of an embodiment of a large-scale autonomous mobile robot cluster distributed scheduling method provided by the present invention.

[0020] Figure 3 It is a schematic diagram of the discretized map of the mobile robot provided by the present invention.

[0021] Figure 4 It is a schematic diagram of the field of view and surrounding environment of the mobile robot provided by the present invention.

[0022] Figure 5 It is a structural diagram of the decision neural network model provided by the present invention.

[0023] Figure 6 Schematic diagram of the surrounding environment of the mobile robot provided by the present invention.

[0024] Figure 7It is a schematic diagram of an illegal action provided by the present invention.

[0025] Figure 8 It is a structural diagram of the distributed scheduling device for large-scale autonomous mobile robot clusters provided by the present invention.

[0026] Figure 9 It is a structural diagram of the distributed scheduling system for large-scale autonomous mobile robot clusters provided by the present invention.

[0027] Figure 10 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0028] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0029] Figure 1 This is a flow chart of the distributed scheduling method for a large-scale autonomous mobile robot cluster provided by the present invention. Figure 1 As shown, the method includes the following steps: Step 100: Receive a scheduling-related message sent by a first mobile robot; the scheduling-related message includes an identifier of the first mobile robot and is used to instruct the first mobile robot to request a scheduling decision and / or arrive at a target location.

[0030] Step 101: Schedule a first mobile robot based on a scheduling-related message.

[0031] The first mobile robot is any mobile robot in the autonomous mobile robot cluster.

[0032] Specifically, the execution subject of the embodiment of the present invention can be a scheduling system for scheduling each mobile robot in an autonomous mobile robot cluster. The method provided by the embodiment of the present invention realizes distributed scheduling of large-scale autonomous mobile robot clusters through message transmission between each mobile robot and the scheduling system.

[0033] First, the scheduling system receives a scheduling-related message from the first mobile robot. The scheduling-related message can be transmitted through any network communication protocol, which is not limited in the embodiment of the present invention.

[0034] It should be noted that the "first mobile robot" in the embodiments of the present invention does not refer to a specific individual, but rather to any mobile robot in a cluster of autonomous mobile robots. Scheduling-related messages sent by the first mobile robot must carry its identifier (e.g., its mobile robot number). Each mobile robot's identifier is uniquely set before scheduling begins, ensuring that the scheduling system can accurately identify the source of scheduling-related messages and implement differentiated scheduling for different mobile robots.

[0035] In addition to the identification of the first mobile robot, the scheduling-related message may include a scheduling decision request actively initiated by the first robot (for example, path planning or task assignment is required), that is, the scheduling-related message is used to indicate that the first mobile robot requests a scheduling decision; or, the scheduling-related message may include a status notification of the first mobile robot arriving at the target position, that is, the scheduling-related message is used to indicate that the first mobile robot has arrived at the target position; or, the scheduling-related message may include a status notification of the first mobile robot arriving at the target position and a scheduling decision request actively initiated (for example, path planning or task assignment is required), that is, the scheduling-related message is used to indicate that the first mobile robot has arrived at the target position and requests a scheduling decision.

[0036] When the scheduling system receives the scheduling-related message from the first mobile robot, it can proceed with the scheduling decision process. The scheduling system's scheduling decision is a response mechanism to the content of the scheduling-related message: if the scheduling-related message is used to instruct the first mobile robot to request a scheduling decision, the scheduling system can generate a scheduling instruction for the first mobile robot based on factors such as the current state of the autonomous mobile robot cluster, task priority, and environmental constraints; if the scheduling-related message is used to instruct the first mobile robot to arrive at the target location, the scheduling system can perform scheduling operations such as updating the first mobile robot's task status, releasing map node resources, or dispatching subsequent tasks to the first mobile robot.

[0037] By receiving scheduling-related messages from the first mobile robot, the scheduling system can always base its scheduling decisions on the latest robot status information, effectively avoiding scheduling conflicts caused by information lags. The entire process unifies the state changes of individual mobile robots with the coordinated control of the autonomous mobile robot cluster through the instructions in the scheduling-related messages and the identities of the included mobile robots, ensuring both individual autonomy and the overall orderliness of the system.

[0038] The distributed scheduling method for large-scale autonomous mobile robot swarms provided by this invention involves each mobile robot autonomously sending a scheduling-related message containing a unique identifier, indicating that the mobile robot is requesting a scheduling decision and / or reaching a target location. Scheduling decisions are dynamically allocated based on the received messages, and the identities of individual mobile robots are identified and scheduled accordingly, achieving coordination between global tasks and local execution. This optimizes the efficiency of action scheduling and execution in large-scale autonomous mobile robot swarms, ensuring efficient collaborative operation of the swarm.

[0039] In some embodiments, scheduling the first mobile robot based on the scheduling-related message includes: In a case where the scheduling-related message is used to instruct the first mobile robot to request a scheduling decision, determining surrounding environment information of the first mobile robot based on the scheduling-related message; the surrounding environment information of the first mobile robot includes information of all nodes within a preset range centered on the first mobile robot; Determining a scheduling action for the first mobile robot for a next target location based on surrounding environment information, and instructing the first mobile robot to perform the scheduling action; In a case where the scheduling-related message is used to instruct the first mobile robot to arrive at a target location, updating the discretized map and the shortest path information of the first mobile robot to reach the destination based on the scheduling-related message; Among them, the discretized map is a four-connected grid graph obtained by discretizing the actual autonomous mobile robot cluster scheduling scenario, which consists of multiple nodes and multiple edges.

