Collaborative tray scheduling control method based on multiple AGVs

Abstract

The invention discloses a cooperative tray scheduling control method based on multiple AGVs (Automatic Guided Vehicles). According to the method, intelligent task release is realized by establishing a digital identifier for a tray and calculating a task priority; each AGV evaluates the ability of itself and obtains a task through distributed bidding; when the single AGV is insufficient in capability, dynamic alliance collaborative operation is formed; the alliance plans a path through a space-time diagram and eliminates conflicts; during execution, a hierarchical control architecture is adopted to keep synchronization; after the task is completed, performance is evaluated, and the alliance is disintegrated; and the system continuously monitors the health state of the AGV and realizes predictive maintenance. Through three innovations of tray intelligence, dynamic alliance and predictive maintenance, the limitation of traditional centralized scheduling is broken through, the system robustness, task adaptability and operation reliability are improved, and an efficient technical scheme is provided for intelligent logistics.

Description

Technical Field

[0001] This invention belongs to the field of intelligent logistics scheduling technology, specifically relating to a collaborative pallet scheduling and control method based on multiple AGVs. Background Technology

[0002] With the rapid development of intelligent manufacturing and smart logistics, Automated Guided Vehicles (AGVs) are widely used in warehousing, production workshops, and other scenarios. Existing AGV scheduling systems mainly adopt a centralized control architecture, with a central scheduling system uniformly allocating tasks and planning routes.

[0003] However, existing technologies have the following shortcomings: First, centralized scheduling has the risk of single point of failure, and failure of the central node will paralyze the entire system; Second, when faced with overweight or oversized pallets, a single AGV cannot complete the transportation task independently, and existing systems lack flexible coordination mechanisms; Third, traditional scheduling methods adopt a post-event maintenance strategy, which cannot prevent the impact of AGV failures on system operation; Fourth, pallets are only passive transportation objects and cannot participate in scheduling decisions, resulting in low efficiency in task allocation.

[0004] Therefore, there is an urgent need for a multi-AGV collaborative pallet scheduling and control method that can achieve autonomous AGV collaboration, support dynamic task allocation, and have predictive maintenance capabilities. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a collaborative pallet scheduling and control method based on multiple AGVs. Through pallet digital identification, a dynamic AGV alliance mechanism, and a predictive scheduling strategy, it achieves efficient collaborative operation of a multi-AGV system. The technical solution of this invention is as follows: A collaborative pallet scheduling and control method based on multiple AGVs includes the following steps: Step S1, Pallet Digital Identification and Task Issuance: Assign a unique 96-bit digital identifier to each pallet and establish a data structure containing static and dynamic attributes; calculate task priority based on time urgency, cargo value, customer importance, temperature control requirements, and queue status; and issue transportation tasks through an event-driven mechanism.

[0006] Step S2, AGV Capability Assessment and Task Matching: Each AGV's transportation capacity is assessed from five dimensions: energy, load, health, location, and experience; the comprehensive matching degree of computing power matching, distance factors, collaborative potential, and historical performance is calculated; tasks are allocated through a distributed bidding mechanism.

[0007] Step S3, Dynamic Alliance Determination and Formation: Determine the required number of AGVs based on weight, size, stability, and turning requirements; the main AGV broadcasts a collaboration request, including task information and incentive mechanisms; the responding AGVs assess feasibility and benefits; the Hungarian algorithm is used to optimize member allocation; clock synchronization and role assignment are performed.

[0008] Step S4, Collaborative Path Planning and Conflict Resolution: Construct a spatiotemporal graph model to optimize the path by comprehensively considering distance, time, energy consumption, safety, and comfort; adopt a spatiotemporal reservation mechanism to avoid conflicts; and resolve path conflicts based on dynamic priority negotiation.

[0009] Step S5, Synchronous Cooperative Execution Control: Maintain formation using a virtual structure method; achieve synchronous control through a consensus protocol; dynamically allocate pallet loads; monitor the safety envelope and execute emergency braking.

[0010] Step S6, Task Completion and Alliance Dissolution: Multi-sensor fusion achieves precise positioning; position, attitude, velocity, stability, and integrity conditions are met; performance indicators are calculated; the alliance is dissolved according to the protocol.

[0011] Step S7, Predictive Maintenance and Scheduling Optimization: Collect multi-source sensor data and extract time-frequency domain features; establish a hierarchical health assessment system; predict remaining lifetime and failure probability; dynamically adjust scheduling weights; optimize maintenance strategies and timing.

[0012] Beneficial effects: 1. Enhanced system robustness: Through a distributed decision-making architecture, the single point of failure problem of centralized scheduling is avoided. Even if some AGVs or communication nodes fail, the system can still operate in a degraded manner.

[0013] 2. Enhanced Task Adaptability: The dynamic alliance mechanism enables the system to flexibly respond to the transportation needs of pallets of different sizes, especially the collaborative handling of overweight and oversized pallets, thus expanding the system's task processing capabilities.

[0014] 3. Improved operational reliability: Predictive maintenance strategies can identify potential AGV failures in advance and proactively schedule maintenance, avoiding sudden failures of AGVs when performing critical tasks and improving the overall reliability of the system.

