Multi-Alos robot cooperative sorting method for warehouse logistics and related device

Abstract

The invention provides a multi-Alos robot cooperative sorting method for warehouse logistics and a related device, and relates to the technical field of warehouse logistics. Receiving real-time order data, and performing analysis according to the real-time order data to generate a task set including cargo positions, target sorting areas and priorities; dynamically calculating a task matching degree by combining state information of each robot through a distributed task allocation mechanism, and optimizing global task allocation by adopting a reinforcement learning model to generate a global task allocation result; modeling a storage environment into a space-time diagram structure on the basis of a global distribution strategy; generating a conflict-free path by adopting a path-finding algorithm in combination with the space-time diagram structure, and optimizing a trajectory in combination with a task time limit constraint; and according to the target position in the conflict-free path, goods are recognized through a visual system, and grabbing and placing are completed by combining with a mechanical arm. The technical effects that the sorting efficiency and adaptability are improved, and the operation and maintenance cost is reduced are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of warehouse logistics, in particular to a multi-Aelos robot collaboration sorting method for warehouse logistics and a related device. BACKGROUND

[0002] In the field of warehouse logistics, cargo sorting as a core link directly determines the operation cost and customer satisfaction of the overall logistics system. Currently, traditional sorting schemes mainly rely on manual operation or fixed automatic equipment, but there are significant limitations in actual application. Manual sorting has flexibility, but is low in efficiency and high in error rate, especially when orders surge. As labor costs rise, logistics systems that rely on manual labor face significant economic pressure. While fixed automatic equipment (such as conveyors and robotic arms) can replace some manual labor, their rigid design results in insufficient flexibility and adaptability. Such equipment is usually developed for specific tasks and is difficult to adapt to dynamic warehouse environments and diverse order demands. The high deployment and maintenance costs and strict requirements for space layout make it difficult to modify when business expands or order fluctuations increase.

[0003] In recent years, multi-robot collaboration systems have become a research hotspot. By distributing tasks and collaborating, they significantly improve sorting efficiency and flexibility. However, existing technologies still face many bottlenecks: centralized task allocation relies on a single control node, which has high communication overhead and poor real-time performance, making it difficult to adapt to dynamic task changes; traditional path planning algorithms (such as A , Dijkstra) ignore the time dimension, leading to frequent path conflicts and collisions; collaboration control relies on preset rules, making it difficult to meet the high requirements of precision and stability. Although some solutions introduce automation technology, they still have high single-point failure risk, frequent path conflicts, and insufficient adaptive capacity.

[0004] With the rapid development of artificial intelligence, the Internet of Things, and robotics, warehouse logistics is accelerating its transformation towards intelligence and flexibility. The market's demand for efficient and low-cost dynamic sorting is increasingly urgent, and there is an urgent need for an innovative solution that integrates dynamic task allocation, spatiotemporal path planning, and precise collaboration control. SUMMARY

[0005] To meet the demand for efficient and low-cost dynamic sorting, the present application provides a multi-Aelos robot collaboration sorting method for warehouse logistics and a related device.

[0006] In a first aspect, the present application provides a multi-Aelos robot collaboration sorting method for warehouse logistics, which adopts the following technical solution: A multi-Aelos robot collaboration sorting method for warehouse logistics, comprising: Receiving real-time order data, parsing according to the real-time order data to generate a task set containing goods location, target sorting area and priority; Through a distributed task allocation mechanism, dynamically calculating task matching degree combined with the state information of each robot, and using a reinforcement learning model to optimize the global task allocation to generate a global task allocation result; Based on the global allocation strategy, modeling the warehouse environment as a space-time graph structure; Using a pathfinding algorithm combined with the space-time graph structure to generate a conflict-free path and combining with the task time limit constraint to optimize the trajectory; According to the target position in the conflict-free path, identifying the goods through a vision system and completing the grabbing and placing combined with a force-controlled robot arm.

[0007] Optionally, generating a task set includes: Automatically identifying the weight, size and fragility characteristics of the goods; Promotional orders are automatically marked as the highest priority and assigned to an emergency processing channel; When new orders arrive, immediately refresh the task queue to ensure timeliness.

[0008] Optionally, dynamically calculating the task matching degree includes: Comprehensive multi-dimensional state evaluation of real-time robot position, remaining power, current load and task urgency; Automatically adjusting the allocation strategy weight based on historical sorting data to achieve reinforcement learning optimization; Setting up a fast lane for high-priority orders to skip the regular queuing process to achieve a bidding-style task allocation.

[0009] Optionally, constructing a space-time graph structure includes: Adding timestamps to key locations for dynamic node labeling; Automatically increasing the passage cost according to the real-time robot density of the channel; Reserving exclusive passage time periods for urgent tasks to achieve time window constraint integration.

