Multi-device cooperative industrial logistics robot cluster scheduling and path optimization management and control system

Abstract

The application belongs to the technical field of industrial logistics, and particularly relates to a multi-device coordinated industrial logistics robot cluster scheduling and path optimization management and control system, which comprises a vehicle-mounted subsystem arranged on each robot body and a central arbitration module arranged on a central scheduling layer; the motion intention broadcast module comprises broadcasting local motion intention information to other robots within a communication range through a vehicle-to-vehicle communication link at a preset frequency; spatiotemporal overlap detection is carried out in combination with the local intention perception atlas, a conflict severity index is calculated, and risk classification is carried out. The prepositioning and active information interaction mode can predict the path operation situation of the robot cluster in advance, does not need to rely on real-time detection of the field conditions by sensors, greatly improves the early warning amount of robot conflict perception, reduces the probability of sudden collision and path interference from the source, effectively makes up for the lagging defect of the traditional post-perception mechanism, and reserves sufficient scheduling processing time for subsequent conflict detection and elimination.

Description

Technical Field

[0001] This invention belongs to the field of industrial logistics technology, specifically a multi-device collaborative industrial logistics robot cluster scheduling and path optimization management system. Background Technology

[0002] With the deepening of intelligent manufacturing and Industry 4.0, industrial logistics robots have been widely used in scenarios such as warehousing and sorting, production line material distribution, and workshop turnover. Single robot operations are evolving towards multi-robot cluster collaborative operations. In large-scale cluster scenarios, there are many robots, high operation density, and frequent path intersections. How to achieve efficient scheduling and path optimization at the cluster level has become a core issue restricting the overall throughput capacity and operational safety of the logistics system.

[0003] Existing industrial logistics robot cluster scheduling technologies are mainly divided into two categories: one is a centralized scheduling architecture, in which a central scheduling server uniformly plans the global path of all robots and issues motion commands; although this method can guarantee global optimization, it has the drawbacks of heavy computational burden on the central node, sensitivity to communication latency, and high risk of single point of failure. When the number of robots increases dramatically, the scheduling response latency increases significantly, making it difficult to meet the real-time requirements of high-density operation scenarios. The other is a purely distributed obstacle avoidance architecture, in which each robot relies only on onboard sensors for local perception and independent collision avoidance, lacking interaction of motion intentions and collaborative decision-making mechanisms between robots; although this method has better real-time performance, it is prone to multiple robots getting stuck in a deadlock or mutual yielding in areas such as intersections and narrow passages, and local optimal decisions often lead to low global path efficiency and congestion spread.

[0004] Existing technologies have the following shortcomings in conflict resolution mechanisms: First, existing industrial logistics robot clusters lack proactive interaction and spatiotemporal alignment mechanisms for robot movement intentions, and conflict detection relies on post-event onboard sensor perception, resulting in insufficient early warning. Second, existing cluster conflict resolution strategies are simplistic, failing to classify risks and differentiate treatment based on conflict severity. In low-risk scenarios, over-reliance on negotiation leads to scheduling efficiency losses, while in high-risk scenarios, insufficient distributed negotiation capabilities prevent timely conflict resolution. Third, existing distributed negotiation lacks an effective hierarchical connection mechanism between it and central arbitration, preventing negotiation deadlocks from escalating to central arbitration in a timely manner, which can easily lead to multiple robots getting stuck in deadlocks at intersections, narrow passages, and other areas for extended periods. Furthermore, the negotiated passage order and time window constraints lack closed-loop execution methods embedded in the local path planner. Therefore, to address these issues, a multi-device collaborative industrial logistics robot cluster scheduling and path optimization management system is proposed. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies and solve at least one of the technical problems mentioned in the background, this invention proposes a multi-device collaborative industrial logistics robot cluster scheduling and path optimization management system.