[0040] Specifically, when the scheduling-related message is used to indicate that the first mobile robot requests a scheduling decision, the scheduling system can use the identifier of the first mobile robot in the scheduling-related message and, with the first mobile robot as the center, extract all node information within a preset range as the surrounding environment information of the first mobile robot.

[0041] It should be noted that the node in the embodiment of the present invention refers to the basic unit in the discretized map, and the preset range may refer to the grid unit within the sensor coverage area or the communication radius.

[0042] The discretized map in the embodiment of the present invention adopts a four-connected grid structure, where each node is connected only to four adjacent nodes above, below, left and right. This topological structure can both simplify the computational complexity and ensure the feasibility of path planning.

[0043] Based on the surrounding environment information of the first mobile robot, the scheduling system can use specific algorithms (such as reinforcement learning and imitation learning, etc.) to calculate the next scheduling action of the first mobile robot and feed back the instructions of the scheduling action to the first mobile robot for execution.

[0044] It can be understood that, with a four-connected grid structure, the scheduling action may include moving to four adjacent nodes and stopping moving, a total of five scheduling actions.

[0045] When a dispatch-related message indicates that the first mobile robot has reached its target location, the dispatch system triggers a map update mechanism. This mechanism corrects the node states (e.g., occupancy) in the discretized map based on the first mobile robot's actual trajectory and recalculates the shortest path for the first mobile robot to reach its destination. This dynamic update mechanism ensures the real-time nature of the discretized map information and enables subsequent scheduling decisions for each mobile robot to be optimized based on the latest discretized map and each mobile robot's shortest path information.

[0046] In some embodiments, updating the discretized map and the optimal path information for the first mobile robot to reach the destination based on the scheduling-related message includes: updating the discretized map based on the identifier of the first mobile robot and the current discretized map; Based on the node where the first mobile robot is currently located, the optimal path information for the first mobile robot to reach the destination is updated.

[0047] Specifically, when the scheduling system receives a scheduling-related message indicating that the first mobile robot has arrived at the target location, it may first perform a map update based on the unique identifier of the first mobile robot and the current discretized map state.

[0048] Based on this dispatch-related information, the scheduling system can determine the real-time status of the first mobile robot's previous and current nodes and modify the corresponding node attributes in the discretized map grid, for example, marking the first mobile robot's current node as "occupied." This process is implemented using a four-connected grid topology, ensuring that connectivity constraints between adjacent nodes are maintained.

[0049] After the map is updated, the scheduling system can recalculate the optimal path for the first mobile robot to reach the destination based on the current node location of the first mobile robot. It should be noted that "optimal" here refers to the shortest path the first mobile robot can take to reach the destination without considering obstacles from other mobile robots. This "optimal path" can be calculated using the A* algorithm.

[0050] It should be noted that the "optimal path" can serve as important information when making the next action decision. Each mobile robot not only considers its own optimal path when making the next action decision, but also considers the optimal paths of other mobile robots in the surrounding area, thereby achieving a balance between self-interest and altruism in decision-making and improving the overall efficiency of the system. The process of calculating the optimal path is relatively time-consuming, while updating the map itself requires less computation and does not require high real-time performance. Therefore, in the embodiment of the present invention, the process of calculating the optimal path is processed together with the map update. In addition, the optimal path information can be stored in the form of a matrix. The matrix has the same shape as the discretized map, and points not on the path are marked as 0, while points on the path are marked with the distance between the point and the target.

[0051] In some embodiments, determining the surrounding environment information of the first mobile robot based on the scheduling-related message includes: Determine all nodes within a preset range centered on the first mobile robot based on the identifier of the first mobile robot and the current discretized map; The surrounding environment information of the first mobile robot is determined based on a matrix obtained based on the relevant information of all nodes within a preset range.

[0052] Specifically, in determining the surrounding environment, the scheduling system first determines all nodes within a preset range centered on the first mobile robot based on the first mobile robot's identifier and the current discretized map. The preset range can refer to a grid area corresponding to the mobile robot's field of view or communication coverage.

[0053] After locating the node where the first mobile robot is located through the identifier of the first mobile robot, the scheduling system can take the node as the center and expand outward a preset number of levels to extract a set of surrounding nodes.

[0054] After determining the node range, the scheduling system can organize the relevant information of all nodes within the preset range into a matrix format. This node information can include primitive node information (such as obstacles, the target node of the first mobile robot, other mobile robots, and target system nodes of other mobile robots), as well as optimal path information for the node (such as the optimal path for the first mobile robot and the optimal paths for other mobile robots). The rows and columns of the matrix represent the relative coordinate positions of the nodes in the discretized map, and each matrix element stores the attribute value of the corresponding node.

[0055] This matrix processing allows the scheduling system to structure the local environment around the first mobile robot. This matrix data can be directly used as input for subsequent scheduling decisions, such as conflict detection (checking whether other robots occupy nodes in the matrix) or path planning (calculating the optimal movement direction based on the matrix).

