Single-robot intelligent scheduling method and system based on dynamic priority learning

Abstract

The application discloses a single-robot conveying intelligent scheduling method and system based on dynamic priority learning, and relates to the technical field of intelligent control in a conveying process.The single-robot conveying intelligent scheduling method and system based on dynamic priority learning comprises the following steps: S1, collecting and preprocessing receiving constraint data, motion trajectory data and running feedback data of the conveying process, and constructing a standardized conveying state data set; S2, analyzing the resource occupation degree of the single robot, and dynamically adjusting the conveying priority order corresponding to each device; S3, evaluating the scheduling sequence stability, and continuously adjusting the numerical interval between the scheduling values of each device; and S4, evaluating the running load of the executable action sequence, and driving the data update of the strategy network.The problems that the existing single-robot scheduling method is difficult to maintain a sustained and stable balance between the improvement of device utilization and the suppression of empty moving distance due to the lack of a long-term running data-driven adaptive trade-off mechanism are solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for transport processes, specifically to a single-robot intelligent scheduling method and system for transport based on dynamic priority learning. Background Technology

[0002] With the continuous improvement of flexible manufacturing and intelligent logistics in industrial settings, material handling by single robots in multi-equipment production environments has become an important component of discrete manufacturing systems, widely used in processing unit connections, inter-process transfers, and multi-station collaborative operations. Existing material handling scheduling methods typically generate robot execution sequences based on information such as equipment status, task queues, or spatial distances. By comprehensively judging equipment idle status, task arrival order, and robot current position, basic scheduling and path planning for material handling tasks are achieved. In practical applications, some scheduling schemes also incorporate equipment utilization statistics, task waiting time recording, and path length evaluation to reduce unnecessary robot movement overhead while ensuring production cycle time, thereby constructing material handling control logic with a certain real-time response capability.

[0003] For example, invention patent CN112396286B discloses an automatic pickup and delivery method and system for mobile robots. The method includes the following steps: receiving pickup orders pushed by a vending machine server; creating delivery orders and assigning target robots to the delivery orders; pushing the status of the delivery orders and the status of the target robots to the vending machine server until the target robots move to the pickup point; interacting with the vending machine server until the target robots enter, connect, and exit the vending machine; and continuing to push the status of the delivery orders to the vending machine server until the target robots move to the target delivery point. This invention, through the order and information pushes from the scheduling server and the vending machine server, enables robots to automatically receive and deliver goods placed in smart vending machines. In particular, it can utilize robots to move and deliver some relatively heavy pre-ordered items, thereby providing users with efficient, convenient, and automated intelligent services and improving user experience.

[0004] For example, the invention patent with announcement number CN119005830B relates to the field of intelligent scheduling technology for multi-robot collaborative operations. Specifically, it discloses an intelligent scheduling method and system for multi-robot collaborative operations. The method includes: confirmation of handling robots, extraction of delivery information, confirmation of delivery robots, and analysis of delivery saturation. This invention confirms the handling robots assigned to each item by combining the load-bearing compatibility between the goods and the handling robots, and confirms the delivery robots assigned to each item by combining the power compatibility between the delivery robots and the delivered goods. By combining the total number of delivered goods and the delivery availability of the delivery robots, the invention analyzes the delivery saturation of each delivery sub-area, thereby improving the safety and stability of goods handling, avoiding affecting the timely delivery of goods, thus preventing customer loss or increased complaints, improving the accuracy and effectiveness of decision-making, and allowing for reasonable adjustments and optimizations based on actual conditions.

[0005] However, in long-term operation scenarios with multiple devices and a single robot, existing scheduling methods generally rely on fixed rules or short-term indicators for decision-making, making it difficult to characterize the long-term trade-off between improving equipment utilization and suppressing the robot's idle travel distance. One type of method focuses on shortening the single transport path or local idle travel distance, which can easily lead to waiting accumulation for high-load equipment during continuous production. Another type of method prioritizes equipment utilization and task throughput, often causing robots to frequently move across areas, resulting in increased idle travel distance. Due to the lack of a mechanism for continuously correcting and evolving scheduling strategies based on newly generated data during operation, scheduling decisions are difficult to dynamically adjust with changes in load structure, spatial congestion, and task distribution at different production stages. This can easily lead to trade-off imbalances in long-term operation, thereby restricting overall transport efficiency and system stability.

[0006] To address the above issues, there is an urgent need for a single-robot intelligent scheduling method and system for material handling based on dynamic priority learning. Summary of the Invention

[0007] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a single-robot intelligent scheduling method and system based on dynamic priority learning. This solves the problem that existing single-robot scheduling methods lack an adaptive trade-off mechanism driven by long-term operational data, making it difficult to maintain a continuous and stable balance between improving equipment utilization and suppressing idle travel distance.

[0008] Technical solution To achieve the above objectives, this invention provides the following technical solution: a single-robot intelligent scheduling method and system for material handling based on dynamic priority learning, comprising: S1, collecting receiving constraint data, motion trajectory data, and operational feedback data during the material handling process, and preprocessing the collected receiving constraint data, motion trajectory data, and operational feedback data to construct a standardized material handling state dataset; S2, based on the standardized material handling state dataset, performing scheduling analysis on the resource occupancy of a single robot performing a material handling task at the current moment from the perspective of equipment operation continuity requirements and the robot's current physical operating state, and dynamically adjusting the material handling priority order of each device based on the scheduling analysis results; S3, based on the standardized material handling state dataset, performing operational evaluation on the scheduling order stability from the operational feedback relationship between the robot's actual motion consumption and the equipment's real-time load performance, and continuously adjusting the numerical spacing between the scheduling values ​​of each device based on the operational evaluation results; S4, using the operational evaluation results as input, comprehensively evaluating the operational load of the executable action sequence from the robot's path unfolding state and motion execution state, and driving the data update of the strategy network based on the comprehensive evaluation results.

[0009] Furthermore, the specific steps for collecting the receiving constraint data, motion trajectory data, and operational feedback data during the transport process are as follows: Collect receiving constraint data during the equipment's operation, including: the quantity of materials being processed by each device, the number of available channels at the material inlet, the number of workstations actually performing actions, and the start and end times of the transport actions; Collect motion trajectory data during the robot's walking execution, including: the robot's real-time pose coordinates, chassis size parameters, travel distance, idle time, direction adjustment angle, instantaneous passage delay value, instantaneous path length increment, instantaneous walking speed, instantaneous acceleration, instantaneous angular velocity, the angle between the robot's current orientation and the target device's receiving direction, as well as the spatial positions of fixed devices in the environment, the passable boundaries of the passage area, and the real-time obstacle positions; Collect operational feedback data during a single transport closed-loop process, including: the remaining continuous operating time of the equipment, waiting time, and the number of completed delivery tasks.

[0010] Furthermore, the specific steps for preprocessing the collected received constraint data, motion trajectory data, and operational feedback data to construct a standardized transport state dataset are as follows: For the received constraint data, an alignment processing algorithm is performed according to the timestamp order, and a neighbor consistency deduplication algorithm is performed to delete redundant records for records that are continuously and repeatedly reported and whose states remain unchanged; For the motion trajectory data, a sampling record with abnormal jumps is deleted using an adjacent sampling point displacement continuity detection algorithm; For records that appear repeatedly in the operational feedback data, an event deduplication and aggregation algorithm is performed to delete invalid records with missing key timestamps; The received constraint data, motion trajectory data, and operational feedback data after standardization processing are normalized using a range normalization algorithm to construct a standardized transport state dataset; Based on the robot's real-time pose coordinates and bottom... The robot chassis dimensions, spatial positions of fixed devices in the environment, environmental boundary information, and real-time obstacle positions output by real-time obstacle detection sensors are discretized in a unified coordinate system. The robot chassis shape is then geometrically expanded according to the chassis dimensions to determine impassable areas. Connectivity analysis is performed on the remaining spatial areas to form a set of nodes composed of traversable area units. Node connectivity is established based on the existence of continuous passageways between areas to construct a traversable space topology. In the traversable space topology, the robot's real-time pose coordinates are used as the starting node, and the spatial position of the target device is used as the target node. The length of the path edge is accumulated using a feasible path search algorithm to obtain the length of the unloaded feasible path from the robot to the target device, while simultaneously recording the instantaneous unloaded path length increment at the corresponding moment.