[0006] The technical solution adopted by the present invention to solve its technical problem is: the multi-device collaborative industrial logistics robot cluster scheduling and path optimization control system of the present invention includes a vehicle-mounted subsystem deployed on each robot body and a central arbitration module deployed on the central scheduling layer. The vehicle-mounted subsystem includes: a motion intent broadcasting module, an intent receiving and parsing module, a local conflict autonomous detection module, a distributed right-of-way negotiation module, and a negotiation commitment execution module; The motion intent broadcasting module includes broadcasting its own motion intent information to other robots within the communication range at a preset frequency via a vehicle-to-vehicle communication link; The intent receiving and parsing module interacts with the motion intent broadcasting module, including receiving motion intent information broadcast by neighboring robots, parsing and spatiotemporally aligning multi-source intent information, and generating a local intent perception map. The intent receiving and parsing module interacts with the motion intent broadcasting module, receives motion intent information broadcast by the neighboring robot, parses and performs spatiotemporal alignment processing on the multi-source intent information, generates a local intent perception map and outputs it to the local conflict autonomous detection module. The local conflict autonomous detection module is connected to the intent receiving and parsing module. It retrieves the local motion intent information, combines it with the local intent perception map to perform spatiotemporal overlap detection, calculates the conflict severity index and performs risk classification, and outputs the conflict identification result, which includes the conflict location, conflict duration, the robot involved, and the conflict risk level, to the distributed right-of-way negotiation module. The distributed right-of-way negotiation module is connected to the local conflict autonomous detection module. It executes a differentiated negotiation triggering strategy based on the conflict risk level: when the conflict risk level is low, it generates a local deceleration and avoidance command and sends it directly to the negotiation commitment execution module; when the conflict risk level is medium, it establishes a point-to-point negotiation session with all involved robots, negotiating the passage order based on the robot's task urgency, remaining battery power, load status, and path occupancy duration. If the negotiation reaches an agreement, the passage order and time window constraints are output to the negotiation commitment execution module; if the negotiation is deadlocked, an arbitration request is initiated to the central arbitration module; when the conflict risk level is high, it directly initiates an arbitration request to the central arbitration module. The negotiation commitment execution module is connected to the distributed right-of-way negotiation module. It receives deceleration and avoidance instructions or negotiated passage order and time window constraints, and embeds the constraints into the local path planner to drive the robot to perform path tracking under time constraints. The central arbitration module communicates with the distributed right-of-way negotiation module of each robot, receives arbitration requests sent by the distributed right-of-way negotiation module, and issues arbitration passage instructions to the robot involved when point-to-point negotiation fails to reach a consensus or high-risk conflicts are directly reported.

[0007] Preferably, the motion intention information includes the planned path, speed curve, task urgency, load status, and the expected time window to reach each path node; the motion intention broadcasting module broadcasts a broadcast message carrying the motion intention information to other robots within the communication range via a vehicle-to-vehicle communication link at a preset frequency, and the header of the broadcast message includes a unique robot identifier, a message sequence number, and a global timestamp.

[0008] Preferably, the vehicle-to-vehicle communication link is configured with a communication fault tolerance and message completion mechanism, and a formula for determining the effective communication status is defined: ; in For message transmission delay, To preset the maximum allowable communication delay, For indicator functions, This serves as a valid communication identifier; when communication latency exceeds limits or packet loss is detected, a local historical intent data completion and short-term communication reconnection mechanism is automatically triggered, based on the most recent... The historical intent data of the frames is supplemented by linear extrapolation to complete the path nodes and time windows of the missing frames, and the supplemented data is marked as low confidence and input into the intent receiving and parsing module to avoid the loss of intent perception and negotiation interruption caused by communication fluctuations.