[0010] Optionally, generating a conflict-free path includes: Hierarchical planning strategy, using space-time A algorithm to generate the main path, and using a potential field model to avoid dynamic obstacles in real time at the local level; Inserting a smooth curve segment to the sharp turn path to ensure stable turning of the robot to achieve trajectory smoothing; Assigning a straight fast lane to the robot for high-value orders to achieve priority passage rights.

[0011] Optionally, the grabbing and placing operation includes: Visual-force control dual closed loop: visual system millimeter level positioning goods coordinates, force control robot arm according to the characteristics of goods automatically adjust the clamping force; Multi-machine cooperative anti-collision: the central controller assigns the adjacent robots to staggered grabbing time periods; Special handling of fragile goods: automatically enable buffer grabbing mode for glass products and the like.

[0012] Optionally, the method further comprises intelligent fault response: Multi-level early warning mechanism: when the power is less than 20% or the temperature exceeds 70℃, trigger a three-level alarm; Seamless task transfer: automatically transfer tasks according to the position of adjacent robots, idle state and load capacity; Load balancing guarantee: intelligent splitting of overweight tasks into multiple subtasks for distribution.

[0013] In a second aspect, the present application provides a multi-Aelos robot collaboration sorting device for warehouse logistics, comprising: A data acquisition module for receiving real-time order data, analyzing the real-time order data to generate a task set containing goods location, target sorting area and priority; A result distribution module for dynamically calculating task matching degree by a distributed task distribution mechanism combined with the state information of each robot, and optimizing global task distribution to generate global task distribution results by using a reinforcement learning model; A space-time graph module for modeling the warehouse environment as a space-time graph structure based on the global allocation strategy; An optimization module for generating conflict-free paths by using a pathfinding algorithm combined with the space-time graph structure and optimizing trajectories combined with task time limit constraints; An execution module for identifying goods by a visual system according to the target position in the conflict-free path, and completing grabbing and placing by a force control robot arm.

[0014] In a third aspect, the present application provides a computer device, comprising a memory and a processor, wherein the processor executes computer instructions stored in the memory to perform the method as described above.

[0015] In a fourth aspect, the present application provides a computer readable storage medium comprising instructions, which, when executed on a computer, cause the computer to perform the method as described above.

[0016] In summary, the application receives real-time order data, parses the real-time order data to generate a task set containing cargo location, target sorting area and priority; through a distributed task allocation mechanism, dynamically calculates the task matching degree combined with the state information of each robot, and uses a reinforcement learning model to optimize the global task allocation to generate a global task allocation result; based on the global allocation strategy, the warehouse environment is modeled as a space-time graph structure; a pathfinding algorithm is used to generate a conflict-free path combined with the space-time graph structure and optimize the trajectory combined with the task time limit constraint; according to the target position in the conflict-free path, the cargo is identified through a vision system, and a force-controlled robotic arm is used to complete picking and placing. The technical effects of improving sorting efficiency and adaptability and reducing operation and maintenance costs are achieved. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a computer device structure schematic diagram of a hardware running environment related to an embodiment scheme of the application. Figure 2 is a flowchart of a first embodiment of a multi-Aelos robot collaborative sorting method for warehouse logistics of the application. Figure 3 is a dynamic task allocation algorithm flowchart. Figure 4 is a space-time path planning algorithm flowchart. Figure 5 is a structure block diagram of a first embodiment of a multi-Aelos robot collaborative sorting device for warehouse logistics of the application. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical scheme and advantages of the application clearer, the application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application and do not limit the application.

[0019] Referring to Figure 1 , Figure 1 is a computer device structure schematic diagram of a hardware running environment related to an embodiment scheme of the application.

[0020] As Figure 1As shown, the computer device can include a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to realize the connection communication between these components. The user interface 1003 can include a display, an input unit such as a keyboard, and optionally a standard wired interface, a wireless interface. The network interface 1004 can optionally include a standard wired interface, a wireless interface (such as a wireless fidelity (Wi-Fi) interface). The memory 1005 can be a high-speed random access memory (RAM), and can also be a stable non-volatile memory (NVM), such as a disk memory. The memory 1005 can also be a storage device independent of the aforementioned processor 1001.

[0021] Those skilled in the art can understand that Figure 1 The structure shown in the figure does not constitute a limitation on the computer device, and can include more or fewer components than the figure, or combine certain components, or different component arrangements.

[0022] As Figure 1 As shown, the memory 1005 as a storage medium can include an operating system, a network communication module, a user interface module, and a multi-Aelos robot collaborative sorting program for warehouse logistics.

