A multi-robot scheduling control method based on roundabout movement and deadlock prevention

Abstract

The application relates to a kind of multi-robot scheduling control methods based on roundabout movement and deadlock prevention, comprising: constructing annular buffer and dynamic circulation path between picking station and storage area, and modeling finite state machine for dynamic behavior of robot, to obtain the finite state machine model corresponding to different robots;Based on the finite state machine model of robot, combined with resource allocation constraints and preset operation constraints, the multi-robot scheduling strategy is output by executing roundabout movement scheduling algorithm, to control the working state of each robot accordingly.Compared with the prior art, the application can dynamically guide the robot to move in order along the circulation path between the picking station and the storage area, and can actively prevent deadlock, effectively solving the limitations of traditional scheduling strategy in deadlock prevention, resource utilization and efficiency improvement.

Description

Technical Field

[0001] This invention relates to the field of intelligent agent scheduling and control technology, and in particular to a multi-robot scheduling and control method based on island motion and deadlock prevention. Background Technology

[0002] With the rapid development of e-commerce and intelligent manufacturing, the demand for automation and responsiveness in warehousing and logistics systems is increasing. Robotic Mobile Fulfillment Systems (RMFS) have become the mainstream solution for large and medium-sized warehousing scenarios due to their high efficiency. In RMFS, robots need to complete the entire process of picking up goods from the storage area, transporting them to the picking station, and returning them to the storage area according to order requirements. However, as the core area for multi-robot interaction, congestion and deadlock around the picking station have always been key bottlenecks restricting system efficiency.

[0003] In real-world scenarios, when a picking station needs to process orders containing multiple SKUs (Stock Keeping Units), a large number of robots carrying the corresponding SKUs will gather around it. If there is a lack of proper distribution and path management, congestion or even deadlock can easily occur. This type of deadlock arises from four necessary conditions for its occurrence: 1. Mutual exclusion, meaning that a certain resource can only be used by one robot at a time. For example, picking stations, aisles, shelves, and waypoints are all exclusive resources. Once one robot occupies it, other robots cannot use it. 2. Occupying and waiting: This means that the robot occupies some resources while waiting for other resources. For example, the robot carries a shelf (resource 1) and waits to be picked at a location near the picking station (resource 2). 3. No preemption: Once a robot occupies a picking station or other path location, it cannot be forcibly taken back. 4. Cyclic waiting, which means that multiple robots form a cyclic waiting, such as multiple robots forming a closed-loop blocking chain in the intersection area of ​​the path.

[0004] like Figure 1 As shown (S1, A1, C1, and N1 in the diagram represent different tasks), some robots (the robots carrying S1) may occupy the exit channel, preventing the robots that have completed picking (the robots carrying A1) from leaving the exit channel. When the exit channel is full of robots, the robots that have completed picking (the robots carrying C1) will not be able to enter the channel, and the robots waiting outside (the robots carrying N1) will also not be able to enter the picking station to pick, thus creating a deadlock situation where "external robots cannot enter and internal robots cannot leave".

[0005] To address the aforementioned issues, existing technologies explore solutions from multiple dimensions: In terms of path optimization, multi-agent path planning algorithms generate conflict-free paths, and by combining priority mechanisms and conflict detection, they reduce access conflicts for robots at key nodes, ensuring the efficiency and safety of the path. In terms of local traffic control, the right-of-way for robots is temporarily regulated by setting up virtual traffic lights and dynamically adjusting priorities, thereby alleviating lane conflicts. Regarding the waiting area and queuing mechanism, a buffer zone is set up near the picking station, and robots are released in an orderly manner according to certain rules to avoid disorderly occupation of the passage. Regarding the task batching and dynamic allocation mechanism, order tasks are processed in staggered time, and task allocation is adjusted according to the real-time load of the picking station to balance the pressure of each node. In addition, intelligent scheduling methods based on multi-agent reinforcement learning and neural networks are used to dynamically optimize paths and task priorities by sensing the system status in real time, thereby further improving deadlock prevention capabilities.

[0006] The technologies mentioned above improve the efficiency and safety of multi-robot systems from different perspectives, laying the foundation for the development of warehouse automation. Among them, the buffer method, as a common solution to alleviate congestion at picking stations in multi-robot warehouse systems, achieves orderly queuing of robots by defining dedicated waiting areas, thus reducing local chaos caused by disorderly competition to a certain extent. However, in scenarios where robot density continues to increase and order complexity significantly improves, its inherent limitations gradually become apparent, making it difficult to meet the requirements for efficient and stable system operation, as detailed below: 1. The static nature of space configuration leads to insufficient resource adaptability: The location and capacity of traditional buffer zones are usually fixed and cannot be dynamically adjusted according to the number of robots and order types in real time. During peak order periods, buffer zones are prone to congestion again due to capacity saturation; while during off-peak periods, the idleness of fixed space will result in resource waste and will not achieve efficient utilization of warehouse space.