[0009] Preferably, the intent receiving and parsing module is configured with a multi-source intent spatiotemporal alignment algorithm, uniformly using the system's global clock as a reference to complete timestamp calibration, and mapping the path data of each robot's local coordinate system to the global coordinate system through a coordinate transformation matrix, satisfying: ; in For global coordinate vectors, This is the robot's local coordinate vector. Let be a rotation matrix. The translation vector is used. After eliminating the local clock deviation and coordinate reference difference of multiple robots, the path trajectory, running speed and node time window data of all neighboring robots are normalized and aligned. Using the robot's unique identifier as an index, the path space interval and time sequence window parameters are associated and matched to construct a structured local intent perception map. Each intent record in the local intent perception map carries a confidence label. The supplementary data from the communication fault tolerance module is marked as low confidence, and the data from real-time broadcast is marked as high confidence.

[0010] Preferably, the local conflict autonomous detection module is configured with a spatiotemporally coupled conflict severity grading model, which quantifies the path space overlap ratio, time window overlap ratio, and robot relative speed factor, and defines the conflict severity index calculation formula: ; in As a conflict severity index, This represents the area of ​​the overlapping region in the path space. This refers to the footprint of a single machine. The duration of time window overlap. This represents the width of the time window for the corresponding path node. The relative speed of the robot involved. The maximum permissible speed of the system. , , The weighted coefficients are the spatial, temporal, and velocity dimensions and satisfy the following conditions: ; Preset two-level risk thresholds , and Risk classification is performed according to the following rules: When When it is judged as low risk, When it is determined to be medium risk, The path is identified as high-risk; CSI calculation is initiated only when there is an intersection in the path space and the time window overlap duration is greater than zero, eliminating pseudo-collision interference where there is no spatial intersection or the time window is completely misaligned.

[0011] Preferably, the distributed right-of-way negotiation module constructs a multi-dimensional priority quantification algorithm with nonlinear amplification of urgency and a four-stage negotiation state machine; The four core indicators—task urgency, remaining battery power, load status, and path occupancy time—are normalized to a range of 0-1. A nonlinear weighted fusion model is then used to calculate the robot's passage priority score. ; in For the first Priority score for each robot , , , These are the normalized task urgency, remaining battery power, load status, and path occupancy time metrics. This is an amplified item for the task urgency index. For preset magnification factor and >0, , , , For scene-adaptive dynamic weights and satisfying + + + = 1; The four-stage negotiation state machine includes: a proposal stage, in which each involved robot broadcasts its own... In the proposed passage time window phase, the robot with lower priority adjusts its proposed passage time window based on the highest priority in its neighborhood and rebroadcasts it. In the commitment phase, when the proposed passage time windows of all robots do not overlap in two consecutive rounds of interaction, each robot sends a commitment message. In the confirmation phase, after receiving commitment messages from all involved robots, a unified passage sequence is generated. A fixed number of negotiation iteration rounds is preset. ,achieve If the negotiation has not yet reached the commitment stage, it is considered a deadlock and an arbitration request will be automatically initiated to the central arbitration module.

[0012] Preferably, the scene-adaptive dynamic weights have a real-time iterative update mechanism, constructing a dynamic weight adjustment model with saturation constraints: ; in For the updated version Item weight, For the weights before the update, Adjusting the step size for fixed weights and , For the first The scenario adjustment coefficients corresponding to each indicator are obtained by linearly fusing three types of scenario features: factory congestion density, proportion of urgent task batches, and proportion of low-battery robots. ; in To account for the density of traffic congestion in the factory area, The percentage of urgent batches. For the proportion of low-power robots, , , These are the preset scene feature mapping coefficients; the path occupancy time corresponding to high congestion scenarios. Increase thus enlarge In emergency mission scenarios, the urgency level corresponds to... Increase thus enlarge After the update, the system is normalized to meet the requirements again. .