[0056] In some embodiments, determining a scheduling action for the first mobile robot for a next target location based on surrounding environment information, and instructing the first mobile robot to perform the scheduling action, includes: Inputting surrounding environment information into the trained decision model to determine the next target position of the first mobile robot and the scheduling action of the first mobile robot for the next target position; Determine whether the scheduling action is legal and update illegal scheduling actions to legal ones; legal scheduling actions are those that do not cause collisions among the autonomous mobile robot clusters; Based on the identifier of the first mobile robot, the first mobile robot is instructed to perform a legal scheduling action, and the state of the node corresponding to the next target position in the discretized map is updated to be occupied by the first mobile robot.

[0057] Specifically, in an embodiment of the present invention, a hybrid learning method of reinforcement learning and imitation learning can be used to train a decision model, so that the surrounding environment information of the mobile robot can be input into the decision model to obtain the next scheduling action of the mobile robot.

[0058] The scheduling system first inputs the acquired surrounding environment information (a matrix consisting of node information within a preset range) into a trained decision model. The decision model can output the next target position of the first mobile robot and its corresponding scheduling action based on the current position of the first mobile robot and its surrounding environment information.

[0059] The scheduling system can then verify the legality of the scheduling action output by the decision model. The core criterion for legality is whether the scheduling action will cause a collision between mobile robots within the autonomous mobile robot cluster. If the scheduling action causes a collision between mobile robots within the autonomous mobile robot cluster (for example, the node to which the scheduling action is directed is the same as the node where the second mobile robot is located), the scheduling action is illegal. If the scheduling action does not cause a collision between mobile robots within the autonomous mobile robot cluster, the scheduling action is legal.

[0060] If a scheduling action is judged to be illegal, an alternative action can be generated by pre-set rules (such as selecting a suboptimal path, inserting a wait instruction) or re-calling the decision model to ensure that the final executed scheduling action meets the safety constraints.

[0061] In some embodiments, determining whether a scheduling action is legal and updating an illegal scheduling action to a legal scheduling action includes: Based on the discretized map, determine the state of the node corresponding to the next target location; If the state of the node corresponding to the next target position is an obstacle, a map boundary, or a state occupied by any mobile robot other than the first mobile robot, the scheduling action is judged to be illegal; if the state of the node corresponding to the next target position is vacant, the scheduling action is judged to be legal; If the scheduling action is illegal, the scheduling action is changed to a stop action.

[0062] Specifically, in the process of determining whether a scheduling action is legal and updating an illegal scheduling action to a legal scheduling action, the scheduling system can query the node state corresponding to the next target position of the first mobile robot according to the current discretized map.

[0063] The node states can include three illegal situations: obstacles (impassable areas), map boundaries (outside the dispatch scenario), or occupied by other mobile robots (dynamic conflict risk), and one legal state: vacant (safe to traverse). This query process can be implemented using the indexing mechanism of the four-connected grid map, for example, by directly retrieving the attribute label of the target node based on its coordinates.

[0064] After determining the node status, the scheduling system enters the scheduling action legality judgment link: if the node status corresponding to the next target position is one of the three illegal situations mentioned above, the current scheduling action is judged to be illegal; if the node status corresponding to the next target position is "vacant", it is judged to be legal.

[0065] If the judgment result is illegal (for example, the target node is occupied by other robots), the scheduling system can modify the original scheduling action (such as the move instruction) to a stop action to ensure that the robot suspends the current path planning result to avoid conflicts.

[0066] After obtaining a valid dispatch action, the dispatch system can issue the valid dispatch action instruction to the corresponding first mobile robot based on its identifier. It also updates the target node's status in the discretized map to "occupied by the first mobile robot," preventing other mobile robots from repeatedly planning to that node. This mechanism, through a simple stopping strategy linked to the map status, provides a fundamental safeguard for dynamic conflict avoidance in distributed scheduling.

[0067] The distributed scheduling method for a large-scale autonomous mobile robot cluster provided by the present invention is further explained below through embodiments in specific application scenarios.

[0068] This embodiment provides a large-scale autonomous mobile robot cluster distributed scheduling method and simulation system to achieve real-time dynamic scheduling of large-scale autonomous mobile robot clusters. Figure 2FIG. 1 is a flow chart of an embodiment of a large-scale autonomous mobile robot cluster distributed scheduling method provided by the present invention. Figure 2 As shown, the method includes: Step 110 , obtaining and parsing the mobile robot's message, wherein the message is network communication information that includes the mobile robot number and message type and meets specific format requirements, and the message type includes two types: “arrival location” and “decision request”.

[0069] Here, the mobile robot's message is transmitted to the application via network communication. This message can be transmitted using any network communication protocol, but it must contain two pieces of information: the mobile robot number and the message type. The mobile robot number is set before scheduling begins and is unique.

[0070] Step 120: Based on the message type of the message, if the message type is "arrival location", update the discretized map and the node where the mobile robot is located. The discretized map is a four-connected grid graph obtained by discretizing the actual scheduling scenario, which includes several nodes and edges.

[0071] Here, the discretized map is a map stored inside the application. This map is an abstract expression of the actual scheduling scenario and will be updated in real time as the scheduling progresses, providing necessary information support for subsequent decision-making. Figure 3 is a schematic diagram of the discretized map of the mobile robot provided by the present invention, such as Figure 3 As shown, each square represents a node. The map includes boundaries and several obstacles. There are 5 mobile robots in total, each of which occupies a node and has its own goal.