[0007] 2. The static state of robots reduces system response efficiency: Robots entering the buffer zone are mostly in a static waiting state, lacking a dynamic flow mechanism. This not only prolongs the ineffective waiting time of robots, but also leads to low resource turnover efficiency at the picking station—when the picking station becomes idle, robots that are far away in the buffer zone cannot respond quickly, resulting in resource mismatch and affecting the order processing rhythm.

[0008] 3. Limitations of collaborative scheduling constrain global optimization: In complex scenarios such as multi-picking station collaboration and mixed operations of multiple types of robots, traditional buffers are difficult to achieve global collaborative scheduling. Robots only queue within a single buffer and cannot be dynamically distributed according to the idle status of other picking stations, which can easily lead to local congestion and imbalance in global resource utilization, reducing the overall throughput of the system.

[0009] 4. Incomplete deadlock prevention poses potential risks: Although traditional buffers can reduce direct channel blockage, they cannot fundamentally eliminate the "circular waiting" condition. If the transition channel between the buffer and the picking station is poorly designed, stationary robots may still block the critical path, leading to a deadlock risk where "external robots cannot enter and internal robots cannot leave," affecting system stability.

[0010] In summary, traditional buffer methods suffer from a sharp drop in efficiency when faced with high-density robot clusters and highly complex orders due to rigid scheduling, or they cannot completely avoid deadlocks due to insufficient adaptability, making it difficult to meet the needs of modern intelligent warehousing systems for efficient and stable operation. Summary of the Invention

[0011] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a multi-robot scheduling and control method based on island motion and deadlock prevention, which can effectively solve the limitations of traditional scheduling strategies in deadlock prevention, resource utilization and efficiency improvement.

[0012] The objective of this invention can be achieved through the following technical solution: a multi-robot scheduling and control method based on island motion and deadlock prevention, comprising the following steps: S1. Construct a circular buffer and dynamic loop path between the picking station and the storage area, and perform finite state machine modeling for the dynamic behavior of the robot to obtain the finite state machine model corresponding to different robots. S2. Based on the finite state machine model of the robot, combined with resource allocation constraints and preset operation constraints, the multi-robot scheduling strategy is obtained by executing the island motion scheduling algorithm to control the working state of each robot accordingly.

[0013] Furthermore, the process of modeling the robot's dynamic behavior using a finite state machine in step S1 includes: abstracting the robot's dynamic process into a series of discrete states, which correspond to different physical locations and task stages of the robot in the warehouse. Define binary discrete event variables and key events, and construct the robot's state transition function. The binary discrete event variables are used to represent whether a key event occurs at discrete time t, and the state transition function is used to describe the transition relationship between the current state and the next state of the machine based on the key event.

[0014] Furthermore, the discrete state specifically includes the following states: The system is either stationary in the storage area or waiting for task assignment. Head to the shelf; Pick up your order from the shelf; Proceed to an available picking station; To navigate around the island or return to the island's original path; Working at the sorting station; Return to storage area.

[0015] Furthermore, the binary discrete event variable specifically refers to: The key events include: , represented as robot r i Assign new picking tasks; , indicating robot r i Arrive at the designated shelf; This indicates that the target picking station is idle and available for robot operation; , indicating that it is assigned to robot r i The picking task has been fully completed; , indicating robot r i Return to storage area.

[0016] Furthermore, the state transition function of the robot is specifically as follows: in, This is the robot's next state. This represents the robot's current state. These are the key events between the robot's current state and the next state. This is the state transition function, which indicates that the robot will transition to the next state after encountering a critical event in its current state.

[0017] Furthermore, the resource allocation constraints in step S2 include the correspondence between orders and workstations, the correspondence between SKUs and shelves, and the correspondence between shelves and robots. These resource allocation constraints remain fixed during the execution of the island motion scheduling algorithm.

[0018] Furthermore, the order-workstation correspondence specifically refers to the fact that before operation, each order is uniquely assigned to a specific picking workstation; The SKU-shelf correspondence specifically refers to the fact that each SKU is pre-loaded on a unique shelf pod; The shelf-robot correspondence specifically refers to the fact that during the initialization phase, each shelf is exclusively assigned to a single robot.

[0019] Furthermore, the preset operating constraints in step S2 are specifically set through the built-in operating rules in the simulation environment, including picking station occupancy constraints, loop node conflict control constraints, and loop empty space forced constraints.

[0020] Furthermore, the picking station occupancy constraint specifically means that within any given discrete time step, no more than one robot is allowed to occupy the same picking station. The loop node conflict control constraint specifically means that each node on the loop path can be occupied by at most one robot at any given time. The mandatory constraint on loop vacancy specifically refers to ensuring that at least one position on the loop path is always vacant.