[0013] Preferably, the negotiation commitment execution module is configured with a time-constraint closed-loop verification algorithm, which embeds the negotiated passage order and node time window constraints as hard constraint parameters into the local path planner, and iteratively generates a smooth tracking path that satisfies the time constraints; the time deviation calculation formula is defined as follows: ; in This is the timing deviation value. The actual arrival time of the robot. Set a timeframe for negotiation; set a tolerance threshold for deviations. ,when If the timing of path tracking fails, the local conflict detection module is restarted to conduct a second conflict detection. If a medium-risk or higher conflict still exists after the second detection, the distributed right-of-way negotiation module is triggered to conduct a second negotiation to achieve dynamic closed-loop error correction. When the timing deviation is caused by communication interruption, the communication fault tolerance and completion mechanism is triggered first before the second conflict detection is performed.

[0014] Preferably, the central arbitration module is configured with a globally optimal arbitration decision algorithm containing fairness constraints. After receiving an arbitration request, it simultaneously collects the priority score of the robot involved, the global road network congestion occupancy rate, the cluster task time slack, and the number of times each robot has been suppressed by arbitration in the past, and constructs a multi-objective optimized arbitration decision model: ; in The single-robot negotiation priority loss function is defined as the ratio of the sum of priority scores of the robots suppressed by arbitration to the total priority scores of all involved robots. The cluster global scheduling latency loss function is defined as the ratio of the sum of the additional waiting time of all involved robots under the arbitration scheme to the baseline scheduling time. The arbitration fairness loss function is defined as the range of the number of times each robot has been suppressed by arbitration in its history: ; in For the first The number of times Taiwan's robot history has been suppressed by arbitration; , , To balance the weights and satisfy By solving the minimization problem The globally optimal passage sequence scheme is obtained, and a standardized arbitration passage instruction is generated and synchronously issued to all involved robots, forcibly covering the results of local negotiation disagreements, serving as a rigid constraint for unified execution of the cluster.

[0015] Preferably, the central arbitration module has a Bayesian-optimized iterative archiving and strategy optimization mechanism for arbitration results. After a single arbitration execution is completed, it automatically collects and stores data on conflict type, reasons for negotiation deadlock, arbitration instruction parameters, robot execution time, actual scheduling delay, and fairness indicators; and constructs a system based on balanced weights. , , To optimize variables, the average loss of arbitration events within a sliding window is used. For the Bayesian optimization problem of the objective function, a Gaussian process is used to model the mapping relationship between the weight vector and the average loss, and the acquisition function is improved based on the expected value: ; in For the weight vector, This is the current optimal weight vector; each accumulated... After each arbitration event, a weight update iteration is performed, and the weight vector that maximizes the expected improvement is output as the balancing weight of the arbitration decision model in the next stage.

[0016] The beneficial effects of this invention are: This invention provides a multi-device collaborative industrial logistics robot cluster scheduling and path optimization management system. Through vehicle-to-vehicle communication links, it proactively broadcasts its own motion intention information at a preset frequency, while simultaneously receiving multi-source motion intention data from neighboring robots in real time. It then performs unified parsing and spatiotemporal alignment of this multi-source information to accurately construct a local intention perception map. This proactive and preemptive information interaction method can predict the path operation status of the robot cluster in advance, eliminating the need for real-time sensor monitoring of the field conditions. This significantly increases the early warning rate of robot conflict perception, reducing the probability of sudden collisions and path interference at the source. It effectively compensates for the lag in traditional post-event perception mechanisms, reserving sufficient scheduling time for subsequent conflict detection and resolution.

[0017] This invention provides a multi-device collaborative industrial logistics robot cluster scheduling and path optimization management system. Based on a local intent-aware graph, it performs spatiotemporal overlap detection, accurately calculates the conflict severity index, and classifies conflict risks. For different conflict levels, it matches specific handling strategies: for low-risk conflicts, it eliminates the need for multi-robot negotiation, directly resolving the conflict through local deceleration and avoidance commands, avoiding the redundant scheduling process of traditional indiscriminate negotiation and significantly improving cluster passage efficiency in low-risk scenarios; for medium-risk conflicts, it conducts point-to-point right-of-way negotiation based on multi-dimensional core parameters such as robot task urgency, remaining battery power, load status, and path occupancy time, ensuring the rationality and adaptability of the negotiation results; for high-risk conflicts, it directly skips the distributed negotiation process and initiates central arbitration, avoiding delays in conflict resolution due to ineffective negotiation in high-risk scenarios. Through a differentiated, layered processing mode, it balances the efficiency and security of cluster scheduling, completely solving the technical pain points of poor adaptability and the inability to balance efficiency and security in traditional single resolution strategies.