[0011] Furthermore, the specific steps for scheduling and analyzing the resource occupancy of a single robot performing a transport task at the current moment based on the standardized transport status dataset, considering the continuity requirements of equipment operation and the current physical operating state of the robot, are as follows: Multiply the remaining sustainable operating time of the equipment by the length of the unloaded feasible path from the robot to the target equipment, add one, and take the natural logarithm to obtain the time path saturation mapping value; add the robot's instantaneous passage delay value to the angle between the robot's current orientation and the target equipment's receiving direction to obtain the passage attitude deflection composite value; perform an arctangent operation on the passage attitude deflection composite value to obtain the passage impedance mapping value; multiply the number of available channels at the target equipment's material inlet by the number of workstations where the target equipment is actually performing actions, and then add this to the quantity of material being processed by the target equipment to obtain the equipment's instantaneous processing scale value; perform a square root operation on the equipment's instantaneous processing scale value to obtain the equipment receiving capacity mapping value; divide the product of the time path saturation mapping value and the passage impedance mapping value by the equipment receiving capacity mapping value to obtain the instantaneous scheduling value.

[0012] Further, the specific steps for dynamically adjusting the transport priority order of each device based on the scheduling analysis results are as follows: The instantaneous scheduling values ​​corresponding to each device are mapped to a fixed-length vector structure according to the device number order. Using the scheduling time as the row index, the corresponding vector is written into the current row position of the matrix data structure, forming a dynamic priority matrix representing the scheduling priority relationship of each device at the current time. The number of matrix columns corresponds one-to-one with the number of devices, and the number of matrix rows increases with the scheduling time. After writing the data to the current row of the dynamic priority matrix, a pairwise comparison-based sorting calculation is performed on the instantaneous scheduling values ​​stored in each column position of the current row. Through a sorting process where the number of comparisons increases logarithmically with the number of devices, a device execution sequence is generated, arranged in ascending order of instantaneous scheduling values. The device corresponding to the first position of the sorting result is determined as the target device for the next transport task. The system is as follows: After the robot completes the current transport and releases the material, it rereads the latest row of data in the dynamic priority matrix and repeats the sorting process to determine the subsequent transport target. When the number of available channels of the equipment changes, the amount of material being processed by the equipment changes, the robot's real-time passage delay changes, or the robot's real-time pose coordinates change, the instantaneous scheduling value of the corresponding equipment is recalculated, and the updated value is written to the corresponding column position in the current row of the dynamic priority matrix. At the same time, the sorting process is re-executed in the current row of the dynamic priority matrix. When the sorting result changes and the robot has not yet issued the next transport instruction, the updated first-ranked equipment in the sorting is used as the new transport target to generate the path and issue the execution instruction. When the robot is already performing a transport action, the next target is determined based on the latest sorting result after the current transport is completed, without interrupting the already executed motion instruction.

[0013] Furthermore, the specific steps for evaluating the scheduling stability based on the standardized transport status dataset and the operational feedback relationship between the robot's actual motion consumption and the equipment's real-time load-bearing performance are as follows: A transport cycle is defined as the continuous operating time period from when the robot completes one material delivery and releases the material until the next material delivery and release; the robot's moving distance within a transport cycle is cumulatively calculated to obtain a cumulative moving distance value; the equipment's waiting time within a transport cycle is cumulatively calculated to obtain a cumulative equipment-side waiting time value; the robot's idle time within a transport cycle is cumulatively calculated to obtain a cumulative robot idle time value; the number of available channels for the equipment within a transport cycle is cumulatively calculated to obtain a cumulative number of available channels for the equipment; and so on. The robot's orientation adjustment angles are cumulatively calculated within each transport cycle to obtain the cumulative orientation adjustment angle value. The number of delivery tasks completed within a transport cycle is cumulatively calculated to obtain the total number of delivery tasks completed. An inverse hyperbolic sine operation is performed on the ratio of the robot's cumulative travel distance in the current transport cycle to the cumulative waiting time on the equipment side to obtain the idle displacement waiting mapping value. An inverse hyperbolic sine operation is performed on the ratio of the robot's cumulative idle time in the current transport cycle to the cumulative number of available channels on the equipment side to obtain the idle receiving mapping value. The ratio of the robot's cumulative orientation adjustment angle in the current transport cycle to the total number of delivery tasks completed is incremented by one, and the natural logarithm is taken to obtain the attitude task mapping value. The idle displacement waiting mapping value, the idle receiving mapping value, and the attitude task mapping value are added together to obtain the running state self-calibration value.

[0014] Further, the specific steps for continuously adjusting the numerical spacing between the scheduling values ​​of each device based on the operational evaluation results are as follows: After completing the calculation of the operational self-calibration value for the current transport cycle, the sorting result of the instantaneous scheduling values ​​corresponding to each device in the current row of the dynamic priority matrix remains unchanged. The difference between the instantaneous scheduling values ​​of adjacent devices is calculated for the sorted instantaneous scheduling values ​​to obtain the adjacent scheduling difference. The adjacent scheduling difference is then continuously proportionally mapped to the operational self-calibration value to obtain the updated adjacent scheduling difference. While keeping the instantaneous scheduling value corresponding to the first device unchanged, the current row of the dynamic priority matrix is ​​reconstructed sequentially based on the updated adjacent scheduling difference, so that the numerical spacing between the instantaneous scheduling values ​​of each device in the matrix changes synchronously with the operational self-calibration value. In subsequent transport decisions, the device sorting result is read based on the reconstructed dynamic priority matrix. When the adjacent scheduling difference corresponding to adjacent devices is continuously amplified N times, the sorting result remains consistent within N transport cycles. When the adjacent scheduling difference corresponding to adjacent devices is continuously compressed N times, the sorting result is allowed to undergo position exchange within N transport cycles.

[0015] Furthermore, the specific steps for comprehensively evaluating the operational load of the executable action sequence based on the robot's path deployment state and motion execution state, using the operational evaluation results as input, are as follows: multiply the robot's instantaneous path length increment by its instantaneous walking speed to obtain the path motion flux value; square the robot's instantaneous acceleration to obtain the acceleration / deceleration inertial dissipation value; square the robot's instantaneous angular velocity to obtain the turning attitude dissipation value; add the path motion flux value, acceleration / deceleration inertial dissipation value, and turning attitude dissipation value to the robot's instantaneous idle path length increment to obtain the basic path motion cost value; perform an exponential operation on the natural constant e with the operational self-calibration value as the exponent to obtain the operational self-calibration modulation value; multiply the basic path motion cost value by the operational self-calibration modulation value to obtain the modulated instantaneous value; and integrate the modulated instantaneous value over time within the time interval from the start to the end of the current conveying action to obtain the execution pruning effect value.

[0016] Further, the specific steps for updating the data of the strategy network based on the comprehensive evaluation results are as follows: Real-time comparison of the current pruning action amount with the pruning action threshold, where the pruning action threshold includes a first pruning threshold and a second pruning threshold, and the first pruning threshold is greater than the second pruning threshold; when the pruning action amount is less than or equal to the second pruning threshold, the actual walking trajectory, speed change curve, angular velocity change curve, and empty path length of the corresponding conveying action sequence are written into the data acquisition buffer of the strategy network according to the complete conveying cycle, and the equipment waiting time, equipment receiving completion time, and robot material release time within the conveying cycle are bound and recorded, and used as high-confidence running sample input in subsequent strategy network updates at a fixed proportion not less than all samples; when When the pruning effect is greater than the second pruning threshold and less than or equal to the first pruning threshold, the number of path replanning events, the duration of acceleration / deceleration segments, the cumulative length of idle segments, and the actual receiving delay of the corresponding devices in the current transport cycle are written into the data acquisition buffer of the policy network. In the next policy network update, these data are proportionally extracted to participate in the iterative calculation of policy parameters. When the pruning effect is greater than the first pruning threshold, the pending actions in the current action sequence that have not yet been sent to the motion control interface are terminated. The pose trajectory segments that have occurred, the length of the incomplete path segments, the current receiving status of the devices, and the environmental congestion indicators are synchronously written into the key event data area of ​​the policy network. These data are then used as mandatory samples in the subsequent policy network update process to participate in the update of policy parameters.

[0017] The second aspect of this invention provides a single-robot intelligent scheduling system for material handling based on dynamic priority learning, comprising: a multi-source state perception module, used to collect receiving constraint data, motion trajectory data, and operational feedback data during the material handling process, and to preprocess the collected receiving constraint data, motion trajectory data, and operational feedback data to construct a standardized material handling state dataset; a dynamic priority matrix generation module, used to perform scheduling analysis on the resource occupancy of a single robot performing a material handling task at the current moment based on the standardized material handling state dataset, considering the continuous operation requirements of the equipment and the current physical operating state of the robot, and to dynamically adjust the material handling priority order corresponding to each equipment based on the scheduling analysis results; a self-updating trade-off mapping module, used to perform operational evaluation on the scheduling order stability based on the operational feedback relationship between the robot's actual motion consumption and the equipment's real-time load performance, and to continuously adjust the numerical spacing between the scheduling values ​​of each equipment based on the operational evaluation results; and an operational load-driven strategy update module, used as input to comprehensively evaluate the operational load of the executable action sequence from the robot's path unfolding state and motion execution state, and to drive the data update of the strategy network based on the comprehensive evaluation results.