[0023] In Figure 1 In the computer device shown, the network interface 1004 is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with a user; the processor 1001 and the memory 1005 in the present application can be arranged in the computer device, and the computer device calls the multi-Aelos robot collaborative sorting program for warehouse logistics stored in the memory 1005 through the processor 1001, and executes the multi-Aelos robot collaborative sorting method for warehouse logistics provided in the present application.

[0024] The present application provides a multi-Aelos robot collaborative sorting method for warehouse logistics, which is described with reference to Figure 2 , Figure 2 The flowchart of the first embodiment of the multi-Aelos robot collaborative sorting method for warehouse logistics in the present application is shown.

[0025] In this embodiment, the multi-Aelos robot collaborative sorting method for warehouse logistics includes the following steps: Step S10: receiving real-time order data, and parsing the real-time order data to generate a task set containing a cargo location, a target sorting area, and a priority.

[0026] It should be noted that the dynamic task allocation algorithm flowchart is as shown in Figure 3 Generating a task set includes automatically identifying cargo weight, size, and fragility characteristics; automatically marking promotional orders as the highest priority and assigning an emergency processing channel; and refreshing the task queue immediately when new orders arrive to ensure timeliness.

[0027] In a specific implementation, after the central control system receives real-time order data transmitted by the warehouse management system, it parses and generates a sorting task set, each task containing key parameters such as cargo location, target sorting area, priority, and cargo attributes. The system uses a distributed task allocation mechanism.

[0028] Step S20: Through the distributed task allocation mechanism, dynamically calculate the task matching degree combined with the state information of each robot, and use the reinforcement learning model to optimize the global task allocation to generate the global task allocation result.

[0029] In a specific implementation, dynamically calculating the task matching degree includes: multi-dimensional state evaluation based on real-time location, remaining power, current load capacity, and task urgency of the robot; automatically adjusting the allocation strategy weight based on historical sorting data to achieve reinforcement learning optimization; setting a fast channel for high-priority orders to skip the regular queuing process to achieve a bidding-style task allocation.

[0030] It should be noted that each Aelos robot dynamically calculates the task matching degree based on its real-time state (such as location, remaining power, and load capacity), and submits the task value evaluation result to the central system through a bidding-style strategy. The central system optimizes global task allocation decisions based on a reinforcement learning model, prioritizing high-priority tasks to the most efficient robots, while triggering real-time task redistribution when orders change or robots fail, ensuring the continuity and response speed of the sorting process. For example, when a promotional order is detected to be delayed, the system automatically adjusts the allocation weight, strengthens the time constraint, and ensures that urgent tasks are completed first.

[0031] It should be noted that task modeling and state representation include: each sorting task is defined as a four-tuple containing spatiotemporal constraints: The cargo location Real-time positioning by laser SLAM (accuracy ±2 cm), task time limit Dynamic setting according to order urgency (such as express order seconds), priority Task importance differentiation (range 1-5), cargo attributes Record weight, size, and special handling requirements. Robot state is represented by a real-time vector: where Updated by the UWB indoor positioning system (frequency 50 Hz), Monitor remaining power, Record current load (maximum load ), for the task queue to be executed. These parameters are synchronized to the central controller at a frequency of 10 Hz, providing a real-time data basis for dynamic decision-making.

[0032] Step S30: Based on the global allocation strategy, model the warehouse environment as a space-time graph structure.

[0033] It should be noted that constructing the space-time graph structure includes: adding timestamps to key locations for dynamic node labeling; automatically increasing the passage cost according to the real-time robot density of the channel; reserving exclusive passage time for emergency tasks to achieve time window constraint integration.

[0034] It should be noted that during the sorting process, the system models the warehouse environment as a space-time graph structure, and the nodes of key locations (such as shelves and sorting areas) are timestamped to record the robot occupancy state, and the edge weight dynamically reflects the degree of path congestion. The global path planning uses an improved space-time A algorithm, combined with time window constraints and curvature smoothing optimization, to generate conflict-free paths. The robot uses laser radar and visual sensors to perceive dynamic obstacles (such as other robots or temporary stacked goods) in real time, and uses a potential field model to adjust the motion trajectory, avoiding collisions and optimizing passage efficiency. For example, when the channel is congested, the system automatically guides the robot to detour or wait, ensuring that high-priority tasks have priority in passing, ultimately reducing the path conflict rate from 12% in traditional systems to 3.5%.