[0015] 4. Optimized scheduling efficiency: The digitalization of pallets makes them proactive task initiators. Combined with the AGV distributed bidding mechanism, it reduces scheduling decision time and improves task allocation efficiency.

[0016] 5. Excellent scalability: The distributed architecture gives the system excellent scalability. New AGVs can participate in scheduling simply by registering, without needing to modify the central control logic. Attached Figure Description

[0017] Figure 1A flowchart illustrating the steps of the method described in this invention is shown. Detailed Implementation

[0018] Exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] Combination Figure 1 This invention provides a collaborative pallet scheduling and control method based on multiple AGVs, comprising the following steps: Step S1: Pallet Digital Identification Establishment and Task Issuance In this step, the system establishes a unique digital identifier for each pallet entering the scheduling area, and uses this identifier to achieve intelligent task assignment and tracking. When a pallet enters the system's scheduling area through the inlet, an RFID reader array deployed at the inlet automatically reads the passive RFID tag on the bottom of the pallet. For pallets without RFID tags, the system uses an industrial camera to identify the QR code or barcode on the pallet surface, ensuring that all pallets can be identified and tracked by the system. The system uses a 96-bit encoding structure to generate a pallet digital identification code, which is calculated using the following formula: in: For pallet digital identification code; It is a 96-bit base code used to distinguish different warehouse systems; This indicates a bitwise XOR operation; This is the facility code, which occupies 16 digits and identifies the specific warehouse number and area. The pallet type code occupies 16 bits, of which the first 8 bits represent the material type and the last 8 bits represent the specification level; This is a timestamp, occupying 32 bits, using the Unix timestamp format, accurate to milliseconds; This is the serial number, occupying 16 bits, and increments daily starting from 0; It is a 16-bit cyclic redundancy check code.

[0020] Based on the aforementioned identification code, the system establishes a complete attribute data structure for each pallet. Static attributes are entered into the system when the pallet first enters the warehouse, including the pallet's physical dimensions, load-bearing capacity, and material properties. Specifically, the static attribute vector is defined as follows: ;in, The length of the pallet is in meters. Pallet width, in meters; The height of the pallet is in meters. Maximum load capacity, in kilograms; The coefficient of static friction; The coefficient of kinetic friction; The elastic modulus of the material is expressed in Pascals.

[0021] At the same time, the system updates the dynamic attributes of the tray in real time, which are continuously monitored and updated through a sensor network: ;in, The actual load on the pallet at time t is obtained by measuring the pressure sensor. , , These are the three-dimensional coordinates of the center of gravity, calculated through multi-point pressure distribution; Surface temperature, used for monitoring cold chain logistics; Relative humidity; Vibration intensity is used for monitoring the transportation of fragile goods.

[0022] Pallet task generation employs an event-driven mechanism, with the system recognizing four types of triggering events: inbound events, outbound events, location adjustment events, and emergency transfer events. When any of these triggering events is detected, the system automatically generates the corresponding transportation task. To ensure the rationality of task scheduling, the system uses a multi-factor weighted method to calculate task priority. ;in, The weight coefficient of the i-th factor satisfies the normalization condition. ; Let be the i-th influencing factor. Specifically, the time pressure factor. The time pressure of the task is reflected by the sigmoid function. ,in: This is a time-sensitivity coefficient, set according to warehouse operation requirements, with a typical value of 0.001. The deadline for the task; The current system time; To estimate the task execution time.

[0023] Value Factor Reflecting the economic value of goods: ;in, The current value of the goods; As a benchmark for reference value, the average value of goods in the warehouse is usually taken.

[0024] Customer Factors This demonstrates the importance of the customer: ;in, The customer base importance score is assigned, with a value range of [0,1]. This is used to identify VIP customers; the value is 0 for regular customers and 1 for VIP customers.

[0025] Temperature control factor For special goods requiring temperature control: , Other situations ;in, and These are the upper and lower limits for temperature control, respectively.

[0026] queue factor Consider the current congestion status of the task queue: ;in, This is the queue length decay coefficient, typically 0.1; This represents the current task queue length.

[0027] Task information is broadcast to the AGVs within the system in JSON format via a dedicated 5.8GHz wireless network, ensuring low latency and high reliability of information transmission.

[0028] Step S2: AGV Capability Assessment and Task Matching In this step, each AGV assesses its capabilities based on its own status and achieves optimal matching with the task through a distributed bidding mechanism.

[0029] First, the AGV calculates its overall transportation capacity through multi-dimensional evaluation. The capacity assessment model comprehensively considers five dimensions: energy state, load capacity, health status, location advantage, and historical experience, constructing a capacity vector: ; Energy Capacity It considers not only the current battery level, but also the degree of battery aging and charging convenience: in: Remaining electricity, in watt-hours; Nominal electricity volume, unit: watt-hour; This is the battery aging factor, reflecting the degradation of battery performance, with a typical value of 0.2; This represents the current charge / discharge cycle count. The number of cycles is the rated number. The charging convenience factor is typically 0.3. The distance to the nearest charging station; The normalized distance constant is set to 50 meters.