[0018] Beneficial effects The present invention has the following beneficial effects: (1) The single robot transport intelligent scheduling method and system based on dynamic priority learning reduces the waiting accumulation of high-load equipment in the continuous production stage by dynamically rearranging the priority order of equipment when the task arrives and the equipment status changes.

[0019] (2) The single-robot intelligent scheduling method and system based on dynamic priority learning suppresses the amplification of invalid movement caused by frequent cross-regional transport by synchronously constraining the robot’s empty path unfolding behavior in the scheduling decision.

[0020] (3) The single-robot intelligent scheduling method and system based on dynamic priority learning avoids generating unexecutable transport actions under conditions of environmental congestion or path restriction by adjusting the action sequence in real time during execution.

[0021] (4) The single-robot intelligent scheduling method and system based on dynamic priority learning realizes the continuous optimization and adaptive convergence of scheduling decisions in long-term operation by linking the path planning results and operation evaluation results to the strategy update process.

[0022] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0023] Figure 1This is a flowchart of the intelligent scheduling method for single-robot transport based on dynamic priority learning according to the present invention. Figure 2 This is a structural diagram of the single-robot intelligent scheduling system for material handling based on dynamic priority learning, as described in this invention. Figure 3 This invention relates to a dynamic change diagram of the value of the clipping action. Figure 4 This is a schematic diagram of the single-robot handling priority dynamic update and execution process involved in the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figures 1-4 This invention provides a technical solution: a single-robot intelligent scheduling method and system for material handling based on dynamic priority learning, comprising: S1, collecting receiving constraint data, motion trajectory data, and operational feedback data during the material handling process, and preprocessing the collected receiving constraint data, motion trajectory data, and operational feedback data to construct a standardized material handling state dataset; S2, based on the standardized material handling state dataset, performing scheduling analysis on the resource occupancy of a single robot performing a material handling task at the current moment from the perspective of equipment operation continuity requirements and the robot's current physical operating state, and dynamically adjusting the material handling priority order corresponding to each device based on the scheduling analysis results; S3, based on the standardized material handling state dataset, performing operational evaluation on the scheduling order stability from the operational feedback relationship between the robot's actual motion consumption and the equipment's real-time load performance, and continuously adjusting the numerical spacing between the scheduling values ​​of each device based on the operational evaluation results; S4, using the operational evaluation results as input, comprehensively evaluating the operational load of the executable action sequence from the robot's path unfolding state and motion execution state, and driving the data update of the strategy network based on the comprehensive evaluation results.

[0026] Specifically, the steps for collecting receiving constraint data, motion trajectory data, and operation feedback data during the conveying process are as follows: Collect receiving constraint data during the equipment operation process. The receiving constraint data is uniformly recorded using the equipment control coordinate system as a spatial reference and the equipment's local control clock as a time reference. The receiving constraint data includes the quantity of materials being processed by each piece of equipment, the number of available channels at the material inlet, the number of workstations actually performing actions, the start time of the conveying action, and the end time of the conveying action. All time records are written using timestamps and cross-equipment time alignment is completed in the subsequent processing stage.

[0027] The system collects motion trajectory data during the robot's walking process. The motion trajectory data uses the robot's map coordinate system as a unified spatial reference and is continuously sampled using the robot's odometer clock as a time reference. The motion trajectory data includes the robot's real-time pose coordinates, chassis size parameters, travel distance, idle time, orientation adjustment angle, instantaneous passage delay value, instantaneous path length increment, instantaneous walking speed, instantaneous acceleration, instantaneous angular velocity, and the angle between the robot's current orientation and the target device's receiving direction. Simultaneously, environmental spatial information is collected. The spatial positions of each fixed device in the environment and the passable boundaries of the passage area are statically calibrated based on the map coordinate system. The real-time obstacle positions are dynamically updated based on the robot's perception coordinate system and mapped to the map coordinate system through coordinate transformation, thus synchronizing with the robot's motion trajectory data on the same time axis.

[0028] The system collects operational feedback data during a single transport loop. The operational feedback data is aligned with the end time of the transport action and is compiled in conjunction with the equipment operation records and robot execution records. The operational feedback data includes the remaining time for the equipment to continue operating, the waiting time, and the number of completed delivery tasks. The relevant data forms a complete record after each transport loop ends, which is used as feedback input for subsequent scheduling evaluation and strategy update processes.

[0029] In this implementation plan, by unifying the spatial coordinate reference and time synchronization method, the equipment-side operation carrying information, robot walking execution information, and conveying closed-loop feedback information are collected and aligned in a consistent manner. This enables data from different sources to form a continuous and associative state expression within the same spatiotemporal framework, thereby providing basic data support with physical and temporal consistency for subsequent scheduling analysis, priority calculation, and trade-off mapping updates, and avoiding state interpretation deviations caused by coordinate mixing and time drift.

[0030] Specifically, the preprocessing of the collected receiving constraint data, motion trajectory data, and operation feedback data to construct a standardized transport status dataset involves the following steps: For the receiving constraint data, timestamp sorting based on total order relation is performed according to the order of the collected timestamps, so that the same device forms a strictly monotonically increasing time series within a continuous sampling period; After time alignment, adjacent consistency deduplication is performed on records with completely consistent status fields within a continuous sampling period and a duration exceeding one minimum sampling period, retaining only key records where the status has changed, thereby eliminating redundant data caused by high-frequency repeated reporting.

[0031] For motion trajectory data, anomaly detection processing is performed based on the displacement continuity of adjacent sampling points in a unified coordinate system. When the spatial displacement between two adjacent sampling points exceeds the maximum motion range that the robot can reach in a single sampling period, the sampling record is determined to be an abnormal jump and is removed to ensure the continuity of trajectory data in space and time.

[0032] For the operational feedback data, the records are grouped into event-level groups according to the start and end times of the transfer action. For feedback records that occur repeatedly within the same transfer event and have the same timestamp, event aggregation processing is performed, and invalid records that are missing key timestamp identifiers are deleted, so that each transfer loop retains only one set of complete and valid operational feedback information.

[0033] After completing the above standardization process, range normalization processing based on the minimum and maximum value interval mapping is performed on the received constraint data, motion trajectory data, and operation feedback data in their respective dimensions to ensure that different physical quantities are consistent on the numerical scale and to convert all collected data into dimensionless data for subsequent formula calculations, thereby constructing a transport state dataset with a unified expression.

[0034] Based on the robot's real-time pose coordinates, chassis size parameters, spatial positions of fixed devices in the environment, environmental boundary information, and obstacle positions output by real-time obstacle detection sensors, the environmental space is discretized into a grid in a map coordinate system. The grid resolution is not less than the robot's minimum controllable displacement within a single control cycle. In the discrete space, the robot chassis shape is geometrically expanded according to the chassis size parameters and a fixed safety margin to define the spatial extent of impassable areas. After removing impassable areas, connectivity analysis is performed on the remaining space. Any two spatial units that can be geometrically reached through continuous collision-free motion trajectories are classified as the same traversable area unit, and a node set is formed with the geometric center of each traversable area unit as a node. When there is a continuous passage between two area units that does not intersect with impassable areas, a connectivity relationship is established between the corresponding nodes, and the Euclidean distance between the node centers is used as the edge length attribute, thereby constructing a topological graph of the traversable space. In the traversable space topology graph, the robot's pose node at the current scheduling moment is used as the starting point of the path search, and the node corresponding to the spatial location of the target device is used as the ending point of the path search. By accumulating the length of each connected edge in the path through a feasible path search process that expands node by node, the length of the unloaded feasible path for the robot to reach the target device at the current scheduling moment is obtained. At the same time, the difference in the length of the unloaded feasible path between adjacent scheduling moments is used as the instantaneous unloaded path length increment to reflect the instantaneous changes in the path structure caused by pose changes.

[0035] In this implementation plan, by establishing clear time alignment rules, spatial modeling rules, and data filtering rules, the received constraint information, motion trajectory information, and operation feedback information are uniformly organized and constrained. This ensures that all types of data maintain a continuous, stable, and comparable state expression under the same time scale and spatial coordinate reference. Based on this, a spatial topology representation with clear passage boundaries and path structure is formed, thereby providing reliable and reproducible basic data support for subsequent calculation of empty path length, characterization of real-time path changes, and evaluation of scheduling values.