[0035] In specific implementation, the core of task allocation is to calculate the matching degree between the robot and the task : This function optimizes the spatial distance, time urgency, and load balancing through a three-item weighted sum. The weight coefficients (initial value) are dynamically adjusted through reinforcement learning, and the reward function is designed as: The first reward is for the timely completion of high-priority tasks (delay time The exponential decay of the algorithm strengthens the time constraint, the second term penalizes load imbalance, and the last term suppresses path conflict. The Q-learning algorithm uses... -greedy strategy ( After 1000 iterations of training, the system converges to a stable policy, enabling it to adaptively adjust weight parameters. For example, when a high-priority task's latency rate exceeds a threshold, it automatically increases the priority level. The weight is increased to 0.4 to strengthen the time constraint.

[0036] It should be noted that the actual task allocation adopts a bidding mechanism with time and space constraints as follows: Among them, waiting time The exponential decay term prevents task starvation, and the emergency task bias term ( (For the indicator function) to ensure priority response to critical orders. The central controller runs the Hungarian algorithm every 200ms to solve for the optimal allocation matrix, while simultaneously monitoring multiple condition-triggered events: When the robot's battery level drops below 20%, the remaining task time is less than 60 seconds, or there are more than 3 collisions in a single area, the system will initiate task reallocation within 50ms and update the environmental access costs. This formula uses the congestion coefficient ( ) and channel capacity limitations ( Dynamically adjust path weights. For example, when three robots are simultaneously present in a certain passage, the passage cost increases to [a certain value]. This guides subsequent robots to choose alternative paths.

[0037] Step S40: Use a pathfinding algorithm combined with the spatiotemporal graph structure to generate a conflict-free path and optimize the trajectory in combination with task time constraints.

[0038] In practical implementation, the flowchart of the spatiotemporal path planning algorithm is as follows: Figure 4 As shown. Generating conflict-free paths includes: a hierarchical planning strategy, employing spatiotemporal A at the global level. The algorithm generates the main path and avoids dynamic obstacles in real time at the local layer using a potential field model; it inserts smooth curve segments into sharp turns to ensure stable robot turning and achieve trajectory smoothing; and it allocates straight fast lanes to robots handling high-value orders to achieve priority passage.

[0039] In practical implementation, spatiotemporal path planning and dynamic obstacle avoidance include: First, the warehousing environment is abstracted into a spatiotemporal diagram structure. , where nodes Key locations such as shelves, sorting areas, charging stations, etc. are represented and time-stamped The occupancy status of the robot is recorded (e.g. the time window in which the robot enters and leaves a node). The weights of the connections between adjacent nodes The real-time passage cost of the path is dynamically reflected, and the calculation formula is: wherein, is the basic length of the path, represents the number of robots passing through the path within the time window is the channel capacity threshold.

[0040] The global path planning uses an improved space-time A algorithm, which ensures the conflict-free of multi-robot paths by introducing time window constraints and curvature smoothing optimization. The heuristic function is designed as: wherein, the first term is the Euclidean distance from the current node to the target, and the second term is the time penalty term, which is used to strengthen the time constraints of urgent tasks. If the path curvature exceeds the threshold , smooth turning is achieved by inserting B-spline control points to ensure the stability of robot motion, and the curvature calculation formula after path smoothing is: Local dynamic obstacle avoidance combines real-time perception and potential field model to solve the interference of temporary obstacles (such as other robots or moving goods). The robot obtains environmental point cloud data through laser radar and vision sensor, and constructs a local obstacle map using occupancy grid mapping method. When the grid occupancy probability exceeds 0.7, it is marked as an obstacle. The motion trajectory adjustment is based on the virtual potential field model, and the total force is composed of attractive force, repulsive force and traffic guiding force: Attractive force: guide the robot to move along the global path to the target, and the calculation formula is: Repulsive force: avoid obstacles, and the calculation formula is: Traffic guiding force: dynamically adjust the path weight according to the degree of channel congestion, guide the robot to bypass high-density areas, and the calculation formula is: ​To improve dynamic adaptability, the system updates the environment state every 200 ms and shares the real-time position and task progress of each robot through the ROS communication network. When a path conflict or temporary obstacle is detected, the central controller triggers local path re-planning within 50 ms and generates a smooth bypass trajectory combined with the potential field model. For example, in a channel congestion scenario, the system automatically guides low-priority task robots into the waiting queue while assigning a fast-passing path for high-priority task robots, reducing the overall path conflict rate from 12% in traditional systems to 3.5%.

[0041] In addition, the path planning module cooperates deeply with the task allocation and fault recovery mechanism. When a robot interrupts the task due to a fault, the system not only reallocates the task but also adjusts the path weight of other robots in real time based on the updated space-time graph to avoid resource conflicts. Simulation tests show that when 50 robots work cooperatively, the dynamic obstacle avoidance response time is less than 0.5 seconds, and the path planning success rate is over 98%, significantly improving the robustness and efficiency of the sorting system.