[0021] Furthermore, the execution process of the island-loop motion scheduling algorithm includes: A1. Input order, picking station, robot, shelf and SKU information, and perform initialization processing, assigning orders to picking stations, assigning SKUs to shelves and setting the initial position of robots; A2. Traverse all robots and control each robot to go to the nearest shelf to retrieve goods; A3. Determine if there are any incomplete orders. If no, end the current process. If the determination is yes, then proceed to step A4; A4. Continue to traverse the robot and determine if there is an empty picking station. If an empty picking station is found, proceed to step A5; otherwise, proceed to step A7. A5. Control the robot to go to an empty picking station to perform picking operations. Then determine whether the SKUs carried by the robot have been picked. If the determination is yes, proceed to step A6; otherwise, return to step A4. A6. Control the robot to return to the storage area to wait, and then return to step A3; A7. Control the robot to enter the roundabout path and perform roundabout movement; A8. Determine if there is an available picking station. If there is an available picking station, proceed to step A9; otherwise, return to step A7. A9. Control the robot to leave the roundabout path, and then proceed to step A5.

[0022] Compared with the prior art, the present invention has the following advantages: This invention constructs a circular buffer zone and a dynamic loop path between the picking station and the storage area, and models the dynamic behavior of the robots using finite state machines to obtain finite state machine models corresponding to different robots. Then, based on the robot's finite state machine model, combined with resource allocation constraints and preset operational constraints, a loop motion scheduling algorithm is executed to output a multi-robot scheduling strategy. This dynamically guides robots to move in an orderly loop along the loop path between the picking station and the storage area, effectively alleviating congestion and deadlock problems caused by excessive robot density near the picking station in multi-robot scenarios, proactively preventing deadlock, and improving overall efficiency.

[0023] This invention employs finite state machine modeling to study the dynamic behavior of robots. First, the robot's dynamic process is abstracted into a series of discrete states, corresponding to different physical locations and task stages within a warehouse. Then, a state transition function is constructed by defining binary discrete event variables and key events to describe the transition relationship between the current state and the next state based on key events. By organically integrating the robot's key events with state transitions, the dynamic evolution process of the robot can be comprehensively and accurately described. This ensures that the scheduling and movement of each subsequent robot can be carried out in an orderly manner based on the corresponding key events, guaranteeing precise control over the robot's sequential state transitions.

[0024] This invention sets resource allocation constraints and preset operating constraints. The resource allocation constraints remain fixed throughout the entire scheduling algorithm execution process, thus eliminating the need for additional explicit optimization constraints. This allows the scheduling algorithm to focus more on robot state transitions, circular path coordination, and deadlock prevention. Furthermore, the preset operating constraints are set through built-in operating rules in the simulation environment, which also eliminates the need for mathematical modeling. This accurately reflects the actual scheduling principles of multiple robots and ensures stability under high-density operating conditions.

[0025] This invention designs a circular island motion scheduling algorithm that dynamically adjusts the robot's entry and exit loop paths based on the idle status of the picking station. This ensures that the robots maintain dynamic flow and can be diverted in real time according to the idle status of the picking station. Relying on the dynamic response capability of the circular island path, when the picking station is idle, the robots on the loop path can quickly switch to the working state, reducing the ineffective waiting caused by static queuing. The robots no longer occupy path nodes and wait for picking station resources, avoiding the state of "holding some resources and waiting for other resources". At the same time, the unidirectional flow of the circular island path cuts off the possibility of multiple robots forming a closed loop blockage, completely solving the deadlock risk of "external robots cannot enter and internal robots cannot leave" in the traditional buffer strategy, thereby greatly alleviating queuing and deadlock problems and improving overall operating efficiency. Attached Figure Description

[0026] Figure 1 This is a diagram illustrating a deadlock scenario; Figure 2 This is a schematic diagram of the method flow of the present invention; Figure 3 This is a schematic diagram of robot state transitions in the embodiment; Figure 4 This is a schematic diagram of the island-loop motion scheduling algorithm. Figure 5 This is a schematic diagram illustrating the scheduling effect of the queuing-free mechanism in the embodiment. Figure 6 This is a schematic diagram illustrating the scheduling effect of the buffer strategy used in the embodiment; Figure 7 This is a schematic diagram illustrating the scheduling effect of the method of the present invention used in an embodiment. Detailed Implementation

[0027] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0028] Example To address the problems of incomplete deadlock prevention, low resource utilization, and long order processing cycles in existing multi-robot warehousing systems, this solution, based on the traditional buffer zone approach, aims to propose a multi-robot scheduling and control method to achieve multi-dimensional and in-depth system optimization, such as... Figure 2 As shown, the method includes the following steps: S1. Construct a circular buffer and dynamic loop path between the picking station and the storage area, and perform finite state machine modeling for the dynamic behavior of the robot to obtain the finite state machine model corresponding to different robots. S2. Based on the finite state machine model of the robot, combined with resource allocation constraints and preset operation constraints, the multi-robot scheduling strategy is obtained by executing the island motion scheduling algorithm to control the working state of each robot accordingly.