[0036] Specifically, based on a standardized transport status dataset, the scheduling analysis of the resource occupancy of a single robot performing a transport task at the current moment is conducted from the perspectives of equipment operation continuity requirements and the robot's current physical operating state. The specific steps are as follows:

[0037] The time path saturation mapping value is obtained by multiplying the remaining sustainable operating time of the equipment by the length of the unloaded feasible path from the robot to the target equipment, adding one, and taking the natural logarithm. The remaining sustainable operating time of the equipment and the length of the unloaded feasible path are both non-negative real measurements. Adding one is used to avoid numerical discontinuity caused by taking the logarithm of zero. The time path saturation mapping value increases monotonically with any input quantity and takes values ​​in the non-negative real number range.

[0038] The robot's instantaneous passage delay value is added to the angle between the robot's current orientation and the target device's receiving direction to obtain the passage attitude deflection composite value. Both the instantaneous passage delay value and the angle are non-negative real measurements. An arctangent operation is performed on the passage attitude deflection composite value to obtain the passage impedance mapping value. The passage impedance mapping value remains monotonically increasing in the non-negative input range and has a limited output range, thereby avoiding extreme value amplification from interfering with subsequent sorting.

[0039] Multiply the number of available channels at the material inlet of the target equipment by the number of workstations where the target equipment is actually performing actions, and then add the value of the quantity of material being processed by the target equipment to obtain the real-time processing scale value of the equipment. All input quantities are non-negative counts. Perform a square root operation on the real-time processing scale value of the equipment to obtain the equipment receiving capacity mapping value. The square root operation ensures that the mapping result remains monotonically increasing in the non-negative interval, while compressing the large-scale value range to enhance the comparability between different equipment.

[0040] The instantaneous scheduling value is obtained by dividing the product of the time path saturation mapping value and the passage impedance mapping value by the device receiving capability mapping value. The denominator is always greater than zero to ensure calculation stability. The instantaneous scheduling value maintains continuous change characteristics during the input change process, which is used to support direct sorting comparison between different devices.

[0041] The formula for calculating the instantaneous scheduling quantity is: ; In the formula, It indicates the remaining time that the equipment can continue to operate. It is used to quantify the time span that the equipment can maintain continuous operation in its current operating state before any shutdown or blockage occurs. It is a direct indicator of the urgency of the equipment's operation and is derived from the real-time process status data output by the PLC. It represents the length of the unloaded feasible path from the robot to the target device. It is used to quantify the actual walking distance required for the robot to move from its current position to the corresponding device without carrying any materials. It is a basic indicator reflecting the cost of spatial movement and is derived from the path calculation results of the robot's real-time pose information and the device layout map. It represents the robot's real-time passage delay value, which is used to quantify the waiting and deceleration that the robot may experience on the path under the current passage conditions. It is a direct indicator reflecting the channel occupancy status and is derived from the feedback signal of the channel occupancy sensor. It represents the angle between the robot's current orientation and the target device's receiving direction. It is used to quantify the directional adjustment range that the robot needs to make before performing the transport task. It is a direct indicator of the cost of attitude adjustment and comes from the robot's attitude sensor and the device's receiving direction identification information. This indicates the quantity of material being processed by the target equipment. It is used to quantify the processing load level occupied by the equipment at the current moment and is a direct indicator reflecting the processing saturation of the equipment. It is derived from the material processing quantity information output by the equipment operation status interface. This indicates the number of available channels at the material inlet of the target equipment. It is used to quantify the number of material channels that the equipment can still directly receive at the current moment. It is a direct numerical indicator reflecting the equipment's real-time receiving capability and is derived from the channel occupancy detection data output by the equipment's inlet sensors. It indicates the number of workstations where the target equipment is actually performing actions. It is used to quantify the number of workstations that are performing processing and handling actions inside the equipment at the current moment. It is a direct numerical indicator reflecting the real-time occupancy level of the equipment and is derived from the execution count data of the equipment workstation execution signal output.

[0042] This implementation plan unifies the time occupation, spatial movement, passage restrictions, posture adjustments, and real-time reception and processing capabilities that a single robot must simultaneously bear when facing multiple devices into a single, directly comparable continuous value. This provides a unique, objective, and real-time updatable quantitative basis for determining the transport target. The formula, through a joint mapping of the remaining sustainable operating time of the equipment and the robot's idle path length, characterizes the basic scale of time and space occupation the robot needs to avoid equipment operation interruptions. Through a nonlinear mapping of passage delay and posture deflection, it reflects the actual motion impedance faced by the robot in completing the path under the current channel state and posture conditions. Simultaneously, it introduces the quantity of materials being processed by the equipment, the number of available receiving channels, and the number of workstations currently executing, quantitatively constraining the equipment's true absorption capacity for external transport behavior at the current moment. All these directly collected quantities are uniformly calculated in the same formula, enabling the instantaneous scheduling value to truly reflect the immediate intensity of the occupation of the robot's overall operating resources when selecting a particular device as the next transport target.

[0043] Specifically, the steps for dynamically adjusting the transport priority order of each device based on the scheduling analysis results are as follows: The instantaneous scheduling values ​​corresponding to each device are mapped to a fixed-length vector structure according to the device number order. Using the scheduling time as the row index, the corresponding vector is written into the current row position of the matrix data structure, forming a dynamic priority matrix that characterizes the scheduling priority relationship of each device at the current time. The number of columns in the matrix corresponds one-to-one with the number of devices, and the number of rows in the matrix increases with the scheduling time. After writing the data to the current row of the dynamic priority matrix, a pairwise comparison-based sorting calculation is performed on the instantaneous scheduling values ​​stored in each column position of the current row. Through a sorting process where the number of comparisons increases logarithmically with the number of devices, a device execution sequence is generated, arranged in ascending order of instantaneous scheduling values. The device corresponding to the first position of the sorting result is determined as the target device for the next transport task. After the robot completes the current transport and releases the material, the latest row of the dynamic priority matrix is ​​reread, and the sorting process is repeated to determine the subsequent transport target. When the number of available channels, the quantity of material being processed, the robot's real-time passage delay, or the robot's real-time pose coordinates change, the instantaneous scheduling value of the corresponding device is recalculated, and the updated value is written to the corresponding column position in the current row of the dynamic priority matrix. At the same time, the sorting process is re-executed in the current row of the dynamic priority matrix. The change in the number of available channels is triggered by inconsistent counts in adjacent sampling periods but a consistent directional change over three consecutive sampling periods. The change in the quantity of material being processed is triggered by an increase or decrease in the count in adjacent sampling periods but a non-zero net change over three consecutive sampling periods. The change in the robot's real-time passage delay is triggered by the cumulative delay increment in the same direction over three consecutive control periods exceeding the control delay increment. The system uses the cycle time resolution as a trigger condition. The real-time pose coordinate change of the robot is triggered by the migration of the map grid cell into which the pose projection falls, resulting in a change in the reconnection of the reachable path of the target device. This suppresses frequent recalculations caused by single jitter. When the sorting result changes and the robot has not yet issued the next transport command, the updated first device in the sorting is used as the new transport target to generate a path and issue an execution command. When the robot is already performing a transport action, the next target is determined based on the latest sorting result after the current transport is completed. The already executed motion command is not interrupted. At the same time, safety boundary conditions are added. When the obstacle avoidance sensor outputs a collision risk trigger signal and the motion control interface returns a forced stop flag, the pending actions are immediately terminated and the current motion command is safely braked. After the risk is cleared, a new executable path is generated based on the latest sorting result of the dynamic priority matrix and issued.

[0044] In this implementation plan, by organizing instantaneous scheduling values ​​in a matrix and combining them with trigger recalculation rules constrained by jitter suppression, the priority order of equipment can be updated in a timely manner as the operating status changes. At the same time, the continuity of instructions is maintained during the transport execution phase to avoid frequent interruptions that cause motion instability. On this basis, clear safety boundary conditions are introduced to ensure that scheduling updates are connected in an orderly manner while meeting obstacle avoidance and risk control requirements, thereby achieving stable coordination between scheduling decision updates, path generation, and motion execution.