[0018] This invention provides a multi-device collaborative industrial logistics robot cluster scheduling and path optimization management system. By establishing a comprehensive distributed negotiation and escalation arbitration mechanism, when negotiation deadlocks occur in high-risk scenarios or distributed negotiations fail to reach a consensus, the conflict can be quickly escalated to the central arbitration module for intervention. This establishes a linkage channel between the underlying distributed scheduling and the top-level central scheduling, effectively preventing multiple robots from getting stuck in critical areas such as intersections and narrow passages for extended periods, thus ensuring the stability of traffic flow in industrial logistics scenarios. Simultaneously, through the negotiation commitment execution module, core conditions such as the agreed-upon passage order and time window constraints are directly embedded into the robot's local path planner, achieving a closed-loop management process of "intent perception - conflict detection - negotiation arbitration - constraint execution." This ensures that the scheduling results of negotiation and arbitration are implementable and executable, eliminating the problem of scheduling decisions being disconnected from actual robot operation, and significantly improving the accuracy, reliability, and collaborative operation efficiency of robot cluster path scheduling. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention.

[0020] In the attached diagram: Figure 1 This is a schematic diagram of the system flow of the present invention; Figure 2 This is a schematic diagram of the motion intent information module in this invention; Figure 3 This is a schematic diagram of the vehicle-to-vehicle communication link configuration module in this invention; Figure 4This is a schematic diagram of the intent receiving and parsing module in this invention. Detailed Implementation

[0021] 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.

[0022] Specific implementation examples are given below.

[0023] Please see Figures 1-4 This invention provides a multi-device collaborative industrial logistics robot cluster scheduling and path optimization management system. In the process of industrial logistics robot cluster scheduling, due to the surge in the number of robots and the increase in work density, the existing technology has the following defects: there is a lack of active interaction and spatiotemporal alignment mechanism between robots to determine their motion intentions; conflict detection relies on post-event on-board sensor perception, resulting in insufficient early warning lead time; the conflict resolution strategy does not classify risks according to the severity of the conflict, resulting in efficiency loss due to excessive reliance on negotiation in low-risk scenarios and insufficient distributed negotiation capabilities in high-risk scenarios; there is a lack of hierarchical connection mechanism between distributed negotiation and central arbitration, and negotiation deadlocks cannot be escalated to central arbitration in a timely manner, which can easily lead to multiple robots getting stuck at intersections or narrow passages.

[0024] For example, in a high-density e-commerce warehouse sorting area, multiple logistics robots simultaneously plan paths to perform delivery tasks through narrow passages. Due to communication delays, the motion intention information broadcast by each robot fails to align in time, and the local intention perception map shows that the paths of multiple robots overlap in time and space, with the conflict severity index indicating a medium risk. The distributed right-of-way negotiation module initiates point-to-point negotiation, but because the indicators of the robots involved are similar and there is a lack of a dynamic weight adjustment mechanism, the negotiation reaches a stalemate. The central arbitration module fails to receive the arbitration request in a timely manner, causing the robots to stop and yield to each other at the entrance of the passage, affecting the continuity of material flow.

[0025] If the above problems are not resolved, the deadlock and impasse will spread from local to the entire road network, causing global congestion, increasing the probability of collisions, increasing scheduling response latency, and ultimately limiting the large-scale deployment of industrial logistics automation systems.