[0030] Load capacity Taking into account the fatigue effects caused by the current load status and continuous working time: in: Current load, in kilograms; Maximum load capacity, in kilograms; This is the load sensitivity index, with a typical value of 1.5; This is the fatigue effect coefficient, with a typical value of 0.3; Continuous working time, in hours; The maximum permissible continuous working time is typically 8 hours.

[0031] Health Capacity By fusing data from multiple sensors, various health indicators are synthesized using a product approach: ,in, The number of monitoring indicators; Let i be the i-th health indicator, with a value range of [0,1]. Let be the weight of the i-th indicator.

[0032] Each health indicator Calculation based on the deviation between sensor measurements and standard values: ,in, This represents the measured value of the i-th sensor; The ideal value for the i-th index; Let be the standard deviation of the i-th indicator.

[0033] Positioning ability This reflects how advantageous the AGV's current position is for performing the task. in: This is the distance attenuation coefficient, typically 0.02; This represents the path distance to the starting point of the task. The number of obstacles on the path; This represents the maximum acceptable number of obstacles, typically 10.

[0034] experience and ability Historical task execution performance based on AGVs: in: Based on the success rate; This is the success reward coefficient, with a typical value of 0.2. The number of similar tasks that were successfully completed; This represents the total number of tasks. This is the failure penalty coefficient, typically 0.5. This represents the number of failed tasks.

[0035] The overall ability value is obtained by weighted summation: The weight vector is dynamically adjusted according to the task type to ensure that the capability assessment matches the task requirements.

[0036] Next, the system calculates the matching degree between each AGV and the task. The matching degree comprehensively considers four aspects: capability matching, distance factor, collaborative potential, and historical performance. The total matching degree is calculated by multiplying the matching degrees of each component: ; Ability matching Ensure that the AGV's capabilities meet the task requirements: in: Let be the overall capability value of the j-th AGV; Let i represent the capability requirements for the i-th task.

[0037] Distance matching degree Reflecting the spatial proximity of the AGV to the task: in: Let be the distance from the j-th AGV to the i-th task starting point; This is the normalized distance constant.

[0038] Collaborative matching degree Consider the possibility of collaboration between neighboring AGVs: in: Let be the set of neighbors of AGV j; Let k be the coordination coefficient of neighbor k; Let j be the distance between AGV j and k. The neighborhood radius is typically 20 meters.

[0039] Historical matching degree Historical success rate of similar tasks performed using AGVs: in: The historical success rate of AGV j performing task type i; The average success rate of all AGVs; The standard deviation of the success rate; It is the hyperbolic tangent function.

[0040] Task allocation employs a distributed bidding mechanism, where each AGV submits a bid based on its matching degree, and the system selects the AGV with the highest matching degree to execute the task. This mechanism ensures optimality and fairness in task allocation.

[0041] Step S3: Dynamic Alliance Determination and Formation After gaining the right to execute a task, the AGV needs to determine whether it can complete the pallet transport task independently. If the capacity of a single AGV is insufficient, a dynamic alliance needs to be formed for collaborative operation.

[0042] The alliance's requirements are determined based on a comprehensive evaluation across multiple dimensions. First, the system calculates the minimum number of AGVs required to complete the task: in: This represents the total number of AGVs required. The number of AGVs calculated based on weight requirements; The number of AGVs calculated based on size requirements; The number of AGVs calculated based on stability requirements; The number of AGVs calculated based on turning requirements.

[0043] The weight requirement calculation takes into account the total weight of the pallet and goods, as well as the safety factor: in: The weight of the goods is expressed in kilograms. The weight of the pallet is in kilograms. This is a weight safety factor to ensure safety during transportation; a typical value is 0.75. The maximum carrying capacity of a single AGV, in kilograms; The weight of the AGV is in kilograms. This represents the function for rounding up.

[0044] Size requirements are based on the pallet geometry and the AGV's gripping capacity: in: The length of the pallet is in meters. Pallet width, in meters; This is a dimensional safety factor, typically 0.85. The effective clamping length of a single AGV, in meters.

[0045] Stability requirements ensure the dynamic balance of the pallet during transportation: in: The total moment of inertia of the pallet and goods is expressed in kilograms per square meter. Maximum permissible angular velocity, in radians per second; The controllable rotational inertia of a single AGV, expressed in kilograms per square meter. The maximum output torque of a single AGV is expressed in Newton-meters (N·m).

[0046] Turning requirements ensure that large pallets can pass smoothly through narrow aisles: in: Minimum turning radius of the warehouse aisle, in meters; The maximum turning angle of a single AGV, in radians.

[0047] When calculated At this time, the main AGV initiates the alliance formation procedure. The main AGV constructs and broadcasts a collaboration request message, which includes three parts: task details, collaboration requirements, and incentive mechanism.

[0048] The incentive mechanism for collaboration request messages is designed as follows: in: Basic task rewards; This is the urgency-based reward coefficient, with a typical value of 0.3. Prioritize tasks; This is the distance reward coefficient, with a typical value of 0.2; For the estimated transport route length; This is the reference path length.