[0029] Specifically, this solution constructs a circular buffer zone and dynamic loop path between the picking station and the storage area, guiding robots to move continuously along the loop and avoiding prolonged static waiting. By breaking the "occupy and wait" and "circular waiting" conditions of deadlock (robots move continuously and dynamically along the loop, no longer statically occupying path nodes while waiting for the picking station, avoiding the state of "holding some resources and waiting for other resources"; the unidirectional flow of the loop path eliminates the possibility of multiple robots forming a closed loop blockage, fundamentally cutting off the circular chain of "mutually waiting for each other to release resources"), combined with reasonable control of "mutual exclusion" and "non-preemptive" characteristics (such as ensuring orderly occupation through unique path node constraints), it fundamentally prevents deadlock from occurring—eliminating the closed loop of "circular waiting" through the dynamic flow of path nodes, and avoiding the static resource locking of "occupy and wait" through the continuous movement of robots, ultimately achieving fundamental prevention of deadlock and solving the potential risk of "external robots being unable to enter and internal robots being unable to leave" caused by unreasonable transition channel design in traditional buffer zones.

[0030] Meanwhile, leveraging the dynamic flow of the circular path, robots on the circular path can quickly respond and enter operational status when the picking station becomes idle, reducing the ineffective waiting time caused by static queuing. Combined with precise control of robot state transitions using a finite state machine model (such as real-time switching from "circular movement state" to "heading to picking station state"), the system achieves immediate task response, improving the processing efficiency of the entire order process from "allocation-picking-return," and shortening the single order completion time and the total system order fulfillment time. Furthermore, the circular buffer does not require fixed waiting positions and can dynamically adjust space utilization according to the real-time number of robots, avoiding the waste caused by traditional static configurations and achieving intensive collaboration between warehouse space, picking equipment, and robot capacity.

[0031] To address complex scenarios such as surging robot density, diverse order SKUs, and collaborative operations across multiple picking stations, this solution combines model constraints (e.g., path node uniqueness, picking station exclusivity) with a ring-loop motion scheduling algorithm to ensure stable system operation under high load. Simultaneously, the dynamic mechanism of the ring-loop path flexibly adapts to changes in the number of robots (e.g., newly added robots can be directly integrated into the loop path) and increased order complexity (e.g., collaborative picking of multi-SKU orders), providing technical support for the large-scale expansion and business upgrades of warehousing systems. Furthermore, by defining mathematical variables and constraints for order allocation, shelf selection, and robot state transitions, and combining this with the state transition logic of a finite state machine, a quantifiable and reusable multi-robot scheduling theory system is formed. This system can not only be directly applied to existing warehousing scenarios but also provide a reference for other multi-agent collaborative scheduling fields (e.g., workshop AGVs, intelligent delivery), promoting the standardization and universalization of automated scheduling technology.

[0032] In other words, this solution aims to achieve the comprehensive optimization goals of "no deadlock, high efficiency, high utilization, and strong adaptability" of the multi-robot warehousing system through the innovative design of the island path mechanism, providing a complete theoretical solution and practical path for the efficient and stable operation of intelligent warehousing and logistics.

[0033] To facilitate understanding of this solution, this embodiment first clarifies the various physical parameters involved in the method execution process, as shown in Table 1: Table 1 make Indicates a single order The completion time is defined as the discrete time step in which all SKUs required for the order are successfully picked, and the total order fulfillment time. Defined as the maximum value among all order completion times: =max{ |i=1,2,…} in, and All are discrete variables. The timeframe is reached when all picking stations have completed their assigned order picking tasks, i.e., when T = ... At that point, the system considers all order processing to be complete.

[0034] The specific application process of this embodiment includes: I. Finite State Machine Modeling To accurately characterize the dynamic behavior of robots in a cyclic path-based warehousing system, this solution introduces the Finite State Machine (FSM) method. The FSM provides a systematic and modular framework for modeling the operational logic and task transitions of each robot. Specifically, the dynamic process of each robot is abstracted into a series of discrete states, corresponding to its different physical locations and task stages in the warehouse. By clearly defining these states and their transition relationships, it is possible to rigorously describe and analyze how robots interact with the environment, respond to scheduling commands, and collaborate with other intelligent agents under complex system conditions. In this embodiment, the possible states that each robot may experience are defined as shown in Table 2.