[0026] To address this, this application proposes a multi-device collaborative industrial logistics robot cluster scheduling and path optimization management system, including an onboard subsystem deployed on each robot and a central arbitration module deployed at the central scheduling layer. The onboard subsystem includes: a motion intent broadcasting module, an intent receiving and parsing module, a local conflict autonomous detection module, a distributed right-of-way negotiation module, and a negotiation commitment execution module. This motion intent broadcasting module broadcasts its own motion intent information to other robots within the communication range at a preset frequency via the vehicle-to-vehicle communication link. The intent receiving and parsing module interacts with the motion intent broadcasting module, receives motion intent information broadcast by neighboring robots, parses and spatiotemporally aligns multi-source intent information, generates a local intent perception map, and outputs it to the local conflict autonomous detection module.

[0027] The local conflict autonomous detection module is connected to the intent receiving and parsing module. It retrieves the local motion intent information, combines it with the local intent perception map to carry out spatiotemporal overlap detection, calculates the conflict severity index and performs risk classification, and outputs the conflict identification results, including the conflict location, conflict duration, the robot involved, and the conflict risk level, to the distributed right-of-way negotiation module.

[0028] This distributed right-of-way negotiation module is connected to the local conflict autonomous detection module. It executes a differentiated negotiation triggering strategy based on the conflict risk level: when the conflict risk level is low, it generates a local deceleration and avoidance command and sends it directly to the negotiation commitment execution module; when the conflict risk level is medium, it establishes a point-to-point negotiation session with all involved robots, and conducts negotiation on the passage order based on the robot's task urgency, remaining battery power, load status, and path occupancy time. If the negotiation reaches an agreement, the passage order and time window constraints are output to the negotiation commitment execution module; if the negotiation is deadlocked, an arbitration request is initiated to the central arbitration module; when the conflict risk level is high, an arbitration request is directly initiated to the central arbitration module.

[0029] The negotiation commitment execution module is connected to the distributed right-of-way negotiation module. It receives deceleration and avoidance instructions or negotiated passage order and time window constraints, and embeds the constraints into the local path planner to drive the robot to perform path tracking under time constraints.

[0030] The central arbitration module communicates with the distributed right-of-way negotiation modules of each robot, receives arbitration requests sent by the distributed right-of-way negotiation modules, and issues arbitration passage instructions to the robots involved when point-to-point negotiation fails to reach a consensus or when high-risk conflicts are directly reported.

[0031] This system adopts a layered hybrid scheduling architecture: the onboard subsystem is responsible for local autonomous decision-making and distributed collaboration, while the central arbitration module provides global arbitration when distributed negotiation is blocked; the motion intent broadcasting module and the intent receiving and parsing module realize real-time sharing and spatiotemporal alignment of motion intents among robots, perceiving potential risks before conflicts occur, significantly improving the advance warning and accuracy of conflict warnings; the local conflict autonomous detection module calculates the conflict severity index and performs risk classification, and the distributed right-of-way negotiation module executes differentiated negotiation triggering strategies to avoid unnecessary complex negotiations in low-risk scenarios, while ensuring that medium- and high-risk conflicts are handled with precision; when distributed negotiation reaches a deadlock, it is promptly upgraded to the central arbitration module, which retains the real-time nature of distributed decision-making and ensures global optimality and system stability through central arbitration, making up for the shortcomings of pure distributed obstacle avoidance architectures that are prone to deadlock; the negotiation commitment execution module embeds the passage order and time window constraints into the local path planner to ensure that the negotiation or arbitration results are executed accurately. In summary, this system constructs a layered hybrid scheduling architecture, which significantly improves the scheduling efficiency, operational safety, and system robustness of industrial logistics robot clusters in complex and high-density scenarios.