[0049] Upon receiving a collaboration request, the AGV makes a response decision by first assessing the feasibility of its participation. This feasibility assessment is derived through a logical AND operation on four conditions: Among them, ability conditions Ensure that the AGV has basic collaborative capabilities: , =1; otherwise, ;in, This is the capability threshold, typically 0.3. Distance condition. Limit the response range to avoid excessively long assembly times: , In other cases, ;in, The distance to the location where the request was initiated; This represents the maximum response distance, typically 30 meters. Time conditions. Ensure that AGVs can participate in collaboration in a timely manner: when , In other cases, ;in, Estimate the available time for the AGV; The deadline for the task; For safety margin, a typical value is 300 seconds. Compatibility requirements. Ensure the AGV can execute the required formation pattern: , In other cases, .

[0050] For AGVs that meet the feasibility criteria, the expected benefits of participating in collaboration are further calculated: ;in, This represents the probability of task success. Basic rewards for the task; Additional rewards for collaboration; The cost of participating in collaboration.

[0051] The probability of success is based on a comprehensive assessment of historical data and the current situation: ;in, The basic success probability is derived from historical statistics; Let be the probability of the i-th risk factor occurring; This represents the total number of risk factors.

[0052] The cost of participation includes three parts: energy cost, time cost, and opportunity cost. ;in, This is the energy cost factor, expressed as a bonus per watt-hour. For additional energy consumption; This is a time cost coefficient, expressed in rewards per second. This is for additional time consumption; Opportunity cost coefficient; The opportunity cost of giving up other tasks.

[0053] After the main AGV collects all responses, an improved Hungarian algorithm is used for optimal matching. A revenue matrix is ​​then constructed. , of which elements Let represent the reward of the i-th responding AGV occupying the j-th position. The optimal allocation is obtained by solving the following optimization problem, as shown in the formula: The constraints are: in, For binary decision variables, a value of 1 indicates that the i-th AGV is assigned to the j-th position, and a value of 0 indicates that it is not assigned.

[0054] After the alliance is formed, each member AGV executes the initialization protocol. First, clock synchronization is performed using the IEEE 1588 Precision Time Protocol (PTP) to achieve microsecond-level synchronization accuracy. Then, a dedicated communication link is established, an independent communication channel is allocated, and a 128-bit AES encryption key is set. Next, the relative positions of each AGV are determined, and the target position for each AGV is calculated based on the pallet size and the selected formation. Finally, control roles are assigned: the master AGV is responsible for navigation decisions and path planning, while the slave AGVs are responsible for collaborative execution and load sharing.

[0055] Step S4: Collaborative Path Planning and Conflict Resolution After the alliance is formed, the main AGV is responsible for planning the collaborative transportation path. The path planning adopts a hierarchical strategy, with the global layer determining the sequence of key nodes and the local layer handling real-time obstacle avoidance and path optimization.

[0056] First, the system constructs an enhanced spatiotemporal graph model to unify the representation of physical space and time dimensions: ;in, A set of nodes; Let it be the set of edges; The time dimension; A collection of obstacles; This is the weighting function. Each node in the node set contains location, orientation, and floor information: ;in, Let be the planar coordinates of node i; Orientation angle; This serves as the floor identifier. Each edge in the edge set contains rich attribute information: ;in, Let $\frac{i}{j}$ be the cost of traveling from node $i$ to node $j$. For the time it takes; This refers to the path width. This represents the risk coefficient.

[0057] The objective function for path optimization comprehensively considers the costs of five aspects: in: For the weighting coefficients, satisfying ; These are the normalized distance cost, time cost, energy cost, safety cost, and comfort cost, respectively.

[0058] Distance cost is calculated by summing the path lengths: Normalized distance cost:

[0059] in: This represents the number of path segments; The coordinates of the k-th path point; The straight-line distance from the starting point to the ending point; This represents the maximum allowed path length.

[0060] Time cost is calculated via path and time:

[0061] Normalization time cost:

[0062] in: The length of the k-th segment; The velocity of segment k; Waiting time; This is the theoretically shortest time. This is the deadline for the task.

[0063] Energy consumption costs consider base power consumption, motion power consumption, and load power consumption: Normalized energy consumption cost:

[0064] in: Base power; The power coefficient of motion; This is the load power factor; For time step; Minimum energy consumption; Maximum energy consumption; This is the maximum load capacity.

[0065] Safety costs are calculated based on the distance to obstacles:

[0066] The cost of normalization security:

[0067] in: The danger factor of obstacle o; Let be the distance from the k-th step to the obstacle o; To prevent division by zero, a small constant is typically 0.01; The upper limit is set at the cost of safety; This is the safe distance threshold.

[0068] The trade-off for comfort is to reduce sharp turns and sudden acceleration / deceleration:

[0069] Normalized comfort cost:

[0070] in: Let be the acceleration of the k-th segment; Let be the path curvature of the k-th segment; The upper limit is achieved by sacrificing comfort. The maximum permissible acceleration; This represents the maximum permissible curvature.