[0045] Specifically, based on a standardized transport status dataset, the operational evaluation of scheduling stability is conducted by examining the operational feedback relationship between the robot's actual motion consumption and the equipment's real-time load-bearing performance. The specific steps are as follows: A transport cycle is defined as the continuous operating time from when the robot completes one material delivery and releases the material until the next material delivery and release. During the transport cycle, all state variables are recorded at a fixed sampling frequency and the sampling time sequence is used as the basis for accumulation. The robot's travel distance within a transport cycle is accumulated segment by segment according to the displacement increment between adjacent sampling times to obtain the cumulative travel distance value; the equipment's waiting time within a transport cycle is accumulated segment by segment according to the time interval corresponding to the sampling cycle to obtain the cumulative equipment waiting time value; the robot's idle time within a transport cycle is accumulated segment by segment according to the time segments within the sampling cycle when no transport action is performed to obtain the cumulative idle time value; the number of available channels of the equipment within a transport cycle is accumulated segment by segment according to the channel number value recorded at each sampling time to obtain the cumulative number of available channels on the equipment side; the robot's orientation adjustment angle within a transport cycle is accumulated segment by segment according to the orientation change angle at adjacent sampling times to obtain the cumulative orientation adjustment angle value; and the number of times a delivery task is completed within a transport cycle is counted and accumulated according to each delivery completion event to obtain the total number of completed delivery tasks. An inverse hyperbolic sine operation is performed on the ratio of the cumulative value of the robot's movement distance to the cumulative value of the waiting time on the equipment side within the current conveying cycle to obtain the idle displacement waiting mapping value. The ratio calculation is based on the cumulative results within the same conveying cycle to ensure consistency of dimensions. An inverse hyperbolic sine operation is performed on the ratio of the cumulative idle time of the robot to the cumulative number of available channels on the equipment side within the current transport cycle to obtain the idle reception mapping value. The ratio calculation is based on the cumulative results within the same transport cycle to ensure consistency of the time scale. The ratio of the cumulative value of the robot's direction adjustment angle in the current delivery cycle to the total number of completed delivery tasks is increased by one and then the natural logarithm is taken to obtain the posture task mapping value. The addition of one is used to avoid discontinuity in logarithmic calculation caused by zero value input. The idle displacement waiting mapping value, idle reception mapping value and attitude task mapping value are added together to obtain the running state self-calibration value, which is used to characterize the comprehensive feedback level of the running state to the scheduling strategy adjustment within the current transport cycle.

[0046] The formula for calculating the self-calibration value in the running state is: ; In the formula, This represents the cumulative distance the robot has traveled during the current transport cycle. It is used to quantify the actual displacement of the robot when it is not carrying any materials. It is a direct indicator of the energy consumed during unloaded travel and is derived from the cumulative displacement output of the robot's odometer during this cycle. This represents the cumulative waiting time of the equipment during the current transport cycle. It is used to quantify the duration during which the equipment is in a waiting state due to the failure to complete the transport in a timely manner. It is a direct indicator reflecting the degree of obstruction to the equipment's operation and is derived from the cumulative time output of the equipment's waiting timer during this cycle. This represents the cumulative idle time of the robot within the current transport cycle. It is used to quantify the duration during which the robot has not performed any effective transport actions within this cycle. It is a direct indicator reflecting the degree to which robot resources are not occupied and is derived from the idle state timing output recorded by the robot controller. This represents the cumulative number of available channels on the equipment side within the current transport cycle. It is used to quantify the physical receiving capacity of the equipment for external transport activities within this cycle. It is a direct indicator reflecting the equipment's receiving capacity and is derived from the cumulative statistical output of the equipment's access point occupancy detection signal within this cycle. This represents the cumulative value of the robot's orientation adjustment angle within the current transport cycle. It is used to quantify the magnitude of the robot's posture correction during the execution of the transport task and is a direct indicator of the complexity of the motion path. It originates from the cumulative output of the robot's posture sensor within this cycle. This indicates the total number of delivery tasks completed within the current transport cycle. It is used to quantify the number of transport tasks that were actually successfully received and completed by the equipment within this cycle. It is a direct indicator reflecting the degree to which the transport results are absorbed by the equipment and is derived from the count output of the task completion confirmation signal within this cycle.

[0047] In this implementation scheme, operational feedback information such as idle displacement, equipment waiting time, robot idle time, and posture adjustment that actually occur within a single transport cycle is uniformly mapped to a continuously changing numerical value. This value is used to characterize the instantaneous mismatch between the robot's motion consumption and the actual absorption capacity of the equipment. By calculating this value, the cumulative deviation generated at the operational level by the current scheduling behavior can be continuously reflected without introducing thresholds, relying on historical data, or performing parameter training. Based on the self-updating operational feedback value, the numerical spacing between the scheduling values ​​of each device in the dynamic priority matrix can be continuously adjusted according to the operational state. This allows the scheduling order to maintain stronger stability in cycles with high operational mismatch and higher adjustment sensitivity in cycles with low operational mismatch. This formula enables the scheduling system to possess a self-calibration capability naturally driven by actual operational feedback. It automatically suppresses frequent sorting switches when equipment waiting time accumulates or idle consumption increases, and allows the scheduling order to respond more quickly to real-time state changes when the operational burden is reduced, thereby achieving a dynamic balance between equipment operational continuity and robot motion load.

[0048] Specifically, the steps for continuously adjusting the numerical spacing between the scheduling values ​​of each device based on the operational evaluation results are as follows: After calculating the self-calibration value of the current transport cycle, the sorting result of the instantaneous scheduling value of each device in the current row of the dynamic priority matrix remains unchanged. The numerical difference between adjacent devices is calculated for each sorted instantaneous scheduling value to obtain the adjacent scheduling difference. The adjacent scheduling difference is then continuously proportionally mapped to the self-calibration value of the current cycle to obtain the updated adjacent scheduling difference. The continuous proportional mapping is based on the change range of the self-calibration value of the current cycle within adjacent transport cycles, so that the adjacent scheduling difference is continuously adjusted with the change of the operating state. While keeping the instantaneous scheduling value of the first device unchanged, the current row of the dynamic priority matrix is ​​reconstructed from top to bottom based on the updated adjacent scheduling difference, so that the numerical spacing between the instantaneous scheduling values ​​of each device in the matrix changes synchronously with the self-calibration value of the current cycle.

[0049] In subsequent transport decisions, the device sorting results are read based on the reconstructed dynamic priority matrix, where N represents the cumulative number of times the adjacent scheduling difference maintains the same trend of change within consecutive transport cycles. This cumulative number is updated at the end of each transport cycle based on the direction of change of the current adjacent scheduling difference relative to the previous transport cycle. When the adjacent scheduling difference corresponding to adjacent devices continues to increase within N consecutive transport cycles, the sorting results remain consistent within the corresponding N transport cycles. When the adjacent scheduling difference corresponding to adjacent devices continues to decrease within N consecutive transport cycles, the sorting results are allowed to undergo position swapping within the corresponding N transport cycles. This allows the scheduling sorting to form a constraint relationship that naturally evolves with the operating state between stability and adjustment flexibility.

[0050] In this implementation scheme, without changing the current priority ranking result, the spacing between adjacent scheduling values ​​is continuously reconstructed by introducing a self-calibrating value in the running state. This allows the equipment ranking to gradually widen and converge as the running state changes, and also suppresses frequent jitter through the continuous number of transport cycles. Thus, a sustainable evolution scheduling ranking update mechanism is formed between stability and responsiveness.

[0051] Specifically, taking the operational evaluation results as input, the operational load of the executable action sequence is comprehensively evaluated based on the robot's path unfolding state and motion execution state. The specific steps are as follows: The path motion flux value is obtained by multiplying the instantaneous path length increment of the robot by the instantaneous walking speed. Multiplication is used to characterize the path propulsion intensity per unit time, so that the path change amplitude and the motion speed form a joint characterization under the same dimension. The instantaneous acceleration of the robot is squared to obtain the acceleration and deceleration inertial dissipation value. The squaring operation is used to amplify the impact of acceleration fluctuations on motion load, so that frequent acceleration and deceleration are given higher numerical weight. The instantaneous angular velocity of the robot is squared to obtain the steering attitude dissipation value. The squaring operation is used to highlight the attitude adjustment cost caused by rapid changes in direction, making the steering intensity contribute more significantly to the cost. The path motion flux value, acceleration / deceleration inertia dissipation value, and turning attitude dissipation value are added to the robot's instantaneous idle path length increment to obtain the basic path motion cost value. The addition is used to linearly superimpose multiple independent sources of consumption at the same moment to form a unified instantaneous motion cost expression. An exponential operation with the self-calibration value of the operating state as the exponent is performed on the natural constant to obtain the self-calibration modulation value of the operating state. The exponential operation is used to map the operating state feedback to the modulation scale in a nonlinear manner, so that the change of the operating state has a continuous amplification or compression effect on the subsequent cost. Multiply the basic cost of path motion by the self-calibration modulation value of the running state to obtain the instantaneous value after modulation. The multiplication is used to realize the overall modulation of the basic motion cost by the running state, so that the two types of information form a coupling relationship at the numerical level. Within the time interval from the start to the end of the current transport action, the modulated instantaneous value is integrated over time to obtain the execution pruning effect value. The time integration is achieved by discretely calculating the modulated instantaneous value corresponding to each sampling moment under a fixed sampling period and multiplying it by the sampling time interval, thereby approximately obtaining the cumulative execution cost during the complete transport action in the sense of continuous time.