[0072] Once the mobile robot ID and the message type are determined to be "arrived at a location," the mobile robot is considered to have arrived at a new node. The discretized map will then change the mobile robot's location from its previous location to its new location. Furthermore, the application maintains an optimal path for each mobile robot, which must be updated as the robot's location changes.

[0073] Step 130 : Based on the message type of the message, if the message type is “decision request”, determine the surrounding environment information of the mobile robot, where the surrounding environment consists of all nodes within a certain range centered on the mobile robot.

[0074] Here, the surrounding environment is centered on the mobile robot and consists of all nodes within a certain range, e.g. Figure 4 is a schematic diagram of the visual range and surrounding environment of the mobile robot provided by the present invention, such as Figure 4As shown in the figure, the dotted line range is the field of view of the mobile robot, and all the nodes within the dotted line constitute the surrounding environment of the mobile robot. It should be noted that if the mobile robot is close to the boundary, such as Figure 4 The mobile robot 5 in FIG has its field of view beyond the map boundary. In this case, the part beyond the boundary will be filled with obstacles.

[0075] After the application determines the mobile robot number and the message type is a "decision request", it will extract the original information from the discretized map according to the current position of the mobile robot and the set field of view. The original information will be stored in the form of a matrix.

[0076] Step 140: Determine the action of the mobile robot based on the surrounding environment information of the mobile robot, where the action includes stopping and moving to an adjacent node.

[0077] Here, the action of the mobile robot is an instruction to direct the mobile robot to move to a specific adjacent node position next. In this embodiment, the actions of the mobile robot are set to five categories, namely, stopping and moving toward four adjacent nodes.

[0078] All the information collected in step 130 is fed into a decision-making model, a neural network model whose input is information about the mobile robot's surroundings and whose output is an action label. This decision-making model determines the specific action of the mobile robot, and after the legitimacy check is completed, the action is transmitted to the mobile robot via network communication for execution.

[0079] It should be noted that the decision neural network is pre-trained through a hybrid learning method of reinforcement learning and imitation learning. Figure 5 This is a schematic diagram of the decision-making neural network model provided by the present invention. The network model's parameters already provide excellent scheduling decision-making capabilities. Specific parameter information is loaded via a .pth file. After the application is initialized, a neural network model is generated based on the set network structure and parameter information. This model is directly called upon to determine the action during decision-making.

[0080] It is understandable that due to the uninterpretability of neural networks and the inability of agents to communicate directly with each other during distributed scheduling, actions decided by neural networks cannot fully guarantee safety, that is, they cannot completely prevent collisions between robots and obstacles or between robots. Therefore, the following steps can be performed to determine the legality of decision-making actions and handle illegal actions. The specific steps are as follows: Based on the action, it is determined whether the action is legal, and illegal actions are processed. It is characterized in that based on the action, it is determined whether the action guarantees safety, and unsafe actions are processed.

[0081] The legality of an action refers to whether the mobile robot, while executing the action, will experience collisions with obstacles or other robots, which could potentially cause personal safety or property damage. If the action is foreseeable and unlikely to cause these situations, the action is considered legal. If these situations are likely to occur, the action is considered illegal.

[0082] First, the robot determines the next node it needs to reach based on the action. The state of that node is then determined based on the discretized map information. Whether the node is occupied determines whether the action is legal. Legal actions are not processed, while illegal actions are stopped to avoid potential collisions.

[0083] Step 150, based on the above method, establish a simulation system for the motion and control of a large-scale autonomous mobile robot cluster, implement the decision-making system established in the above steps, apply the decision-making system to the autonomous mobile robot cluster, and realize the dynamic scheduling of the large-scale autonomous mobile robot cluster.

[0084] Here, based on the simulation system architecture and the determined legal robot actions, after receiving the next legal action from the server, the robot end in the simulation environment will first complete the current target action and then update the received next legal action to the current target action, and continuously receive the next legal action and execute and update the current target action until the target end point is reached.

[0085] Here, the communication between the robot side and the server side in the simulation environment is based on the MQTT communication protocol; the receiving of the next legal action and the execution and updating of the current target action are to avoid the waiting time for the next legal action from the server side after the robot completes the current action and sends a decision request to the server side.

[0086] This system uses EMQX as the client for the MQTT communication protocol; other MQTT clients can also be embedded in this system. After the server and robot are connected to the MQTT broker server, the server subscribes to the "release node" and "decision request" topics published by all robots using wildcards. Upon receiving a "release node" message, the server releases the previous target node occupied by the corresponding robot. Upon receiving a "decision request" message, the server publishes the next target node to the robot. The robot receives the target node issued by the server by subscribing to the server's "control decision" topic.

[0087] It should be noted that the robot will only send a "release node" message to the server when it reaches the current target node; when the robot reaches an appropriate position close to the current target node, such as the midpoint between the previous node and the current target end point, it will send a "decision request" message to the server. The appropriate position is determined by factors such as the robot's size, speed, and the distance between nodes.

[0088] The distributed scheduling method for large-scale autonomous mobile robot clusters provided in this embodiment obtains and parses the robot's messages, classifies the messages, updates the discretized map for the message type "arrival location", determines the environment around the mobile robot for the message type "decision request", decides the mobile robot's actions, and realizes the practical application of the scheduling algorithm based on the simulation platform.