[0071] During route planning, the system needs to detect and handle potential path conflicts in real time. Conflict detection employs a prediction window mechanism. in: Let i be the position of entity i at time t; For a safe distance, a typical value is 1.5 meters.

[0072] When a conflict is detected, the system employs a multi-strategy negotiation mechanism to resolve it. First, the dynamic priorities of the conflicting parties are compared: in, It is a static priority; This represents the time urgency factor, with a typical value of 0.01. The deadline for the task; The current time; This is the waiting penalty coefficient, typically 0.1; The system accumulates the number of waiting attempts. Based on the priority comparison results, the system selects an appropriate conflict resolution strategy.

[0073] If there is a significant difference in priority, the execution speed of the lower priority item will be adjusted accordingly. in, This is the original speed; This is the velocity attenuation coefficient, typically 0.2; Number of conflicts.

[0074] If priorities are similar, avoidance can be achieved through path offset: in, This is the original path; This is the offset distance; This is a unit vector perpendicular to the original path. Through this mechanism, the system can effectively plan the cooperative path of the alliance and handle dynamic conflicts, ensuring the efficiency and safety of pallet transportation.

[0075] Step S5: Synchronous and Cooperative Execution Control After path planning is completed, the alliance enters the collaborative execution phase. In this phase, each AGV needs to maintain a precise relative position, coordinate its movement, and work together to smoothly transport the pallet. Collaborative control is implemented using a three-layer architecture. The top layer is the formation control layer, responsible for maintaining the overall formation structure; the middle layer is the synchronization control layer, ensuring the consistency of movement among the AGVs; and the bottom layer is the execution control layer, implementing specific motion control commands.

[0076] At the formation control layer, the system employs a virtual structure approach, treating the alliance as a virtual rigid body. The motion of the virtual rigid body is determined by the master AGV according to the planned path, while each slave AGV maintains its relative position to the virtual rigid body. The state of the virtual frame is represented as follows: ; in, The center position of the virtual frame; Here is the frame orientation rotation matrix; Let i be the relative position vector of the i-th AGV within the frame; This represents the total number of AGVs in the alliance. Based on the virtual framework, the desired position of each AGV is obtained through coordinate transformation: ; in, Let be the desired position of the i-th AGV. Correspondingly, the desired speed is obtained by differentiating the position and considering the rotation effect: ; in, The linear velocity at the center of the frame; The angular velocity of the frame; This represents the vector cross product operation. In the synchronization control layer, the system employs a distributed consensus protocol to ensure coordinated movement of all AGVs. Each AGV calculates the synchronization error by exchanging status information with its neighboring AGVs. ;in, For AGV i, it is the set of its communication neighbors; These are the adjacency matrix elements of the communication topology; The actual position of AGV i; Let AGVi be the desired position. Based on the synchronization error, generate the synchronization control law: ;in, This is the synchronous proportional gain matrix; The synchronous differential gain matrix; This represents the time derivative of the synchronization error. To ensure communication connectivity, the communication topology is dynamically adjusted based on the distance between AGVs. , ; , ; in, This is the communication attenuation factor, typically 0.5; Let i be the distance between AGV i and j; For effective communication range, a typical value is 10 meters. At the execution control layer, each AGV generates specific wheel speed control signals based on the upper-level control commands. For differential drive AGVs, the left and right wheel speeds are calculated using inverse kinematics: ;in, and These are the angular velocities of the right and left wheels, respectively. This refers to the linear velocity of the AGV. The angular velocity of the AGV; This refers to the distance between the left and right wheels; Let be the wheel radius. To ensure robust control, a sliding mode controller is used to achieve wheel speed tracking. Define the sliding surface: in: and These are the expected angular velocities of the right and left wheels, respectively. and For integral coefficients. The control law uses a saturation function to avoid control chattering: in: and These are the control torques for the right and left wheels, respectively. and For sliding mode gain; and This is the differential gain; It is a saturation function; and This represents the boundary layer thickness.

[0077] During collaborative transportation, the pallet load needs to be rationally distributed among the AGVs. Load distribution considers not only static weight but also dynamic effects. First, the pallet's tilt state is estimated using multiple pressure sensors: in: For the pressure sensor readings at the four corner points; and These are the longitudinal and lateral spacings of the sensor, respectively. and These are the longitudinal and lateral tilt angles, respectively. Based on the tilt and motion states, calculate the load that each AGV should bear: in: Let be the vertical load of the i-th AGV; The total weight of the pallet and the goods; It is the acceleration due to gravity; This is a horizontal compensation item; This is the acceleration compensation term.

[0078] The horizontal compensation term is used to balance the uneven load caused by tray tilt: in: This is the horizontal compensation coefficient, with a typical value of 0.1; Let the height be the height of the i-th support point; This represents the average height of all support points. The acceleration compensation term considers dynamic load transfer. in: For load mass; The center of gravity acceleration vector; Let be the normal unit vector of the i-th support point. To ensure transportation safety, the system continuously monitors synchronization error and safety envelope. When the synchronization error exceeds a threshold, the operating speed is automatically reduced. in, The adjusted speed; Nominal speed; This is the error impact coefficient, with a typical value of 0.5; This represents the current synchronization error. This is the error threshold, typically 0.02 meters.