[0032] In another implementation, if the motion intent information broadcast by the robot is too simple, other robots may be unable to accurately predict its future behavior, increasing the risk of misjudgment in conflicts. To address this, this application further proposes that: the motion intent information includes the planned path, speed curve, task urgency, load status, and the expected time window for reaching each path node; the motion intent broadcasting module broadcasts a broadcast message carrying the motion intent information to other robots within the communication range via a vehicle-to-vehicle communication link at a preset frequency, and the header of the broadcast message includes a unique robot identifier, a message sequence number, and a global timestamp; The planned path is represented by discrete path points or a continuous curve equation; the velocity curve is represented by a time-velocity function; the task urgency is a numerical or enumerated value; the load status is a Boolean value or a specific numerical value; and the estimated arrival time window is a time interval. , The broadcast message header includes a unique robot identifier to distinguish different robots, a message sequence number to detect loss or out-of-order delivery, and a global timestamp provided by GPS timing, NTP, or PTP to ensure consistent time references. Specifically, the planned path is represented by an array containing multiple 3D coordinate points, the speed curve by a piecewise function, the task urgency as an integer value from 1 to 5, the load status as a floating-point number (unit: kg), and the time window as an array of structures containing start and end timestamps. Communication uses a UDP-based broadcast method, implemented through a DSRC module supporting IEEE 802.11p or a 5G module, with a preset frequency of 10Hz. The message header includes a 64-bit MAC address as the unique robot identifier, a 16-bit unsigned integer as the message sequence number, and a 64-bit Unix timestamp (millisecond level). Path planning and speed curves enable robots to accurately predict the future trajectories and speed changes of other robots, improving the accuracy of local intent perception maps; task urgency, load status, and expected arrival time windows provide multi-dimensional decision-making basis for conflict detection and negotiation; unique identifiers, sequence numbers, and global timestamps in message headers ensure the reliability, timeliness, and traceability of information transmission, avoiding misjudgments due to missing or chaotic information, thereby optimizing cluster scheduling efficiency and security.

[0033] In another implementation, the vehicle-to-vehicle communication link may face problems such as unstable signals, fluctuating transmission delays, or packet loss, resulting in incomplete acquisition of intent information. To address this, this application further proposes: configuring a communication fault tolerance and packet completion mechanism for the vehicle-to-vehicle communication link, and defining a formula for determining the effective communication state: ; in For message transmission delay, To preset the maximum allowable communication delay, For indicator functions, This serves as a valid communication identifier; when communication latency exceeds limits or packet loss is detected, a local historical intent data completion and short-term communication reconnection mechanism is automatically triggered, based on the most recent... The frame history intent data is used to complete the path nodes and time windows of missing frames through linear extrapolation, and the completed data is marked as low confidence and input into the intent receiving and parsing module. During operation, the intent receiving and parsing module continuously monitors the communication link quality and performs real-time calculations. And determine whether the communication is valid based on the judgment formula. When the value is 0, the system triggers a local historical intent data completion and short-term communication reconnection mechanism. Using the most recent K frames of historical data, it estimates missing path nodes and time windows through linear extrapolation, marks them as low-confidence data, and inputs them along with real-time high-confidence data into the intent receiving and parsing module. Simultaneously, it attempts to restore the communication connection. Specific example: When robot B receives a message from robot A, the transmission latency is 150ms, preset... If the time limit is 100ms, the communication is deemed invalid. The system retrieves the most recent 3 frames of historical data from A and predicts the current path node and time window through linear extrapolation. The data is marked as "low confidence" and sent to the parsing module. At the same time, the system attempts to reconnect multiple times within 500ms. The communication fault tolerance and message completion mechanism ensures that the intent receiving and parsing module can continue to generate a relatively complete local intent-aware map even under communication fluctuations. This avoids the loss of intent awareness and negotiation interruption due to communication interruption, ensures the continuity of conflict detection and negotiation processes, and significantly improves the robustness and reliability of the system in complex industrial environments.