[0042] Step S50: According to the target position in the conflict-free path, the goods are identified by the vision system, and the grabbing and placing are completed by the force-controlled robotic arm.

[0043] It should be noted that the grabbing and placing operation includes: vision-force control double closed loop: the vision system locates the goods coordinates at millimeter level, and the force-controlled robotic arm automatically adjusts the clamping force according to the characteristics of the goods; multi-machine cooperative anti-collision: the central controller allocates staggered grabbing time periods for adjacent robots; special treatment for fragile goods: automatic buffer grabbing mode is enabled for glass products and the like.

[0044] In specific implementation, the robot identifies the target goods through a high-precision vision system, matches the goods feature database combined with deep learning algorithm, and locates the accurate coordinates on the shelf. The six-axis force-controlled robotic arm dynamically adjusts the clamping force, and realizes stable grabbing and placing of goods through multi-sensor fusion technology (vision feedback, torque detection, inertial navigation). After being carried to the sorting area, the robot classifies the goods to the specified outbound area according to the order information, and ensures the sorting accuracy to reach millimeter level through the vision guidance and positioning system. For example, when grabbing fragile goods, the system automatically reduces the clamping force to avoid damage to the goods.

[0045] It should be noted that the multi-robot communication network based on the ROS framework shares the state information (such as position, task progress, and remaining power) of each robot in real time. When a robot triggers a fault due to insufficient power, excessive temperature, or communication delay, the central system automatically starts a fault response mechanism, selects the optimal support robot according to the remaining load capacity and task priority score of the adjacent robots, and takes over the task. For example, when the load of the faulty robot is 50 kg, the system preferentially allocates 30 kg to the adjacent idle robot, and the remaining 20 kg is shared by other robots, while updating the global path planning to avoid resource conflicts, and the fault recovery time is controlled within 8.5 seconds.

[0046] It can be understood that the method further includes intelligent fault response: multi-level warning mechanism: triggering a three-level alarm when the power is less than 20% or the temperature exceeds 70°C; seamless task transfer: automatically transferring tasks according to the positions, idle states, and load capacities of adjacent robots; load balancing guarantee: intelligently splitting an overweight task into multiple sub-tasks for distribution.

[0047] In a specific implementation, the visual positioning and three-dimensional coordinate mapping include: the robot uses a binocular vision system to capture an RGB-D image of the shelf area in real time, and constructs a high-precision environment point cloud map in combination with an improved ORB-SLAM algorithm. The pixel coordinates of the target goods are mapped to three-dimensional space coordinates by using a camera intrinsic matrix : wherein, is a camera focal length, is a principal point coordinate. For high-reflectivity or low-texture goods, the system introduces a multi-frequency stripe structured light projection technology to compensate for the lack of depth information through phase solving. In combination with a Harris-Laplace corner detection algorithm, sub-pixel level edge features are extracted, and the positioning error is controlled within Under a dynamic warehouse environment, ORB-SLAM updates the goods pose in real time through key frame matching and local map optimization, eliminating the positioning drift caused by goods displacement or stacking deformation.

[0048] The adaptive force control algorithm and clamping optimization include: The six-axis force control robot adjusts the clamping force based on a dynamic PID model, and the target clamping force is dynamically set according to the weight, material, and vulnerability of the goods. The adaptive PID controller is designed as: wherein, the error term is: Real-time feedback from six-axis force / torque sensor, and high-frequency noise is suppressed by Kalman filter. The mapping relationship with the target gripping force is through a pre-calibrated nonlinear function: The implementation ensures that the gripping force is smoothly transitioned within to . For different cargo characteristics, the system automatically switches control modes: when grabbing rigid cargo, the proportional coefficient is increased to enhance anti-interference ability; when handling fragile cargo, the integral term is increased to eliminate steady-state error.

[0049] Multi-modal perception fusion and dynamic correction include: The stability during the grabbing process depends on the deep fusion of vision, torque feedback, and inertial navigation data.

[0050] Vision servo correction: real-time tracking of cargo pose offset by binocular vision, generating joint space compensation instructions: where is the Jacobian matrix of the robot arm. The Lie group space interpolation algorithm is used to ensure smooth and continuous compensation trajectory, avoiding jitter caused by robot kinematic singularity.

[0051] Inertial anti-shake control: based on IMU to detect the vibration acceleration of the robot arm end , using a frequency domain notch filter to suppress high-frequency jitter above.

[0052] Contact force perception: when the force / torque sensor detects a sudden change in contact force (such as cargo slipping), an emergency retreat instruction is triggered. The retreat path is planned based on the Bezier curve, ensuring that the robot retreats with the smallest curvature trajectory.