[0079] Meanwhile, the system defines a dynamic safety envelope, the radius of which varies with velocity and acceleration: ;in, For the safety envelope radius; The static safety radius is typically 1 meter. This is the speed influence coefficient, with a typical value of 0.1; This is the acceleration influence coefficient, with a typical value of 0.05; Current speed; This represents the current acceleration. The system will immediately initiate emergency braking if any of the following emergency conditions are detected: an obstacle appears within the safety envelope, the synchronization error exceeds the maximum allowable value, communication interruption exceeds the timeout threshold, sensor readings are abnormal, or a manual emergency stop signal is received. During emergency braking, all AGVs simultaneously apply maximum reverse acceleration to ensure the stability of the pallet.

[0080] Step S6: Mission Completed and Alliance Disbanded As the system approaches the target location, it enters the precise positioning phase to ensure the tray is accurately placed in the designated position. Precise positioning employs multi-sensor fusion technology, integrating measurement results from encoders, LiDAR, UWB, and visual positioning. The fusion algorithm is based on an extended Kalman filter, and the state vector is defined as: in, These are the position coordinates; Orientation angle; Linear velocity; Angular velocity; This represents the systematic deviations in position and angle.

[0081] The prediction model is built based on the kinematic equations: in, It is a nonlinear state transition function; For control input; The noise is assumed to follow a zero-mean Gaussian distribution.

[0082] The observation model integrates data from multiple sensors: in, It is a nonlinear observation function; To observe noise.

[0083] The optimal state estimate is obtained through the prediction-update loop of Kalman filtering: in, The Kalman gain is calculated by minimizing the estimation error covariance. Task completion requires the simultaneous fulfillment of five conditions. The position accuracy condition requires that the deviation between the final position and the target position be less than an acceptable value. ;in, For the final position; For the target location; For positional tolerance, the typical value is 0.03 meters.

[0084] Posture accuracy conditions ensure the tray is oriented correctly: ;in, For the final orientation; Oriented towards the target; This is for angular tolerance, typically 0.05 radians.

[0085] The speed convergence condition ensures that the tray has come to a complete stop: ;in, The final linear velocity; The final angular velocity; meters per second; Radius per second. The stability condition requires the tray to remain stable at the target position for a certain period of time. ;in, To stabilize the observation time, a typical value is 2 seconds; This is the stability threshold, typically 0.02 meters. Integrity conditions are verified by checking and confirming the pallet is undamaged: ;in, This is a cyclic redundancy check function; and These represent the final and initial states of the tray, respectively.

[0086] Once all conditions are met, the system executes a standardized unloading procedure. First, the main AGV sends an unloading preparation signal to all members, and each AGV confirms its readiness. Then, the consortium synchronously lowers the pallet height to the placement plane. Next, each AGV simultaneously releases its gripping mechanism to ensure the pallet lands smoothly. Subsequently, all AGVs synchronously retreat to a safe distance, typically 2 meters. Finally, sensors confirm that the pallet is stably placed without tilting or sliding. After the task is completed, the system calculates and records several performance indicators. Path efficiency reflects how close the actual path is to the optimal path: ;in, This is the theoretical shortest path length; This represents the actual travel distance. Time efficiency is used to assess the timeliness of task completion. ;in, The planned completion time; This refers to the actual completion time. Deadline; This is a delay penalty coefficient. Energy efficiency reflects the economics of energy use. ;in, This represents the theoretical minimum energy consumption. This refers to the actual energy consumed. Collaboration efficiency assesses the level of cooperation among alliance members: ;in, Let be the standard deviation of the load of the i-th AGV; This represents the average load. The overall performance is calculated using the weighted geometric mean of all indicators: ;in, Let be the weight of the j-th indicator. Alliance dissolution is executed according to a predetermined protocol. The main AGV generates a dissolution message, including task completion status, performance indicators, resource release list, and subsequent arrangements. Upon receiving the message, each member AGV saves its task log, updates its own status parameters, releases the occupied communication channels and path resources, and then decides whether to enter standby mode or execute the next task based on the task queue status.

[0087] Step S7: Predictive Maintenance and Scheduling Optimization The system continuously monitors the operating status of AGVs and establishes a predictive maintenance mechanism, thereby preventing failures before they occur and significantly improving the long-term reliability of the system. The foundation of predictive maintenance is a comprehensive data acquisition system. Each AGV is equipped with multiple sensors to collect operational data in real time. Vibration sensors acquire triaxial acceleration data at a sampling frequency of 1000Hz to detect abnormal vibrations in mechanical components. Temperature sensors monitor the temperature of the motor, battery, controller, and ambient temperature at a sampling frequency of 10Hz. Current sensors record current changes in the left and right drive motors, battery, and system bus to assess the health of the electrical system. In addition, encoders provide precise position and speed information, pressure sensors monitor load distribution, and the communication module records network quality indicators. Based on the collected raw data, the system extracts multi-dimensional features for health status assessment. Time-domain features reflect the statistical characteristics of the signal, including root mean square value, peak factor, kurtosis, skewness, and standard deviation. Frequency-domain features are obtained through Fast Fourier Transform (FFT) to identify periodic failure modes. in: Represented in the frequency domain; The unit is the imaginary unit. Frequency domain energy is calculated based on the spectrum. Spectral entropy reflects the complexity of a signal: in: The power spectral density is normalized. The system employs a hierarchical health assessment system, evaluating health layer by layer from the component level to the system level. The component-level health index is calculated through feature-weighted fusion: ;in, For component health index; The number of features; The weight of the i-th feature; Here is the health mapping function. The health mapping function takes the form of a sigmoid function, mapping eigenvalues ​​to the interval [0,1]. ;in, This is a sensitivity parameter that controls the steepness of the mapping curve; This is a threshold parameter, corresponding to the dividing point of health status.