[0052] The formula for calculating the trimming effect value is as follows: ; In the formula, This indicates the end time of the current conveying action, used to define the termination time of the action integral, and is derived from the completion timestamp record generated when the robot completes the delivery and release of the material; This indicates the start time of the current conveying action, used to define the starting time point of the integral of the action, and is derived from the timestamp record of the execution command issued by the robot controller; It represents the instantaneous path length increment of the robot, used to quantify the actual walking displacement of the robot in the current time slice. It is a direct indicator describing the degree of path unfolding and is derived from the robot pose calculation results and pose changes at adjacent time points. It represents the robot's instantaneous walking speed, which is used to quantify the intensity of the robot's movement at the current moment. It is a direct indicator describing the path execution state and comes from the speed feedback signal output by the robot's drive controller. It represents the robot's instantaneous acceleration, used to quantify the robot's acceleration and deceleration behavior during path execution. It is a direct indicator describing motion smoothness and inertial changes, derived from the change in velocity feedback from the drive controller at adjacent time points. It represents the robot's instantaneous angular velocity, which is used to quantify the intensity of the robot's turning and attitude adjustments during path execution. It is a direct indicator of the complexity of path turning and is derived from the real-time output of the robot's attitude sensor. It represents the instantaneous increment of the robot's idle path length, used to quantify the walking portion that occurs in the current execution phase without carrying materials. It is a direct indicator describing the proportion of idle execution, derived from the joint feedback of task status identifier and pose change. This represents the self-calibration value in the running state, used to quantify the degree of mismatch between the robot's motion consumption and the actual load-bearing capacity of the equipment in the current running phase. It is the direct basis for performing continuous scale modulation on the path-motion action, and is derived from the real-time calculation results of running feedback data such as the idle travel distance, equipment waiting time, robot idle time, number of equipment receiving channels, attitude adjustment angle, and number of task completions within the current transport cycle.

[0053] In this implementation example, within a unified time integration interval, the start and end times of each transfer cycle are set to the same time base to ensure that the data of each transfer cycle are comparable under the same time scale.

[0054] Under the same time reference, the instantaneous path length increment for transport cycle one is set to 0.80, the instantaneous travel speed to 0.60, the instantaneous acceleration to 0.40, the instantaneous angular velocity to 0.30, the instantaneous unloaded path length increment to 0.20, and the running self-calibration value to 0.15. The instantaneous path length increment for transport cycle two is 1.00, the instantaneous travel speed is 0.70, the instantaneous acceleration is 0.50, the instantaneous angular velocity is 0.35, the instantaneous unloaded path length increment is 0.25, and the running self-calibration value is 0.20. The instantaneous path length increment for transport cycle three is 0.90, the instantaneous travel speed is 0.65, the instantaneous acceleration is 0.45, the instantaneous angular velocity is 0.32, the instantaneous unloaded path length increment is 0.22, and the running self-calibration value is 0.18. The instantaneous path length increment for transport cycle four is 1.20, the instantaneous travel speed is 0.80, the instantaneous acceleration is 0.60, the instantaneous angular velocity is 0.40, the instantaneous unloaded path length increment is 0.30, and the running self-calibration value is 0.25. The instantaneous path length increment for transport cycle five is 1.10, the instantaneous travel speed is 0.75, the instantaneous acceleration is 0.55, the instantaneous angular velocity is 0.38, the instantaneous unloaded path length increment is 0.28, and the running self-calibration value is 0.22. The instantaneous path length increment for transport cycle six is ​​1.30, the instantaneous travel speed is 0.85, the instantaneous acceleration is 0.65, the instantaneous angular velocity is 0.45, the instantaneous unloaded path length increment is 0.32, and the running self-calibration value is 0.28. The instantaneous path length increment for transport cycle seven is 1.00, the instantaneous travel speed is 0.70, the instantaneous acceleration is 0.50, the instantaneous angular velocity is 0.36, the instantaneous idle path length increment is 0.26, and the running self-calibration value is 0.19. The execution trimming effect value for each transport cycle is calculated, as shown in Table 1.

[0055] Table 1 Data Table of Cutting Effect Values

[0056] like Figure 3 As shown in Table 1, the dynamic change graph of the execution trimming action value provided in this application example demonstrates the synchronous change relationship between the execution trimming action value and the runtime self-calibration value under different transport cycles. The bar chart represents the execution trimming action value corresponding to each transport cycle, reflecting the comprehensive cumulative level of the robot's motion execution load within the transport cycle; the line chart represents the runtime self-calibration value, depicting the immediate representation of the intensity of strategy updates by runtime feedback. By using a dual-axis approach, the two types of values ​​are presented side-by-side in the same time series, making the linkage between changes in execution layer load and changes in runtime feedback more intuitive and clear. (See Table 1 and...) Figure 3 It can be seen that the execution trimming effect is highest in transport cycle six, indicating that the combined load of the robot's actions reaches its peak under the combined influence of factors such as path length, motion intensity, and posture adjustment within this transport cycle. The execution trimming effect is lowest in transport cycle one, indicating that the robot's motion path is shorter and its motion state is relatively stable within this cycle, resulting in a lower execution load. Simultaneously, the runtime self-calibration value also reaches a high level in transport cycle six, while it is at its lowest in transport cycle one, reflecting the synchronous trend between runtime feedback intensity and execution load changes. The dynamic change graph of the execution trimming effect intuitively reflects the characteristics of the robot's execution trimming effect and runtime self-calibration value co-evolving with changes in runtime state throughout continuous transport cycles, providing intuitive data support for scheduling strategy updates based on runtime feedback.

[0057] Specifically, the data update steps for the strategy network driven by the comprehensive evaluation results are as follows: compare the current pruning action amount with the pruning action threshold in real time. The pruning action threshold includes a first pruning threshold and a second pruning threshold, wherein the first pruning threshold is greater than the second pruning threshold. During the comparison process, the motion safety status indicator and the obstacle avoidance status indicator are checked simultaneously. Subsequent pruning-related processing is only performed under the condition that there is no collision risk, emergency obstacle avoidance trigger, or forced braking constraint. When the amount of clipping action is less than or equal to the second clipping threshold, the actual walking trajectory, speed change curve, angular velocity change curve, and empty path length of the corresponding conveying action sequence are written into the data acquisition buffer of the strategy network according to the complete conveying cycle. The equipment waiting time, equipment receiving completion time and robot material release time within the conveying cycle are bound and recorded. In the subsequent strategy network update process, they are used as high-confidence running sample inputs in a way that accounts for a fixed proportion of all valid samples. This proportion automatically remains stable as the total number of samples changes. When the pruning action is greater than the second pruning threshold and less than or equal to the first pruning threshold, the number of path replanning events, the duration of acceleration / deceleration segments, the cumulative length of idle segments, and the actual receiving delay of the corresponding equipment in the current transport cycle are written into the data acquisition buffer of the policy network. In the next policy network update, the samples participate in the iterative calculation of policy parameters in proportion to complement the high-confidence samples, so that samples under different operating conditions are continuously included in the update process. When the amount of clipping exceeds the first clipping threshold, provided that the current motion state has not triggered obstacle avoidance braking constraints and there is no safety risk, the pending actions in the current action sequence that have not yet been sent to the motion control interface are terminated. The pose trajectory segments that have occurred, the length of the incomplete path segments, the current device receiving status, and the environmental congestion indicators are synchronously written into the critical event data area of ​​the policy network. In each subsequent round of policy network update, the data participates as a mandatory sample in the update of policy parameters, thereby ensuring that the abnormal operating state has a continuous impact in the policy update.

[0058] In this implementation plan, by performing a hierarchical comparison between the pruning action and the pruning threshold, the transport execution information generated under different operating conditions is collected separately according to stability and risk level. Without violating motion safety constraints, the operating data is guided into the strategy update process in an orderly manner. This allows data from the stable operation phase to continuously enhance scheduling stability, data from the fluctuating operation phase to participate in scheduling adjustments, and data from the abnormal load phase to continuously constrain subsequent decisions, thereby achieving long-term adaptation of the strategy update process to changes in the actual operating state.

[0059] The second aspect of this invention provides a single-robot intelligent scheduling system for material handling based on dynamic priority learning, comprising: a multi-source state perception module, used to synchronously collect relevant data on changes in equipment receiving state, changes in robot walking path, and changes in loading and unloading action feedback during the material handling process. The receiving constraint data is used to characterize the permission status of each device for material entry at the current moment, the motion trajectory data is used to depict the actual movement path and posture changes of the robot in space, and the operation feedback data is used to reflect the time consumption, empty walking status, and action execution rhythm formed after the material handling is completed. The above data is processed by timestamp alignment, abnormal fragment removal, and scale unification, so that data from different sources form a continuous and consistent state expression under the same time reference, thereby constructing a standardized material handling state dataset.