[0053] Multi-machine cooperation collision avoidance strategy includes: The central controller assigns the grabbing area through a space-time constraint model to avoid robot workspace overlap. The priority preemption mechanism dynamically calculates the authority distribution weight based on task urgency and robot real-time load: When the paths of two robots overlap, the system dynamically adjusts the trajectory through a repulsive force model: If the distance between the robot arm end and , the system triggers collision avoidance trajectory re-planning, dynamically adjusting the detour strategy combined with global path weight.

[0054] In specific implementations, the multi-level fault detection and dynamic response includes: the robot monitors real-time operation state parameters, including remaining power , motor temperature and communication delay , through built-in sensors. When any parameter exceeds the safety threshold , or , the system triggers a three-level response mechanism: the primary response immediately freezes the motion command of the faulty robot and broadcasts the fault code (including fault type, location and load status); the secondary response calculates the priority score of the support robot based on the real-time state of the adjacent robots (position , remaining load and task priority ): is the Euclidean distance between the support robot and the fault point, is the maximum load capacity of the robot, represents the proportion of its idle time. If the remaining load of a single robot is insufficient to take over the task, the system splits the task into multiple subtasks through the Hungarian algorithm to ensure that the global load deviation rate is less than , and the task splitting formula is: where is the total weight of the tasks to be transferred by the faulty robot, is the minimum splitting number.

[0055] Dynamic task transfer and load balancing includes: During the task transfer process, the system processes the execution queue of the faulty robot in descending order of urgency . For heavy goods, a parallel transfer strategy is adopted: the task is split into subtasks, each with a weight of: and an optimal allocation matrix is generated through an improved Hungarian algorithm, with the objective function being: where is the task priority, is the distance from the support robot to the task point. In the load verification stage, the actual load of each support robot is calculated: If overloaded, the overloaded part is redistributed to ensure global load balancing: Global path coordination and conflict resolution includes: After the task transfer is completed, the system updates the path weight in the space-time graph model , and adjusts the motion trajectory of the support robot based on the dynamic potential field model. When the number of channel robots , the passage cost increases to , guiding the robot to bypass the low-density area; if the support robot and other robots overlap in path, the anti-collision trajectory optimization is triggered, and the repulsive force calculation formula is: High-priority task robot preferentially acquires fast passage path, and low-priority robot enters the waiting queue, minimizing the impact of path conflict on overall efficiency.

[0056] Digital twin verification and self-learning optimization includes: The system simulates the sorting process through the digital twin engine, and preforms the feasibility of the task transfer scheme. If the simulation sorting efficiency and conflict rate , the execution instruction is issued; otherwise, the task splitting granularity and path weight coefficient are recalculated. The historical fault data is input into the reinforcement learning model to optimize the weight parameters of the scoring function (equation 16), and the objective function is: After 5000 iterations of training, the fault recovery time is shortened from the initial to , and the task transfer success rate is improved to .

[0057] Digital twin and performance verification includes: through the deep integration of digital twin technology and deep reinforcement learning, a virtual-real collaborative warehouse sorting simulation environment is constructed to realize the whole life cycle verification and dynamic optimization of the sorting process. The following explains the practical application and implementation details of the technical scheme from the dimensions of digital twin modeling and reinforcement learning training.

[0058] Digital twin modeling and virtual-real synchronization includes: the digital twin engine builds a high-precision virtual warehouse environment based on the Unity3D framework, and the physical layer and the virtual layer realize bidirectional real-time data synchronization through the ROS middleware. The physical layer robot pose data is updated to the virtual environment at a frequency of , and the cargo state (position, grabbing state) is transmitted incrementally through the event-driven mode, with a synchronization error controlled within inside. The key parameters of the virtual environment (shelf layout, robot kinematics, weight distribution of goods) are fully consistent with the physical layer, and the physical interaction behavior of the warehouse scene is dynamically simulated. For example, when there is a surge in promotional orders, the system automatically generates a peak order data stream (such as every hour single), simulates the sorting process through the virtual environment, and verifies the robustness of the path planning algorithm under high load. In addition, the dynamic obstacle generation module can simulate scenarios such as shelf collapse and temporary stacking of goods, testing the dynamic obstacle avoidance and task reassignment capabilities of the robot. The virtual-real synchronization mechanism ensures that the simulation results can be directly mapped to the physical layer, providing reliable pre-validation for actual deployment.