[0035] Table 2 Having clarified the physical meaning of each state, it is necessary to further characterize the discrete events that influence state transitions. Therefore, a binary discrete event variable is defined. , used to indicate whether a critical event occurs at discrete time t, is defined as follows: This variable provides a unified event representation for subsequent state transitions and scheduling decisions of the system. Based on this model, the following key events are defined: , represented as robot r i Assign new picking tasks; , indicating robot r i Arrive at the designated shelf; This indicates that the target picking station is idle and available for robot operation; , indicating that it is assigned to robot r i The picking task has been fully completed; , indicating robot r i Return to storage area.

[0036] Based on the definitions of the variables and events above, we obtain the core of the finite state machine—the state transition function: Robot r i The robot's next state is determined by its current state and key events, i.e., the robot's... The next state at time t+1 is determined by its current state. and triggering events In this embodiment, the state transitions are categorized into the following cases: ∩ The above specifies that after receiving a task assignment, the robot will switch from an idle state to a state of moving toward the shelf, and after arriving at the shelf, it will switch to a sorting state. The above are the decision branches after processing the shelf sorting: if the target picking station is idle, the robot goes to the picking station; otherwise, it joins the island queue to wait. The above ensures that robots waiting in the queue can go to the picking station when it is available; The above indicates that the robot transitions to working mode after arriving at the picking station; The above is used to distinguish between two situations: completed and incomplete orders. If all the SKUs carried by the robot have been picked, the robot switches to the return to storage area state; otherwise, it returns to the island queue to wait for the next target picking station to be available for picking. The above is used to form a closed loop, indicating that the robot becomes idle after returning to the storage area, waiting for the next task assignment.

[0037] This series of state transitions enables the scheduling and movement of each robot in the warehouse environment to proceed in an event-driven order. Based on the aforementioned state transition functions and rules, the state transition logic of each robot in the system based on the finite state machine model is described. Figure 3 As shown in the figure, this diagram illustrates the finite state machine of a single robot in a multi-robot warehouse system. Each box represents a robot state and a key event, and the arrows indicate how the robot transitions between states based on different key events.

[0038] II. Multi-robot motion scheduling using a circular path operation mode The Finite State Machine (FSM) model provides a solid theoretical foundation for the state transitions of robots at different stages of operation, while the scheduling algorithm achieves collaborative optimization of multi-robot path conflicts and resource competition through global task allocation and real-time queue management.

[0039] The core of this solution lies in preventing deadlocks through the implementation of a circular path mechanism. This focus also provides a rationale for the simplified processing adopted in the robot and order allocation strategy. The allocation scheme is entirely determined by the upper-level system, based on the assumption that each shelf stores only one type of SKU and is allocated to a single robot, and these two correspondences remain unchanged throughout the execution process.

[0040] First, resource allocation constraints are pre-defined and remain fixed throughout the execution period, thus eliminating the need to establish explicit optimization constraints. These constraints mainly include: 1) Order-Workstation Mapping: Before the system runs, each order is uniquely assigned to a specific picking workstation. This mapping is determined by the upper-level order allocation system and remains unchanged throughout the entire batch processing.

[0041] 2) SKU-Shelf Mapping: Each SKU (Stock Keeping Unit) is pre-loaded onto a unique shelf (Pod). This one-to-one mapping ensures the clarity of goods storage and avoids dynamic allocation conflicts during task execution.

[0042] 3) Shelf-Robot Correspondence: During the initialization phase, each shelf is exclusively assigned to a single robot. This prevents resource competition for shelf handling rights and simplifies the overall coordination of transportation tasks. (Although the shelf-to-robot ratio is typically much higher in real-world warehouse systems, this scheme intentionally increases robot density to better highlight the effectiveness of the circular path mechanism in high-congestion scenarios.)

[0043] These predefined correspondences eliminate the need to set explicit allocation constraints in the mathematical model, thereby allowing the scheduling focus to shift to robot state transitions, circular path coordination, and deadlock prevention.

[0044] Secondly, to ensure system stability under high-density operating conditions, this scheme emphasizes deadlock prevention and conflict avoidance around the picking workstations and along the circular path. The relevant constraints are not mathematically modeled but implemented through built-in operating rules in the simulation environment, specifically including: 4) Picking station occupancy: Within any given discrete time step, no more than one robot is allowed to occupy the same picking station, thereby avoiding competition for workstation resources.

[0045] 5) Circular node conflict control: Each node on the circular path can be occupied by at most one robot at any given time to prevent collisions during the flow process.

[0046] 6) Loop Vacancy Force: At least one position on the loop path is always kept vacant to ensure the continuous flow of the robot and fundamentally eliminate the possibility of robot collisions and conflicts along the loop path.

[0047] These built-in assumptions reflect the actual scheduling principles of multi-robot warehousing systems and provide a solid foundation for evaluating the effectiveness of circular path mechanisms in congested operating scenarios.