[0088] The system-level health index comprehensively considers the health status of each key component: in: , , These are the health indices for the motor, battery, and controller, respectively. Let j be the importance coefficient of the j-th component; This represents the total number of components. Remaining useful life (RUL) prediction employs a deep learning-based method. The system constructs a time-series feature matrix containing all features within the historical time window: ;in, This refers to the length of the time window, typically 100 sampling periods. Let be the feature vector at time t. Based on historical data and current features, the prediction model outputs the probability distribution of RUL: ;in, The predicted average remaining useful life; This represents the uncertainty in the prediction. The failure probability is calculated using the cumulative distribution function: ;in, The cumulative distribution function of the standard normal distribution; To predict the time span, the system dynamically adjusts the scheduling weights of the AGVs based on their health status and failure probability. The scheduling weights consider multiple factors: ;in, For scheduling weights; Basic weights; Availability function; For quality factor; This is the load balancing factor. The availability function reflects the reliability of the AGV: ;in, This represents the probability of failure. For reliability; For service quality indicators.

[0089] Reliability is based on the Weibull distribution model: ;in, Here, is the scale parameter, representing the feature lifetime; The shape parameter reflects the trend of failure rate changes; For equipment startup time. Quality factor assessment of AGV task execution accuracy: in: This is the quality sensitivity coefficient; Let be the error of the j-th performance index; Tolerance; This refers to the number of performance metrics. Load balancing factors promote the even distribution of tasks: in: This is the equilibrium coefficient; For AGV utilization rate; Average utilization rate; The standard deviation of utilization is used. Maintenance decisions are made using a Markov decision process framework, modeling the maintenance problem as a sequential decision problem. The state space includes five dimensions: health state, reliability, workload, task type, and runtime.

[0090] in: In good health; For reliability; For workload levels; To determine whether or not a critical task is to be performed; This represents the cumulative running time.

[0091] Action space includes five maintenance strategies: State transition probabilities reflect the impact of different actions on the system state: in: The next state; This is the Dirac function.

[0092] The transition of health status follows a degradation-recovery model: , ; , ; , ;in, This is the normal amount of degradation. To maintain the recovery volume; The health status after replacement; This represents a normal distribution.

[0093] The reward function balances operational benefits and maintenance costs: in: For operating revenue; Cost of action; For failure costs; To cover the costs of delays; This is a critical task indicator function.

[0094] Operational costs are determined based on the type of maintenance: , ; , ; , ; , ; , ;in, , , , These are the direct costs of inspection, minor repair, major repair, and replacement, respectively. Downtime cost rate; , , This refers to the corresponding maintenance time.

[0095] The optimal maintenance strategy is solved using a value iteration algorithm: ;in, It is the optimal value function; This is the discount factor, with a typical value of 0.95.

[0096] The optimal strategy is:

[0097] Maintenance scheduling and task scheduling are integrated and optimized to ensure that maintenance activities have minimal impact on normal operations. The system identifies low-load periods as maintenance windows. in: Let be the system load at time t; This is the load threshold; This is a collection of AGVs to be maintained. For the maintenance requirements of AGV i at time t; To maintain the upper limit of the capacity.

[0098] Maintenance priorities are sorted by urgency: in: As a health index; This represents the probability of failure. For criticality; Service life; These are the weighting coefficients.

[0099] Batch maintenance is optimized using integer programming: The constraints include: each AGV must be maintained once, the maintenance capacity cannot exceed the upper limit, and the decision variables are integers between 0 and 1. Among these: Indicates whether to maintain AGV i at time t; Indicates whether maintenance is scheduled at time t; To maintain costs; To prepare costs.

[0100] Through a seven-step iterative process, this invention constructs a complete multi-AGV collaborative pallet scheduling and control system, achieving dual optimization of efficient single-task execution and long-term stable system operation. The distributed architecture ensures scalability and robustness, the dynamic alliance provides task processing flexibility, and predictive maintenance guarantees long-term reliability, offering an innovative technical solution for intelligent warehousing and smart logistics.