[0059] The reinforcement learning model and dynamic optimization include: a deep reinforcement learning model that maximizes sorting efficiency and minimizes conflict rate as the optimization goal, designs a state space (containing robot position, power, load, and task queue length) and an action space (A={assign, preempt, wait}). The reward function integrates task completion rate, conflict frequency, and energy consumption indicators: The first reward is for timely completion of high-priority tasks (exponential decay of delay time time limit constraint), the second reward suppresses path conflict frequency, and the last reward punishes energy consumption. The training process uses a double deep Q network (DDQN) algorithm, with the target network updated every step, the experience replay pool capacity is set to , the exploration rate linearly decays from the initial value to . Prioritized experience sampling (Prioritized Experience Replay) is used to improve the learning efficiency of key states, and the sampling probability calculation formula is: where is the temporal difference error . After iterations of training, the task completion rate improves from the initial to , and the conflict rate stabilizes below . The training results are mapped to the physical layer in real time through the digital twin environment, dynamically adjusting the task assignment weight and path planning parameters, such as automatically increasing the congestion penalty coefficient in the spatio-temporal graph model when detecting a high-frequency path conflict area, guiding the robot to bypass low-density channels.

[0060] The embodiment receives real-time order data, parses the real-time order data to generate a task set containing the location of goods, target sorting area and priority; through a distributed task allocation mechanism, dynamically calculates the task matching degree in combination with the state information of each robot itself, and uses a reinforcement learning model to optimize the global task allocation to generate a global task allocation result; the warehouse environment is modeled as a space-time graph structure based on the global allocation strategy; a conflict-free path is generated using a pathfinding algorithm in combination with the space-time graph structure, and the trajectory is optimized in combination with the task time limit constraint; the goods are identified through a vision system according to the target position in the conflict-free path, and the force-controlled mechanical arm is used to complete the grabbing and placing. The technical effects of improving the sorting efficiency and adaptability and reducing the operation and maintenance cost are achieved.

[0061] In addition, the embodiment of the present application also provides a computer readable storage medium, and the storage medium stores a program for multi-Aelos robot cooperative sorting in warehouse logistics. When the program for multi-Aelos robot cooperative sorting in warehouse logistics is executed by a processor, the steps of the method for multi-Aelos robot cooperative sorting in warehouse logistics are implemented.

[0062] Reference Figure 5 , Figure 5 The structure block diagram of the first embodiment of the multi-Aelos robot cooperative sorting device for warehouse logistics is shown in the figure.

[0063] As Figure 5 shown, the multi-Aelos robot cooperative sorting device for warehouse logistics provided by the embodiment of the present application comprises: A data acquisition module 10 is configured to receive real-time order data, parse the real-time order data to generate a task set containing the location of goods, target sorting area and priority; A result allocation module 20 is configured to dynamically calculate the task matching degree in combination with the state information of each robot itself through a distributed task allocation mechanism, and use a reinforcement learning model to optimize the global task allocation to generate a global task allocation result; A space-time graph module 30 is configured to model the warehouse environment as a space-time graph structure based on the global allocation strategy; An optimization module 40 is configured to generate a conflict-free path using a pathfinding algorithm in combination with the space-time graph structure, and optimize the trajectory in combination with the task time limit constraint; An execution module 50 is configured to identify the goods through a vision system according to the target position in the conflict-free path, and use a force-controlled mechanical arm to complete the grabbing and placing.

[0064] It should be understood that the above is only an example, and the technical solutions of the present application are not limited in any way. In specific applications, those skilled in the art can make settings according to needs, and the present application does not limit this.

[0065] The embodiment receives real-time order data, parses the real-time order data to generate a task set containing cargo location, target sorting area and priority; through a distributed task allocation mechanism, dynamically calculates task matching degree in combination with self-state information of each robot, and adopts a reinforcement learning model to optimize global task allocation to generate a global task allocation result; models the warehouse environment as a space-time graph structure based on the global allocation strategy; adopts a pathfinding algorithm to generate a conflict-free path in combination with the space-time graph structure and optimizes a trajectory in combination with a task time limit constraint; identifies a cargo through a vision system according to a target location in the conflict-free path and completes grabbing and placing in combination with a force-controlled mechanical arm. The embodiment achieves the technical effects of improving sorting efficiency and adaptability and reducing operation and maintenance cost.

[0066] It should be noted that the above-described workflow is merely illustrative and does not limit the protection scope of the present application. In actual application, a person skilled in the art can select part or all of the above-described workflow to achieve the purpose of the embodiment according to actual needs, and the present application is not limited in this regard.

[0067] In addition, technical details not described in detail in the embodiment can be found in the method for multi-Aelos robot collaborative sorting in warehouse logistics provided by any embodiment of the present application, and will not be described here.

[0068] In addition, it should be noted that in this document, the terms "comprise", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or system. Without more limitations, the element defined by the statement "comprises a" does not exclude the presence of another identical element in the process, method, article or system that includes the element.