[0048] Figure 4 The diagram shows the island-loop motion scheduling algorithm designed in this scheme, including: A1. Input order, picking station, robot, shelf and SKU information, and perform initialization processing, assigning orders to picking stations, assigning SKUs to shelves and setting the initial position of robots; A2. Traverse all robots and control each robot to go to the nearest shelf to retrieve goods; A3. Determine if there are any incomplete orders. If no, end the current process. If the determination is yes, then proceed to step A4; A4. Continue to traverse the robot and determine if there is an empty picking station. If an empty picking station is found, proceed to step A5; otherwise, proceed to step A7. A5. Control the robot to go to an empty picking station to perform picking operations. Then determine whether the SKUs carried by the robot have been picked. If the determination is yes, proceed to step A6; otherwise, return to step A4. A6. Control the robot to return to the storage area to wait, and then return to step A3; A7. Control the robot to enter the roundabout path and perform roundabout movement; A8. Determine if there is an available picking station. If there is an available picking station, proceed to step A9; otherwise, return to step A7. A9. Control the robot to leave the roundabout path, and then proceed to step A5.

[0049] To verify the effectiveness of this solution, this embodiment uses a warehouse multi-robot simulation platform to compare three scheduling strategies for mitigating deadlock and congestion near picking stations: a queuing-free mechanism (robots randomly gather, easily leading to congestion and deadlock), a buffer strategy (setting up dedicated buffers to achieve orderly queuing), and the innovative island path mechanism proposed in this solution (using island paths to enable robots to dynamically and orderly cycle, further improving system efficiency). The three strategies are evaluated under the same conditions, focusing on comparing their effectiveness in terms of deadlock occurrence, total order fulfillment time, and average utilization.

[0050] The initial conditions were set as follows: the distance between adjacent grids was 25 units; the robot moved one grid or remained stationary every 0.4 seconds; the robot took 2 seconds to pick up items from the shelf; and the picking time at the picking station was 8 seconds. The robot's initial position remained consistent in all experiments.

[0051] At the start of the experiment, each of the four picking stations was assigned a unique order, with the following correspondences: {O1, S1}, {O2, S2}, {O3, S3}, {O4, S4}. Simultaneously, fifty robots were matched one-to-one with fifty SKUs, with the following mappings: {P1, A1}, ..., {P... 25 Y 25}、{P 26 Y 26}、…、{P 50 Y 50 Each order contains 20 different SKUs, specifically as follows: O1={A1, B1, C1,…, S1, T1}; O2={K1, L1, M1,…, X1, Y1, A2,…, E2}; O3={U1,…, Y1, A2, B2,…, N2, O2}; O4={F2, G2, H2,…, X2, Y2}; Assume that the storage capacity of each SKU on the shelf is sufficient to meet order demand.

[0052] 1. No queuing mechanism group like Figure 5 As shown, if waiting areas are not properly planned near the picking station, robots will randomly distribute themselves around the station after completing their picking tasks. When a robot occupies the exit aisle, other robots that have completed their picking cannot leave; once the exit aisle is completely occupied, robots that have completed their picking cannot exit the picking station, and robots waiting to enter will also be blocked, ultimately creating a deadlock situation where "external robots cannot enter, and internal robots cannot leave."

[0053] 2. Buffer Policy Group like Figure 6 As shown, a dedicated waiting area is set up between the picking station and the shelving area (coordinates 700≤x≤800, 25≤y≤600). Designated waiting positions (marked as points) are provided within the waiting area. Robots returning to the shelving area use the passageway between the two rows of waiting positions. Furthermore, prohibited areas are clearly marked with an "×" to prevent obstruction of the picking station exit passage. Only robots that need to enter the picking station are allowed to leave the waiting area; other robots remain in an orderly, stationary queue within the area. After picking, if the SKU carried by the robot is still needed, it waits in an empty position until the picking station becomes available; otherwise, it returns to the shelving area. This design effectively prevents congestion and deadlock at the exit passage.

[0054] 3. Roundabout Path Mechanism Group A circular path is constructed within a designated area (700≤x≤800, 25≤y≤600). The robot moves continuously counterclockwise along this path, eliminating the need for prolonged waiting at the picking station. The robot can dynamically enter and exit the circular path based on the idle status of the picking station, significantly alleviating queuing and deadlock problems and improving the overall system efficiency. Figure 7 As shown, the robot carrying F1 and C2 is currently performing a picking operation at the picking station, and its status is 5. The robot carrying A1 and V2 has completed its corresponding order and is waiting in the storage area, with a status of 0. The robot carrying C1 has completed its corresponding order and returned to the storage area, with a status of 6. The robot carrying O2 has left the roundabout and is moving towards the picking station, with a status of 3. The robot carrying J2 has completed its picking operation, but still has other orders to process, so it has returned to the path switching area; the remaining robots are all performing roundabout movement operations, and their statuses are all 4.