[0089] Based on any of the above embodiments, if the message type is "arrival location", updating the discretized map and the node where the mobile robot is located includes: Based on the mobile robot number and current information of the discretized map, updating the discretized map to a new state; Based on the node where the mobile robot is currently located, the optimal path information of the mobile robot is determined; the optimal path is the shortest path for the mobile robot to reach the target node without considering obstructions from other robots.

[0090] Here, the discretized map is a map stored inside the application. This map is an abstract expression of the actual scheduling scenario and will be updated in real time as the scheduling progresses, providing necessary information support for subsequent decision-making.

[0091] Based on the mobile robot's new position, the discretized map can be directly read and written. Since the discretized map stores real-time map information via a matrix, the original position of the mobile robot is modified to 0, and the value of the new position is modified to the mobile robot's number. In addition, the application determines an "optimal path" for each mobile robot. This optimal path is the shortest path for the mobile robot to reach its target location without considering obstructions from other mobile robots. As the mobile robot's position changes, the "optimal path" also changes. Therefore, after the discretized map is updated, the A* algorithm is used to recalculate the mobile robot's "optimal path."

[0092] It should be noted that the "optimal path" serves as crucial information for decision-making. A mobile robot considers not only its own optimal path but also the optimal paths of other nearby robots, thereby achieving a balance between self-interest and altruism in its decision-making and improving overall system efficiency. While calculating the optimal path is time-consuming, updating the map itself requires less computation and requires less real-time performance. Therefore, calculating the optimal path is integrated into the map update process. Furthermore, the optimal path information is stored in a matrix that matches the shape of the discretized map. Points not on the path are marked as 0, while points on the path are marked with their distance to the target.

[0093] Based on any of the foregoing embodiments, if the message type is "decision request," determining the surrounding environment information of the mobile robot includes: Determining nodes constituting the surrounding environment based on the mobile robot number and the discretized map information; Based on the nodes constituting the surrounding environment, the obstacle matrix, other mobile robot target matrices, optimal path matrix, etc. are calculated. The various matrices are matrices that store various types of information, and these matrices constitute the surrounding environment information.

[0094] Here, the mobile robot's surroundings are determined by the mobile robot's own position and the set field of view. The surrounding environment information can be divided into obstacles, the mobile robot's own target, other mobile robots, the targets of other mobile robots, the mobile robot's own optimal path, and the optimal paths of other mobile robots.

[0095] For example, Figure 6 is a schematic diagram of the surrounding environment of the mobile robot provided by the present invention, such as Figure 6 As shown, for mobile robot 1 and mobile robot 5, the nodes constituting the surrounding environment of the mobile robot and all surrounding environment information are as follows: Figure 6 Shown on the right.

[0096] Based on the mobile robot's current position and the set field of view, raw information is extracted from the discretized map. This raw information can then be used to determine four types of information: obstacles, the mobile robot's own target, other mobile robots, and the targets of other mobile robots. Based on the mobile robot's current position and the set field of view, the optimal path for the mobile robot itself and the optimal paths for other mobile robots are determined. Based on the stored target position and current position of the mobile robot, the relative distance in the x-direction, the relative distance in the y-direction, and the absolute distance are determined.

[0097] It should be noted that in addition to the aforementioned surroundings, the decision-making process also requires consideration of key information: the location of the mobile robot's own target. This information is broken down into three categories: relative distance in the x-direction, relative distance in the y-direction, and absolute distance. This information can be directly obtained and is independent of the mobile robot's field of view.

[0098] Based on any of the foregoing embodiments, determining the action of the mobile robot based on the surrounding environment information of the mobile robot includes: Based on the surrounding environment information, determining the action of the mobile robot through a decision model, wherein the decision model is a trained neural network model; Based on the action, determine whether the action is legal and process illegal actions, wherein the legal action is an action that can ensure the safety of the mobile robot in the system and prevent collision; Based on the mobile robot number and the legal action, the action is issued to the mobile robot, and the state of the node corresponding to the action in the discretized map is changed to be occupied by the mobile robot.

[0099] The collected information about the mobile robot's surroundings is converted into a tensor form suitable for neural network input. This information is then fed into the decision-making neural network model, which determines the optimal action based on the input information. The model then determines the legality of the action, changing any illegal action to a stop. The application then publishes the action information to the corresponding mobile robot via a network communication protocol, handing it over to the mobile robot for execution. The location the mobile robot has decided to reach is marked on the discretized map to prevent other mobile robots from deciding to proceed there, potentially creating safety issues.

[0100] It should be noted that in order to ensure real-time performance, all "arrival location" and "decision request" messages are processed in parallel. Since the decision-making requires a certain amount of time to calculate the neural network model, this will cause the map to continue to change during the calculation process, resulting in the mobile robot's surrounding environment when the neural network is input and the surrounding environment when the neural network is calculated and output are different, resulting in deviations in the results.

[0101] For example, Figure 7 This is a schematic diagram of illegal actions provided by the present invention, such as Figure 7As shown in the figure, the first case is an obstacle conflict, where the target of the decision is an obstacle; the second case is a robot conflict, where the target of the decision is already occupied by another mobile robot; and the third case is a target conflict, where mobile robot 1 and mobile robot 2 start making decisions at the same time. Assume that mobile robot 1 completes its calculation and occupies the target position in advance. In this case, if the target position of mobile robot 2 after calculation is the same as the target position of mobile robot 1, a target conflict is considered to have occurred. All three cases result in illegal actions.