[0101] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A collaborative pallet scheduling and control method based on multiple AGVs, characterized in that, Includes the following steps: S1. Pallet Digital Identification and Task Issuance: Assign a unique 96-bit digital identification code to each pallet entering the scheduling area, and establish a pallet data structure containing static and dynamic attributes; calculate task priority through weighted summation of time urgency factor, value factor, customer factor, temperature control factor, and queue factor; generate transportation tasks based on four types of triggering events: inbound, outbound, location adjustment, and emergency transfer. S2. AGV Capability Assessment and Task Matching: Each AGV calculates its comprehensive transportation capacity value by weighted summation of five dimensions: energy capacity, load capacity, health capacity, position capacity, and experience capacity; the task matching degree is obtained by multiplying the capability matching degree, distance matching degree, coordination matching degree, and historical matching degree; a distributed bidding mechanism is adopted to select the AGV with the highest matching degree to execute the task. S3. Dynamic Alliance Determination and Formation: The required number of AGVs is calculated based on the four dimensions of weight, size, stability, and turning, and the maximum value is taken as the alliance size; the main AGV broadcasts a collaboration request containing task information, collaboration requirements, and incentive mechanisms; the responding AGVs decide whether to participate through feasibility assessment and benefit calculation; the optimal member allocation is determined using an improved Hungarian algorithm. S4. Collaborative Path Planning and Conflict Resolution: Construct a spatiotemporal graph model including nodes, edges, time, obstacles, and weights; optimize the path by weighted summation of five costs: distance, time, energy consumption, safety, and comfort; reserve the path occupancy time using a spatiotemporal reservation mechanism; and resolve path conflicts based on dynamic priority and multi-strategy negotiation. S5. Synchronous Cooperative Execution Control: Define the virtual rigid body frame of the alliance through the virtual structure method; calculate the synchronization error based on the exchange of neighbor state information and generate a consistency control law; dynamically allocate the load-bearing capacity of each AGV according to the pallet tilt and acceleration state; monitor the safety envelope in real time and execute emergency braking when an anomaly occurs; S6. Task Completion and Alliance Dissolution: Multi-sensor positioning data is fused through extended Kalman filtering; task completion is determined by simultaneously satisfying five conditions: position, attitude, speed, stability, and integrity; four efficiency indicators—path, time, energy, and cooperation—are calculated; resources are released and AGV status is updated according to the dissolution protocol; S7. Predictive maintenance and scheduling optimization: Collect data from multiple sources of vibration, temperature, and current sensors and extract time-frequency domain features; Establish a hierarchical health assessment system from the component level to the system level; predict remaining service life and failure probability based on historical data and current characteristics; dynamically adjust scheduling weights according to health status, failure probability, and load balancing. Optimize maintenance timing and strategies through Markov decision processes.

2. The collaborative pallet scheduling and control method based on multiple AGVs according to claim 1, characterized in that, The digital identification code in step S1 is generated by XORing the base code with the facility code, pallet type, timestamp, serial number and check code. The base code is used to distinguish different warehouse systems, the timestamp adopts the Unix time format accurate to milliseconds, and the check code adopts a sixteen-bit cyclic redundancy check.

3. The collaborative pallet scheduling and control method based on multiple AGVs according to claim 1, characterized in that, The health capability in step S2 is calculated by a weighted product of multiple health indicators. Each health indicator is mapped to the zero-to-one interval based on the deviation between the sensor's measured value and the ideal value through a Gaussian function.

4. The collaborative pallet scheduling and control method based on multiple AGVs according to claim 1, characterized in that, The alliance formation in step S3 adopts a three-phase consensus protocol: the pre-preparation phase is proposed by the master AGV broadcasting; the preparation phase is when each AGV exchanges preparation messages; and the submission phase is when the alliance is formally established after receiving confirmation from more than two-thirds of the participants.

5. The collaborative pallet scheduling and control method based on multiple AGVs according to claim 1, characterized in that, In step S4, the dynamic priority is equal to the static priority multiplied by an exponential function of time urgency, and then multiplied by a penalty term that includes the number of waiting times. Time urgency is calculated by the difference between the deadline and the current time.

6. The collaborative pallet scheduling and control method based on multiple AGVs according to claim 1, characterized in that, The synchronization error in step S5 is a weighted sum of the positional deviations between each AGV and its communication neighbors. The weights are determined by the adjacency matrix of the communication topology, and the elements of the adjacency matrix are in the form of a distance-based sigmoid function.

7. The collaborative pallet scheduling and control method based on multiple AGVs according to claim 1, characterized in that, The comprehensive performance index in step S6 is calculated by weighted geometric average of path efficiency, time efficiency, energy efficiency and collaboration efficiency. Path efficiency is the ratio of the theoretical shortest path to the actual path, time efficiency takes into account delay penalty, and energy efficiency is the ratio of the theoretical minimum energy consumption to the actual energy consumption.

8. The collaborative pallet scheduling and control method based on multiple AGVs according to claim 1, characterized in that, The failure probability in step S7 is calculated using the cumulative distribution function of the standard normal distribution, with the input being the difference between the prediction time span and the remaining useful life divided by the prediction uncertainty.

9. The collaborative pallet scheduling and control method based on multiple AGVs according to any one of claims 1 to 8, characterized in that, The maintenance priority in step S7 is calculated by weighted summation of health status, failure probability, criticality, and service life. The maintenance timing is selected when the system load is below the threshold and the maintenance capability meets the requirements.

Images