[0060] The dynamic priority matrix generation module is used to quantitatively analyze the resource consumption caused by a single robot performing different transport tasks at the current moment, after obtaining a standardized transport status dataset, by combining the demand intensity of each device for continuous material supply during the production process and the changes in the robot's current pose, speed, and orientation. The analysis results are then mapped into a priority expression in matrix form, so that each device forms a clear transport order within the current scheduling cycle, thereby realizing dynamic updates of priority as the operating status changes.

[0061] The self-updating trade-off mapping module is used to compare and analyze the operational feedback relationship between the robot's actual motion consumption and the equipment's real-time load performance during continuous transport, based on a standardized transport state dataset. By comprehensively evaluating the waiting changes, idle changes, and load changes generated by different scheduling sequences in actual operation, the stability of the scheduling sequence is continuously characterized. Based on the evaluation results, the numerical spacing between the corresponding scheduling values ​​of each equipment is continuously adjusted, so that the priority distribution gradually and adaptively evolves with the changes in the operation stage.

[0062] The load-driven strategy update module is used to comprehensively analyze the load level formed by the executable action sequence in actual operation after obtaining the operation evaluation results, combined with the path unfolding and motion execution of the robot when performing the transport action. The analysis results are used as the data update basis for the strategy network, so that subsequent scheduling decisions can be iteratively adjusted according to the actual load changes during continuous operation.

[0063] like Figure 4The diagram illustrates the dynamic update and execution flow of single-robot transport priorities provided in the application example, showcasing a priority update and execution process for single-robot transport scheduling. The process begins at the start node with a decision branch, triggering subsequent processing by detecting whether a new transport task has been triggered. When a new task is triggered, the process enters the state data acquisition phase, gathering current operational information. When no new task is triggered, the process maintains the current scheduling queue to ensure the continuous execution of the action sequence. After completing state data acquisition, the process enters the basic priority calculation phase, calculating the initial priority order of each candidate transport object under the current operating conditions. Subsequently, it enters the learning and correction phase, adjusting the priority ranking results based on operational feedback to ensure the priority order reflects the actual load and capacity of the current operational phase. After correction, the process generates the final priority result and sends the corresponding scheduling instructions to the robot execution end, completing a full scheduling update loop. The overall process starts with task triggering and ends with priority generation and issuance, demonstrating the logical structure of dynamically updating the transport priority order based on real-time state changes during operation.

[0064] In this implementation scheme, the role of the multi-source state perception module is to synchronously collect and uniformly process the changes in the receiving state of the equipment, the changes in the spatial movement trajectory of the robot, and the operational feedback performance during the conveying process. This ensures that the state information from different sources remains consistent in terms of time reference and expression scale, thereby providing a continuous, comparable, and physically consistent state input basis for subsequent scheduling analysis.

[0065] The dynamic priority matrix generation module is designed to quantitatively analyze the resource occupancy formed by the continuous operation requirements of equipment and the current physical operation status of the robot based on the transport status data, and to transform the analysis results into a matrix-based priority order expression, so that each device can form a clear and dynamically adjustable transport order relationship within the current scheduling cycle.

[0066] The self-updating trade-off mapping module is used to continuously evaluate the stability of scheduling and ranking during continuous operation by using the operational feedback relationship formed between the robot's actual motion consumption and the real-time load performance of the equipment. Based on this, it gradually corrects the relative spacing between the scheduling values ​​of each equipment, so that the priority distribution maintains adaptive evolution characteristics as the operation stage changes.

[0067] The role of the load-driven strategy update module is to combine the robot's path deployment state and motion execution state to comprehensively evaluate the load level formed by the executable action sequence in actual operation, and transform the evaluation results into the data update basis of the strategy network, so that the scheduling strategy can be continuously optimized and adjusted according to the real load changes during continuous operation.

[0068] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0069] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A single-robot intelligent scheduling method for material handling based on dynamic priority learning, characterized in that: include: S1: Collect receiving constraint data, motion trajectory data and operation feedback data during the transfer process, and preprocess the collected receiving constraint data, motion trajectory data and operation feedback data to construct a standardized transfer status dataset; The specific steps for receiving constraint data, motion trajectory data, and operation feedback data during the acquisition and transfer process are as follows: The data collected during the operation of the equipment includes the following constraints: the quantity of material being processed by each piece of equipment, the number of available channels at the material inlet, the number of workstations actually performing actions, and the start and end times of the conveying actions. The robot's motion trajectory data during its walking process is collected. The motion trajectory data includes: the robot's real-time pose coordinates, chassis size parameters, travel distance, idle time, direction adjustment angle, instantaneous passage delay value, instantaneous path length increment, instantaneous walking speed, instantaneous acceleration, instantaneous angular velocity, the angle between the robot's current orientation and the target device's receiving direction, as well as the spatial position of each fixed device in the environment, the passable boundary of the passage area, and the real-time obstacle position. Collect operational feedback data during a single transport closed-loop process. The operational feedback data includes: remaining time for the equipment to operate continuously, waiting time, and the number of times the delivery task has been completed. S2, based on a standardized transport status dataset, performs scheduling analysis on the resource occupancy of a single robot performing a transport task at the current moment, considering the continuous operation requirements of the equipment and the current physical operating status of the robot, and dynamically adjusts the transport priority order of each device based on the scheduling analysis results. The specific steps for scheduling and analyzing the resource occupancy of a single robot performing a transport task at the current moment, based on the standardized transport status dataset and considering the continuity requirements of equipment operation and the current physical operating state of the robot, are as follows: Multiply the remaining sustainable operating time of the equipment by the length of the unloaded feasible path from the robot to the target equipment, add one and take the natural logarithm to obtain the time path saturation mapping value; The robot's instantaneous passage delay value is added to the angle between the robot's current orientation and the target device's receiving direction to obtain the passage attitude deflection composite value; the arctangent operation is performed on the passage attitude deflection composite value to obtain the passage impedance mapping value. Multiply the number of available channels at the material inlet of the target equipment by the number of workstations where the target equipment is actually performing actions, and then add the value of the quantity of material being processed by the target equipment to obtain the real-time processing capacity value of the equipment; perform a square root operation on the real-time processing capacity value of the equipment to obtain the equipment receiving capacity mapping value. The instantaneous scheduling value is obtained by dividing the product of the time path saturation mapping value and the passage impedance mapping value by the device receiving capability mapping value. S3, based on a standardized transport state dataset, evaluates the stability of scheduling by considering the operational feedback relationship between the robot's actual motion consumption and the real-time load performance of the equipment, and continuously adjusts the numerical spacing between the scheduling values ​​of each equipment based on the operational evaluation results. S4 takes the operational evaluation results as input, comprehensively evaluates the operational load of the executable action sequence from the robot's path unfolding state and motion execution state, and drives the data update of the policy network based on the comprehensive evaluation results.

2. The intelligent scheduling method for single-robot transport based on dynamic priority learning according to claim 1, characterized in that: The specific steps for preprocessing the collected receiving constraint data, motion trajectory data, and operation feedback data to construct a standardized transport status dataset are as follows: For received constraint data, the timestamp sorting algorithm is used for alignment according to the timestamp order, and the adjacent consistency deduplication algorithm is used to delete redundant records for records that are continuously and repeatedly reported and whose status remains unchanged; for motion trajectory data, the sampling records that have abnormal jumps are deleted by the adjacent sampling point displacement continuity detection algorithm. The event deduplication and aggregation algorithm is executed on the records that appear repeatedly in the operation feedback data to delete invalid records with missing key timestamps; The received constraint data, motion trajectory data and operation feedback data after standardization are normalized by the range normalization algorithm to construct a standardized transport status dataset. Based on the robot's real-time pose coordinates, chassis size parameters, spatial positions of fixed devices in the environment, environmental boundary information, and real-time obstacle positions output by real-time obstacle detection sensors, the environmental space is discretized in a unified coordinate system, and the robot chassis shape is geometrically expanded according to the chassis size parameters to determine the impassable areas. Perform connectivity analysis on the remaining spatial region to form a set of nodes composed of passable regional units. Establish node connectivity relationships based on whether there are continuous passageways between regions to construct a passable spatial topology map. In the traversable space topology map, the robot's real-time pose coordinates are used as the starting node and the target device's spatial location is used as the target node. The length of the path edge is accumulated through a feasible path search algorithm to obtain the length of the unloaded feasible path from the robot to the target device. At the same time, the instantaneous unloaded path length increment at the corresponding moment is recorded.