[0102] At time 1, for mobile robot 1, all surrounding locations can be reached, but because decision-making requires a certain amount of time, when the calculation is completed at time 2, node A on the right side of mobile robot 1 is occupied in advance by mobile robot 2, that is, the decision result of mobile robot 2 is to reach node A. Since the collected information about the surrounding environment of the mobile robot is at time 1, the decision of mobile robot 1 may still be to reach node A, resulting in a decision conflict and an illegal action.

[0103] Based on any of the above embodiments, determining whether the action is legal based on the action and processing illegal actions includes: Based on the action, determining a node position to which the mobile robot is about to go corresponding to the action, and determining a state of the node based on the discretized map information; Based on the node status, if the node is in one of the following three states: an obstacle, a map boundary, or occupied by another agent, the action is considered illegal; if the node is vacant, the action is considered legal; Based on whether the action is legal, if the action is illegal, the action is changed to stop; if the action is legal, the original action is maintained; First, the next node the mobile robot needs to reach is determined based on the action, and the node's state is determined based on the discretized map information. Based on the node's state, if the node has any of the following conditions: an obstacle, a map boundary, or is occupied by another mobile robot, the action will result in a collision, and the action is marked as illegal. If the node is vacant, the action will not result in a collision, and the action is marked as legal. Legal actions are not processed, while illegal actions are stopped to avoid potential collisions.

[0104] Based on any of the above embodiments, this embodiment provides a simulation system. This simulation system is applicable to all existing robot simulation environments and any model of mobile robot. It can be deployed in all existing robot simulation environments, such as Coppeliasim, Pybullet, Gym, and Gazebo. Simply by deploying the discretized four-connected map and mobile robots in the simulation environment, this system can be used to achieve the motion and control of large-scale autonomous mobile robot swarms.

[0105] This simulation system consists of a server-based scheduling algorithm and a robot simulation environment. The server-based scheduling algorithm is responsible for making decisions about robot scheduling: determining the next target node for the mobile robot based on information about its surroundings. The robot simulation environment receives these decisions from the server and controls the robot's movement toward the target node. Communication between the robots in the simulation environment and the server-based scheduling algorithm is accomplished via MQTT. Each robot is directly scheduled by the server-based scheduling algorithm, without interfering with other robots.

[0106] Based on the MQTT communication method described above, the server's scheduling algorithm and the robot simulation environment need to be connected to the MQTT BROKER server to achieve communication; the server needs to subscribe to the "release node" and "decision request" published by each robot through a wildcard method to determine when to release the node occupied by the robot and when to publish a new target node to the corresponding robot; each robot needs to subscribe to the decision topic published by the server according to its own robot number to obtain the server's decision information.

[0107] Based on the MQTT communication method described above, after the server and robot are connected to the MQTT broker, the server subscribes to the "decision request" and "release node" topics published by all robots using wildcards. When receiving a "release node" message, the server releases the previous target node occupied by the corresponding robot. When receiving a "decision request" message, the server publishes the next target node to the robot. The robot receives the target node issued by the server by subscribing to the "control decision" topic on the server.

[0108] When the robot reaches an appropriate position close to the target current target node, the appropriate position is determined by factors such as the size, speed and distance between nodes of the robot. The robot obtains the next target node in advance by sending a "decision request" to the server. When the robot reaches the current target node, it goes directly to the next target node to avoid slowing down and stopping to wait for the next target node after reaching the target node. When the robot reaches the current target node, it sends a "release node" message to the server to notify the server to release the previous target node occupied by the corresponding robot, and at the same time update the current target node to the received next target node.

[0109] When the scheduling algorithm on the server side receives a "decision request" from the robot side, it gives the robot's next target node through the trained neural network based on information such as the robot's current coordinate position, target end point, and the coordinate positions of surrounding robots, and publishes it to the corresponding robot with the "control decision" topic.

[0110] As the robot moves toward its target location (end point), it undergoes the following communication and movement with the server scheduling algorithm: (1) When the discretized four-connected map and mobile robots are deployed in the initial simulation environment, each robot must send the corresponding "decision request" information according to its own model. After receiving the "decision request" from the robot, the server calculates the next target node of each robot based on the target position of each robot and the location information of each robot, and sends it to each robot by publishing a decision topic.

[0111] (2) After receiving the current target node, the robot starts to move towards the current target node. When it approaches the current target node (the distance is adjusted according to the distance between nodes in the discretized map, the size and forward speed of the mobile robot, and the decision time of the scheduling algorithm), the robot sends a decision request message to the server again to obtain the location of the next target node. There are two situations for the next target node: ① If the next target node is inconsistent with the current target node, then when the robot reaches the current target node, it will update the next target node to the current target node, and at the same time send the release of the previous node information to the server to update the discretized map, and move towards the new current target node; ② Since the next target node is occupied by other robots, the server scheduling algorithm releases the node coordinates consistent with the current target node to the robot. Then, when the robot reaches the current target node and sends the release information of the previous node to the server, it will stay at the current target node for 1 second, and then send a decision request to the server again. If the coordinates given by the server are inconsistent with the current target node, the robot updates the current target node and moves towards the new target node. If they are still consistent, the robot will stay for 1 second again and then send a decision request again.