[0055] Each experiment was repeated 10 times under identical conditions, including the same order demand, SKU distribution, robot initial position, and the same loop area and waiting area locations. The following key metrics were recorded during the experiments (average of the 10 experiments): The number and duration of deadlocks; Total order fulfillment time: The total time required for four picking stations to complete one round of orders, denoted as "Total order fulfillment time". ”; Picking station resource utilization rate: The percentage of time a picking station actually processes picking tasks within the total experimental time, reflecting the effective utilization of picking station resources.

[0056] Among them, sorting station s i The formula for calculating utilization rate is: The formula for calculating the average utilization rate of multiple picking stations is: Although the experimental setup ensured that the number of orders allocated to each picking station, the number of SKUs per order, and the robot picking time were consistent, in reality, due to dynamic scheduling and path conflicts, the order completion time of each picking station may differ. Therefore, this embodiment calculates the resource utilization rate of each picking station separately and uses the average value as the evaluation index. The experimental results are shown in Table 3.

[0057] Table 3 The experimental results show that for the group without a queuing mechanism, the robots form a circular waiting pattern at the picking station exit, leading to a deadlock. Both internal and external robots are blocked, causing long-term congestion and preventing orders from being completed.

[0058] For the buffer strategy group, deadlocks were eliminated. The average total order fulfillment time was 340 seconds. Each picking station processed one round of orders containing 20 SKUs (8 seconds picking time per SKU), with an average utilization rate of 47.05%.

[0059] For the island-loop path mechanism group, deadlock was eliminated. Robots maintain dynamic flow, allowing for real-time order routing based on picking station availability, thus improving response speed. Total order fulfillment time was further reduced to 272 seconds, with an average utilization rate of 58.82%.

[0060] Experiments show that the island path mechanism, while avoiding deadlocks, further shortens the total order fulfillment time, reducing picking time by 68 seconds per round and significantly improving system throughput. Compared to traditional buffer strategies, the island path mechanism alleviates peak queuing pressure, enhances resource flow around the picking station, and increases the average utilization rate of the picking station by 11.77%. Although moving along the island path increases energy consumption, it significantly shortens the total order fulfillment time—especially suitable for large-scale warehousing systems—effectively improving overall efficiency and resource utilization.

[0061] In summary, the warehouse multi-robot scheduling and deadlock prevention method based on the island path mechanism proposed in this invention effectively solves the limitations of traditional scheduling strategies in deadlock prevention, resource utilization and efficiency improvement by constructing a circular buffer and dynamic cyclic path to guide the robot to move continuously along the island, establishing a finite state machine model and a dynamic island scheduling algorithm and a system constraint mathematical model, and demonstrates significant technical advantages.

[0062] This method eliminates the "hold and wait" and "circular wait" conditions that lead to deadlock by guiding robots to move continuously and dynamically along a roundabout path. Robots no longer statically occupy path nodes waiting for picking station resources, avoiding the state of "holding some resources while waiting for others." Simultaneously, the unidirectional flow of the roundabout path cuts off the possibility of multiple robots forming a closed-loop blockage, completely resolving the deadlock risk of "external robots unable to enter, internal robots unable to leave" in traditional buffer strategies. Experimental data shows that after adopting the solution of this invention, the system experiences no deadlocks, significantly outperforming the continuous deadlock state of groups without queuing mechanisms.

[0063] Leveraging the dynamic response capability of the island-like path, robots on the loop can quickly switch to operational status when the picking station is idle, reducing unnecessary waiting caused by static queuing. The finite state machine model's precise control over robot state transitions (such as from "island movement" to "heading to the picking station") further enhances the immediacy of task processing. Experimental results show that, compared to traditional buffer strategies, the total order fulfillment time of this invention is reduced from 340 seconds to 272 seconds, and the average picking station utilization rate increases from 47.05% to 58.82%, significantly accelerating the entire process from order allocation to completion, and significantly enhancing the overall resource efficiency of the system.

[0064] Meanwhile, this invention combines constraints (such as path node uniqueness and picking station exclusivity) with finite state machine algorithms to ensure stable operation even under high-load scenarios with a surge in robot density and a wide variety of order SKUs. The dynamic mechanism of the island path can flexibly adapt to increases or decreases in the number of robots (such as adding new robots directly to the loop path) and increases in order complexity (such as collaborative picking of multi-SKU orders), providing reliable technical support for the large-scale expansion and business upgrade of the warehousing system.

[0065] Applying the method proposed in this invention to practice, through the innovative design of the island path mechanism, it is possible to achieve comprehensive optimization of the multi-robot system in warehousing, which is characterized by "no deadlock, high efficiency, high utilization rate, and strong adaptability", thus providing a practical solution for the efficient and stable operation of intelligent warehousing and logistics.