3. The intelligent scheduling method for single-robot transport based on dynamic priority learning according to claim 1, characterized in that: The specific steps for dynamically adjusting the transport priority order of each device based on the scheduling analysis results are as follows: The instantaneous scheduling values ​​corresponding to each device are mapped to a fixed-length vector structure according to the device number order. The scheduling time is used as the row index, and the corresponding vector is written into the current row position of the matrix data structure to form a dynamic priority matrix that represents the scheduling priority relationship of each device at the current time. The number of columns in the matrix corresponds one-to-one with the number of devices, and the number of rows in the matrix increases with the scheduling time. After writing the data for the current row of the dynamic priority matrix, a pairwise comparison-based sorting calculation is performed on the instantaneous scheduling values ​​stored in each column of the current row. The sorting process generates a sequence of devices arranged in ascending order of instantaneous scheduling values ​​through a sorting process in which the number of comparisons increases logarithmically with the number of devices. The device corresponding to the first position in the sorting result is then determined as the target device for the next transport task. After the robot completes the current transport and releases the material, the latest row of data in the dynamic priority matrix is ​​read again, and the sorting process is repeated to determine the subsequent transport target. When the number of available channels of the equipment changes, the quantity of materials being processed by the equipment changes, the instantaneous passage delay of the robot changes, and the real-time pose coordinates of the robot change, the instantaneous scheduling value of the corresponding equipment is recalculated, and the updated value is written into the corresponding column position of the current row of the dynamic priority matrix. At the same time, the current row of the dynamic priority matrix is ​​re-sorted. When the sorting result changes and the robot has not yet issued the next transport instruction, the updated first device in the sorting is used as the new transport target to generate the path and issue the execution instruction; when the robot is already performing a transport action, the next target is determined based on the latest sorting result after the current transport is completed, and the already executed motion instruction is not interrupted.

4. The intelligent scheduling method for single-robot transport based on dynamic priority learning according to claim 1, characterized in that: The specific steps for evaluating the scheduling stability based on the standardized transport status dataset and the operational feedback relationship between the robot's actual motion consumption and the equipment's real-time load-bearing performance are as follows: A conveying cycle is defined as the continuous operating time from when the robot completes one material delivery and releases the material until it completes the next material delivery and releases the material again. The robot's travel distance within a transport cycle is cumulatively calculated to obtain the cumulative travel distance value; the equipment's waiting time within a transport cycle is cumulatively calculated to obtain the cumulative equipment waiting time value. The robot's idle time within a transport cycle is accumulated to obtain the cumulative value of the robot's idle time; the number of available channels of the equipment within a transport cycle is accumulated to obtain the cumulative value of the number of available channels on the equipment side. The robot's directional adjustment angles are cumulatively calculated within a transport cycle to obtain the cumulative value of the directional adjustment angles; the number of times a delivery task is completed within a transport cycle is cumulatively calculated to obtain the total number of completed delivery tasks. Perform an inverse hyperbolic sine operation on the ratio of the cumulative distance the robot moves during the current transport cycle to the cumulative waiting time on the equipment side to obtain the idle displacement waiting mapping value; Perform an inverse hyperbolic sine operation on the ratio of the cumulative idle time of the robot in the current conveying cycle to the cumulative number of available channels on the equipment side to obtain the idle receiving mapping value; The posture task mapping value is obtained by dividing the cumulative value of the robot's direction adjustment angle in the current delivery cycle by the ratio of the total number of completed delivery tasks, adding one to the ratio, and then taking the natural logarithm. The idle displacement waiting mapping value, the idle reception mapping value, and the attitude task mapping value are added together to obtain the running state self-calibration value.

5. The intelligent scheduling method for single-robot transport based on dynamic priority learning according to claim 1, characterized in that: The specific steps for continuously adjusting the numerical intervals between the scheduling values ​​of each device based on the operational evaluation results are as follows: After completing the calculation of the running self-calibration value for the current transport cycle, the sorting result of the instantaneous scheduling value corresponding to each device in the current row of the dynamic priority matrix remains unchanged. The difference between the instantaneous scheduling values ​​of adjacent devices is calculated for the sorted instantaneous scheduling values ​​to obtain the adjacent scheduling difference. The adjacent scheduling difference is then continuously proportionally mapped to the running self-calibration value to obtain the updated adjacent scheduling difference. While keeping the instantaneous scheduling value corresponding to the first device unchanged, the current row of the dynamic priority matrix is ​​reconstructed sequentially based on the updated adjacent scheduling difference, so that the numerical spacing between the instantaneous scheduling values ​​of each device in the matrix changes synchronously with the running state self-calibration value. In subsequent transport decisions, the device sorting results are read based on the reconstructed dynamic priority matrix. When the adjacent scheduling difference between adjacent devices is amplified N times consecutively, the sorting results remain consistent within N transport cycles. When the adjacent scheduling difference between adjacent devices is compressed N times consecutively, the sorting results are allowed to exchange positions within N transport cycles.

6. The intelligent scheduling method for single-robot transport based on dynamic priority learning according to claim 1, characterized in that: The specific steps for comprehensively evaluating the operational load of the executable action sequence based on the robot's path deployment state and motion execution state, using the operational evaluation results as input, are as follows: Multiply the robot's instantaneous path length increment by its instantaneous walking speed to obtain the path motion flux value. The instantaneous acceleration of the robot is squared to obtain the acceleration and deceleration inertial dissipation value; The square of the robot's instantaneous angular velocity is used to obtain the steering attitude dissipation value; The path motion flux, acceleration / deceleration inertia dissipation, and steering attitude dissipation are added to the robot's instantaneous idle path length increment to obtain the basic path motion cost. Perform an exponential operation on the natural constant e with the running state self-calibration value as the exponent to obtain the running state self-calibration modulation value; Multiply the basic cost of path motion by the self-calibration modulation value of the running state to obtain the instantaneous cost after modulation; Within the time interval from the start to the end of the current conveying action, the modulated instantaneous value is integrated over time to obtain the execution clipping effect value.

7. The intelligent scheduling method for single-robot transport based on dynamic priority learning according to claim 1, characterized in that: The specific steps for updating the data in the policy network driven by the comprehensive evaluation results are as follows: The current pruning action is compared in real time with the pruning action threshold, which includes a first pruning threshold and a second pruning threshold, wherein the first pruning threshold is greater than the second pruning threshold. When the amount of clipping action is less than or equal to the second clipping threshold, the actual walking trajectory, speed change curve, angular velocity change curve and empty path length of the corresponding conveying action sequence are written into the data acquisition buffer of the strategy network according to the complete conveying cycle. The equipment waiting time, equipment receiving completion time and robot material release time within the conveying cycle are bound and recorded. In the subsequent strategy network update process, these are used as high-confidence running sample inputs with a fixed proportion of no less than all samples. When the amount of pruning action is greater than the second pruning threshold and less than or equal to the first pruning threshold, the number of path replanning events, the duration of acceleration and deceleration segments, the cumulative length of idle segments, and the actual receiving delay of the corresponding devices in the current transport cycle are written into the data acquisition buffer of the policy network, and are proportionally extracted to participate in the iterative calculation of policy parameters during the next policy network update. When the pruning action exceeds the first pruning threshold, the pending actions in the current action sequence that have not yet been sent to the motion control interface are terminated. The pose trajectory segments that have occurred, the length of the incomplete path segments, the current device receiving status, and the environmental congestion indicators are synchronously written into the critical event data area of ​​the policy network, and participate in the policy parameter update as a mandatory sample in each subsequent round of policy network update.

8. A single-robot intelligent scheduling system for material handling based on dynamic priority learning, employing the single-robot intelligent scheduling method for material handling based on dynamic priority learning as described in any one of claims 1-7, characterized in that: include: The multi-source state perception module is used to collect receiving constraint data, motion trajectory data and operation feedback data during the transfer process, and to preprocess the collected receiving constraint data, motion trajectory data and operation feedback data to construct a standardized transfer state dataset. The dynamic priority matrix generation module is used to perform scheduling analysis on the resource occupancy of a single robot performing a conveying task at the current moment, based on the standardized conveying status dataset, considering the continuous operation requirements of the equipment and the current physical operating status of the robot. Based on the scheduling analysis results, the module dynamically adjusts the conveying priority order of each device. The self-updating trade-off mapping module is used to evaluate the scheduling order stability based on the standardized transport state dataset, from the operational feedback relationship between the robot's actual motion consumption and the real-time load performance of the equipment, and continuously adjust the numerical spacing between the scheduling values ​​of each equipment based on the operational evaluation results. The runtime load-driven strategy update module takes the runtime evaluation results as input, performs a comprehensive evaluation of the runtime load of the executable action sequence from the robot's path unfolding state and motion execution state, and drives the data update of the strategy network based on the comprehensive evaluation results.

Images