[0112] (3) When the robot reaches the target position (end point) according to the server scheduling information, it needs to send a decision request to the server; since the distributed scheduling method of this large-scale autonomous mobile robot cluster is an uninterrupted scheduling method, when the robot reaches the target position, the server scheduling algorithm will automatically update the target position of the corresponding robot, and then give the robot a new target node based on the robot's current position, the new target position and other map information. The robot then moves to the new target node or stays at the current node (the new target node is consistent with the current position, similar to situation (2)), and the cycle repeats.

[0113] The distributed scheduling device for a large-scale autonomous mobile robot cluster provided by the present invention is described below. The distributed scheduling device for a large-scale autonomous mobile robot cluster described below and the distributed scheduling method for a large-scale autonomous mobile robot cluster described above can be referenced to each other.

[0114] Figure 8 This is a schematic diagram of the structure of the distributed scheduling device for a large-scale autonomous mobile robot cluster provided by the present invention. Figure 8 As shown, the device includes the following modules: The receiving module 800 is configured to receive a scheduling-related message sent by the first mobile robot; the scheduling-related message includes an identifier of the first mobile robot and is configured to instruct the first mobile robot to request a scheduling decision and / or to reach a target location; A scheduling module 810 is configured to schedule the first mobile robot based on the scheduling-related message; The first mobile robot is any mobile robot in the autonomous mobile robot cluster.

[0115] In some embodiments, scheduling the first mobile robot based on the scheduling-related message includes: In a case where the scheduling-related message is used to instruct the first mobile robot to request a scheduling decision, determining surrounding environment information of the first mobile robot based on the scheduling-related message; the surrounding environment information of the first mobile robot includes information of all nodes within a preset range centered on the first mobile robot; Determining a scheduling action for the first mobile robot for a next target location based on surrounding environment information, and instructing the first mobile robot to perform the scheduling action; In a case where the scheduling-related message is used to instruct the first mobile robot to arrive at a target location, updating the discretized map and the shortest path information of the first mobile robot to reach the destination based on the scheduling-related message; Among them, the discretized map is a four-connected grid graph obtained by discretizing the actual autonomous mobile robot cluster scheduling scenario, which consists of multiple nodes and multiple edges.

[0116] In some embodiments, updating the discretized map and the optimal path information for the first mobile robot to reach the destination based on the scheduling-related message includes: updating the discretized map based on the identifier of the first mobile robot and the current discretized map; Based on the node where the first mobile robot is currently located, the optimal path information for the first mobile robot to reach the destination is updated.

[0117] In some embodiments, determining the surrounding environment information of the first mobile robot based on the scheduling-related message includes: Determine all nodes within a preset range centered on the first mobile robot based on the identifier of the first mobile robot and the current discretized map; The surrounding environment information of the first mobile robot is determined based on a matrix obtained based on the relevant information of all nodes within a preset range.

[0118] In some embodiments, determining a scheduling action for the first mobile robot for a next target location based on surrounding environment information, and instructing the first mobile robot to perform the scheduling action, includes: Inputting surrounding environment information into the trained decision model to determine the next target position of the first mobile robot and the scheduling action of the first mobile robot for the next target position; Determine whether the scheduling action is legal and update illegal scheduling actions to legal ones; legal scheduling actions are those that do not cause collisions among the autonomous mobile robot clusters; Based on the identifier of the first mobile robot, the first mobile robot is instructed to perform a legal scheduling action, and the state of the node corresponding to the next target position in the discretized map is updated to be occupied by the first mobile robot.

[0119] In some embodiments, determining whether a scheduling action is legal and updating an illegal scheduling action to a legal scheduling action includes: Based on the discretized map, determine the state of the node corresponding to the next target location; If the state of the node corresponding to the next target position is an obstacle, a map boundary, or a state occupied by any mobile robot other than the first mobile robot, the scheduling action is judged to be illegal; if the state of the node corresponding to the next target position is vacant, the scheduling action is judged to be legal; If the scheduling action is illegal, the scheduling action is changed to a stop action.

[0120] Figure 9 This is a schematic diagram of the structure of the distributed scheduling system for large-scale autonomous mobile robot clusters provided by the present invention. Figure 9 As shown, the system includes: Autonomous mobile robot cluster 900, and distributed scheduling device 910 for large-scale autonomous mobile robot cluster as described above; Among them, any mobile robot 901 in the autonomous mobile robot cluster is used to send scheduling-related messages to the distributed scheduling device 910 and accept the scheduling of the distributed scheduling device 910.

[0121] Figure 10 Schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 10 As shown, the electronic device may include: a processor 1010, a communications interface 1020, a memory 1030, and a communication bus 1040, wherein the processor 1010, the communications interface 1020, and the memory 1030 communicate with each other via the communication bus 1040. The processor 1010 may call the logic instructions in the memory 1030 to execute a distributed scheduling method for a large-scale autonomous mobile robot cluster, which includes: receiving a scheduling-related message sent by the first mobile robot; the scheduling-related message includes an identifier of the first mobile robot and is used to instruct the first mobile robot to request a scheduling decision and / or arrive at a target location; Scheduling the first mobile robot based on the scheduling-related message; The first mobile robot is any mobile robot in the autonomous mobile robot cluster.

[0122] Furthermore, the logic instructions in the aforementioned memory 1030 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0123] On the other hand, the present invention further provides a computer program product, comprising a computer program, which may be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the distributed scheduling method for a large-scale autonomous mobile robot cluster provided by the above methods, the method comprising: receiving a scheduling-related message sent by the first mobile robot; the scheduling-related message includes an identifier of the first mobile robot and is used to instruct the first mobile robot to request a scheduling decision and / or arrive at a target location; Scheduling the first mobile robot based on the scheduling-related message; The first mobile robot is any mobile robot in the autonomous mobile robot cluster.