Claims

1. A multi-robot scheduling and control method based on island-loop motion and deadlock prevention, characterized in that, Includes the following steps: S1. Construct a circular buffer and dynamic loop path between the picking station and the storage area, and perform finite state machine modeling for the dynamic behavior of the robot to obtain the finite state machine model corresponding to different robots. S2. Based on the finite state machine model of the robot, combined with resource allocation constraints and preset operation constraints, the multi-robot scheduling strategy is obtained by executing the island motion scheduling algorithm to control the working state of each robot accordingly.

2. The multi-robot scheduling and control method based on island motion and deadlock prevention according to claim 1, characterized in that, The process of modeling the robot's dynamic behavior using a finite state machine in step S1 includes: abstracting the robot's dynamic process into a series of discrete states, which correspond to different physical locations and task stages of the robot in the warehouse. Define binary discrete event variables and key events, and construct the robot's state transition function. The binary discrete event variables are used to represent whether a key event occurs at discrete time t, and the state transition function is used to describe the transition relationship between the current state and the next state of the machine based on the key event.

3. The multi-robot scheduling and control method based on island motion and deadlock prevention according to claim 2, characterized in that, The discrete states specifically include the following states: The system is either stationary in the storage area or waiting for task assignment. Head to the shelf; Pick up your order from the shelf; Proceed to an available picking station; To navigate around the island or return to the island's original path; Working at the sorting station; Return to storage area.

4. The multi-robot scheduling and control method based on island motion and deadlock prevention according to claim 2, characterized in that, The specific binary discrete event variable is: The key events include: , represented as robot r i Assign new picking tasks; , indicating robot r i Arrive at the designated shelf; This indicates that the target picking station is idle and available for robot operation; , indicating that it is assigned to robot r i The picking task has been fully completed; , indicating robot r i Return to storage area.

5. A multi-robot scheduling and control method based on island motion and deadlock prevention according to claim 4, characterized in that, The robot's state transition function is specifically as follows: in, This is the robot's next state. This represents the robot's current state. These are the key events between the robot's current state and the next state. This is the state transition function, which indicates that the robot will transition to the next state after encountering a critical event in its current state.

6. A multi-robot scheduling and control method based on island motion and deadlock prevention according to claim 1, characterized in that, The resource allocation constraints in step S2 include the correspondence between orders and workstations, the correspondence between SKUs and shelves, and the correspondence between shelves and robots. These resource allocation constraints remain fixed during the execution of the island motion scheduling algorithm.

7. A multi-robot scheduling and control method based on island motion and deadlock prevention according to claim 6, characterized in that, The order-workstation correspondence specifically refers to the fact that before operation, each order is uniquely assigned to a specific picking workstation; The SKU-shelf correspondence specifically refers to the fact that each SKU is pre-loaded on a unique shelf pod; The shelf-robot correspondence specifically refers to the fact that during the initialization phase, each shelf is exclusively assigned to a single robot.

8. A multi-robot scheduling and control method based on island motion and deadlock prevention according to claim 1, characterized in that, In step S2, the preset operating constraints are specifically set through the built-in operating rules in the simulation environment, including picking station occupancy constraints, loop node conflict control constraints, and loop empty space forced constraints.

9. A multi-robot scheduling and control method based on island motion and deadlock prevention according to claim 8, characterized in that, The picking station occupancy constraint specifically means that within any given discrete time step, no more than one robot is allowed to occupy the same picking station. The loop node conflict control constraint specifically means that each node on the loop path can be occupied by at most one robot at any given time. The mandatory constraint on loop vacancy specifically refers to ensuring that at least one position on the loop path is always vacant.

10. A multi-robot scheduling and control method based on island motion and deadlock prevention according to claim 1, characterized in that, The execution process of the island motion scheduling algorithm includes: A1. Input order, picking station, robot, shelf and SKU information, and perform initialization processing, assigning orders to picking stations, assigning SKUs to shelves and setting the initial position of robots; A2. Traverse all robots and control each robot to go to the nearest shelf to retrieve goods; A3. Determine if there are any incomplete orders. If no, end the current process. If the determination is yes, then proceed to step A4; A4. Continue to traverse the robot and determine if there is an empty picking station. If an empty picking station is found, proceed to step A5; otherwise, proceed to step A7. A5. Control the robot to go to an empty picking station to perform picking operations. Then determine whether the SKUs carried by the robot have been picked. If the determination is yes, proceed to step A6; otherwise, return to step A4. A6. Control the robot to return to the storage area to wait, and then return to step A3; A7. Control the robot to enter the roundabout path and perform roundabout movement; A8. Determine if there is an available picking station. If there is an available picking station, proceed to step A9; otherwise, return to step A7. A9. Control the robot to leave the roundabout path, and then proceed to step A5.

Images