[0069] The above-mentioned sequence numbers of the embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0070] Those skilled in the art can clearly understand the above-mentioned embodiment method can be realized by means of software and the necessary general hardware platform, of course, also can be through hardware, but in many cases the former is the better embodiment. Based on such understanding, the technical solutions of the present application essentially or say the part of the contribution to the prior art can be embodied in the form of software products, the computer software product is stored in a storage medium (such as read-only memory (ReadOnly Memory, ROM) / RAM, disk, optical disc), including a number of instructions to make a terminal device (may be a mobile phone, computer, server, or network equipment, etc.) executes the method of each embodiment of the present application. The above is only the preferred embodiment of the present application, and does not limit the patent scope of the present application, any equivalent structure or equivalent process transformation made by using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A multi-Aelos robot collaborative sorting method for warehousing and logistics, characterized in that, include: Receive real-time order data and parse the real-time order data to generate a task set that includes the location of goods, target sorting area and priority. Through a distributed task allocation mechanism, the task matching degree is dynamically calculated by combining the state information of each robot, and a reinforcement learning model is used to optimize the global task allocation and generate the global task allocation result. Based on the global allocation strategy, the warehouse environment is modeled as a spatiotemporal graph structure; A pathfinding algorithm is used in conjunction with the aforementioned spatiotemporal graph structure to generate conflict-free paths, and the trajectory is optimized in conjunction with task time constraints. Based on the target location in the conflict-free path, the goods are identified by a vision system and then grasped and placed by a force-controlled robotic arm.

2. The method according to claim 1, characterized in that, The generated task set includes: Automatically identifies the weight, dimensions, and fragility characteristics of goods; Promotional orders are automatically marked as the highest priority and assigned to an emergency processing channel; The task queue is refreshed immediately when a new order arrives to ensure timeliness.

3. The method according to claim 1, characterized in that, Dynamically calculating task matching degree includes: A multi-dimensional status assessment is performed by integrating the robot's real-time location, remaining battery power, current load, and task urgency. Automatically adjust the weighting of the allocation strategy based on historical sorting data to achieve reinforcement learning optimization; Set up fast lanes for high-priority orders to skip the regular queuing process and enable competitive task allocation.

4. The method according to claim 1, characterized in that, Constructing the spacetime graph structure includes: Add timestamps to key locations for dynamic node marking; Automatically increase passage costs based on real-time robot density in the passageway; Reserve dedicated passage periods for emergency tasks to achieve time window constraint integration.

5. The method according to claim 1, characterized in that, Generating conflict-free paths includes: The hierarchical planning strategy uses the spatiotemporal A* algorithm to generate the main path at the global level and uses a potential field model to avoid dynamic obstacles in real time at the local level. Insert smooth curve segments into sharp turns to ensure stable robot steering and achieve smooth trajectory processing; Assign high-value order robots a direct fast lane to achieve priority passage.

6. The method according to claim 1, characterized in that, Grab and place operations include: Vision-force control dual closed loop: The vision system locates the coordinates of the goods at the millimeter level, and the force-controlled robotic arm automatically adjusts the gripping force according to the characteristics of the goods; Multi-robot collaborative collision avoidance: The central controller allocates staggered grasping time slots for adjacent robots; Special handling for fragile items: Automatically activates a buffer gripping mode for glass products, etc.

7. The method according to claim 1, characterized in that, The method also includes intelligent fault response: Multi-level early warning mechanism: A level three alarm is triggered when the battery level is below 20% or the temperature exceeds 70°C; Seamless task transfer: Automatically transfer tasks based on the location, idle status, and load capacity of nearby robots; Load balancing guarantee: Heavy tasks are intelligently split into multiple sub-tasks for distribution.

8. A multi-Aelos robot collaborative sorting device for warehousing and logistics, characterized in that, include: The data acquisition module is used to receive real-time order data and parse the real-time order data to generate a task set containing the location of goods, target sorting area and priority. The result allocation module is used to dynamically calculate the task matching degree by combining the state information of each robot with the distributed task allocation mechanism, and to optimize the global task allocation by using a reinforcement learning model to generate the global task allocation result. The spatiotemporal graph module is used to model the warehouse environment as a spatiotemporal graph structure based on a global allocation strategy; The optimization module is used to generate conflict-free paths by combining the pathfinding algorithm with the spatiotemporal graph structure and optimize the trajectory by combining the task time constraints. The execution module is used to identify the goods through a vision system based on the target position in the conflict-free path, and then combine it with a force-controlled robotic arm to complete the grasping and placement.

9. A computer device, characterized in that, The device includes a memory and a processor, wherein the processor, when executing computer instructions stored in the memory, performs the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Includes instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 7.

Images