[0124] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the distributed scheduling method for a large-scale autonomous mobile robot cluster provided by the above methods, the method comprising: receiving a scheduling-related message sent by the first mobile robot; the scheduling-related message includes an identifier of the first mobile robot and is used to instruct the first mobile robot to request a scheduling decision and / or arrive at a target location; Scheduling the first mobile robot based on the scheduling-related message; The first mobile robot is any mobile robot in the autonomous mobile robot cluster.

[0125] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0126] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A distributed scheduling method for a large-scale autonomous mobile robot cluster, characterized in that: include: receiving a scheduling-related message sent by the first mobile robot; The scheduling-related message includes an identifier of the first mobile robot and is used to instruct the first mobile robot to request a scheduling decision and / or arrive at a target location; Scheduling the first mobile robot based on the scheduling-related message; The first mobile robot is any mobile robot in an autonomous mobile robot cluster.

2. The distributed scheduling method for large-scale autonomous mobile robot clusters according to claim 1 is characterized in that: Scheduling the first mobile robot based on the scheduling-related message includes: In a case where the scheduling-related message is used to instruct the first mobile robot to request a scheduling decision, determining surrounding environment information of the first mobile robot based on the scheduling-related message; the surrounding environment information of the first mobile robot includes information of all nodes within a preset range centered on the first mobile robot; determining a scheduling action for the first mobile robot to a next target location based on the surrounding environment information, and instructing the first mobile robot to execute the scheduling action; In a case where the scheduling-related message is used to instruct the first mobile robot to arrive at a target location, updating a discretized map and information about a shortest path for the first mobile robot to reach a destination based on the scheduling-related message; The discretized map is a four-connected grid graph obtained by discretizing the actual autonomous mobile robot cluster scheduling scenario, and is composed of multiple nodes and multiple edges.

3. The distributed scheduling method for large-scale autonomous mobile robot clusters according to claim 2, characterized in that: Based on the scheduling-related message, updating the discretized map and the optimal path information for the first mobile robot to reach the destination includes: updating the discretized map based on the identifier of the first mobile robot and the current discretized map; Based on the node where the first mobile robot is currently located, the optimal path information for the first mobile robot to reach the destination is updated.

4. The distributed scheduling method for large-scale autonomous mobile robot clusters according to claim 2, characterized in that: Determining surrounding environment information of the first mobile robot based on the scheduling-related message includes: Determining all nodes within a preset range centered on the first mobile robot based on the identifier of the first mobile robot and the current discretized map; Based on a matrix obtained from relevant information of all nodes within the preset range, the surrounding environment information of the first mobile robot is determined.

5. The distributed scheduling method for a large-scale autonomous mobile robot cluster according to claim 2 or 4, characterized in that: Determining a scheduling action for the first mobile robot to a next target location based on the surrounding environment information, and instructing the first mobile robot to perform the scheduling action, including: Inputting the surrounding environment information into a trained decision model to determine the next target position of the first mobile robot and a scheduling action of the first mobile robot for the next target position; Determining whether the scheduling action is legal, and updating an illegal scheduling action to a legal scheduling action; the legal scheduling action is a scheduling action that does not cause a collision between the autonomous mobile robot cluster; Based on the identifier of the first mobile robot, the first mobile robot is instructed to perform a legal scheduling action, and the state of the node corresponding to the next target position in the discretized map is updated to be occupied by the first mobile robot.

6. The distributed scheduling method for large-scale autonomous mobile robot clusters according to claim 5, characterized in that: Determining whether the scheduling action is legal and updating illegal scheduling actions to legal scheduling actions includes: Determining the state of the node corresponding to the next target position based on the discretized map; If the node corresponding to the next target position is in a state of an obstacle, a map boundary, or occupied by any mobile robot other than the first mobile robot, the scheduling action is determined to be illegal; if the node corresponding to the next target position is in a state of vacant, the scheduling action is determined to be legal; In the case that the scheduling action is illegal, the scheduling action is modified to a stop action.

7. A distributed scheduling device for a large-scale autonomous mobile robot cluster, characterized in that: include: A receiving module, configured to receive a scheduling-related message sent by the first mobile robot; The scheduling-related message includes an identifier of the first mobile robot and is used to instruct the first mobile robot to request a scheduling decision and / or arrive at a target location; a scheduling module, configured to schedule the first mobile robot based on the scheduling-related message; The first mobile robot is any mobile robot in an autonomous mobile robot cluster.

8. A distributed scheduling system for large-scale autonomous mobile robot clusters, characterized in that: include: An autonomous mobile robot cluster, and a distributed scheduling device for a large-scale autonomous mobile robot cluster as claimed in claim 7; Wherein, any mobile robot in the autonomous mobile robot cluster is used to send scheduling-related messages to the distributed scheduling device and accept scheduling by the distributed scheduling device.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the distributed scheduling method for a large-scale autonomous mobile robot cluster as claimed in any one of claims 1 to 6 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the distributed scheduling method for a large-scale autonomous mobile robot cluster as claimed in any one of claims 1 to 6 is implemented.

Images