[0034] In another implementation, different robots may have local clock deviations and coordinate system differences. Directly fusing heterogeneous data can lead to spatiotemporal inconsistencies in the local intent perception map. To address this, this application further proposes that the intent receiving and parsing module be configured with a multi-source intent spatiotemporal alignment algorithm, uniformly using the system's global clock as a reference for timestamp calibration, and mapping the path data of each robot's local coordinate system to the global coordinate system through a coordinate transformation matrix, satisfying the following: ; in For global coordinate vectors, This is the robot's local coordinate vector. Let be a rotation matrix. This is a translation vector. After eliminating clock skew and coordinate reference differences, the path trajectories, running speeds, and node time window data of all neighboring robots are normalized and aligned. Using the robot's unique identifier as an index, the path spatial intervals and temporal occupancy window parameters are associated and matched to construct a structured local intent-aware map. Each intent record carries a confidence label; completed data is marked as low confidence, and real-time broadcast data is marked as high confidence.

[0035] Specific example: The system's global clock is provided by a central scheduling layer GPS synchronization server, and each vehicle subsystem synchronizes periodically via NTP. Robot A broadcasts its local coordinate system pathpoints. and time The parsing module utilizes a pre-stored coordinate transformation matrix. , (Obtained through site lidar scanning and self-positioning calibration) Convert to global coordinates (P_{A_global}), and simultaneously adjust according to NTP clock offset. Calibration The path trajectory is normalized to a discrete sequence of path points in a global coordinate system, the velocity is normalized to a 0-1 percentage of the maximum velocity, and the time window is unified to the start and end times of a global timestamp. The data is stored in a graph database with the robot's unique identifier as an index, forming a local intent-aware graph that includes spatial intervals, temporal occupancy windows, and confidence labels. A unified global clock and coordinate transformation matrix eliminate local clock deviations and coordinate reference differences among multiple robots, ensuring precise alignment of motion intent information in time and space, and improving the accuracy and reliability of the local intent perception map; confidence labels distinguish between real-time data and supplementary data, providing a refined decision-making basis for subsequent conflict handling.

[0036] In another implementation, to more accurately reflect the true severity of the conflict, the local conflict autonomous detection module is configured with a spatiotemporally coupled conflict severity grading model, which quantifies the path space overlap ratio, time window overlap ratio, and robot relative speed factor, and defines the conflict severity index calculation formula: ; in As a conflict severity index, This represents the area of ​​the overlapping region in the path space. This refers to the footprint of a single machine. The duration of time window overlap. This represents the width of the time window for the corresponding path node. The relative speed of the robot involved. The maximum permissible speed of the system. , , The weighted coefficients are the spatial, temporal, and velocity dimensions and satisfy the following conditions: .

[0037] Preset two-level risk thresholds , and Risk classification is performed according to the following rules: When When it is judged as low risk, When it is determined to be medium risk, The system is classified as high-risk; it will only be activated when there is an intersection in the path space and the time window overlaps for a duration greater than zero. The calculation eliminates pseudo-conflicts that are spatially non-intersecting or whose time windows are completely misaligned.

[0038] The multi-dimensional weighted fusion CSI index more accurately reflects the potential harm of conflict, overcoming the limitations of traditional single-dimensional or binary judgments, and avoiding overreaction to minor conflicts and underestimation of serious conflicts. The two-level threshold classification and pseudo-conflict pre-filtering mechanism improve detection accuracy and operational efficiency, providing a precise basis for the distributed right-of-way negotiation module to implement differentiated strategies.

[0039] In another implementation, to efficiently and fairly determine passage priorities and ensure negotiation convergence, the distributed right-of-way negotiation module constructs a multi-dimensional passage priority quantification algorithm with nonlinear amplification of urgency and a four-stage negotiation state machine. The four core indicators—task urgency, remaining battery power, load status, and path occupancy time—are normalized to a range of 0-1. A nonlinear weighted fusion model is then used to calculate the robot's passage priority score. ; in For the first Priority score for each robot , , , These are the normalized task urgency, remaining battery power, load status, and path occupancy time metrics. This is an amplified item for the task urgency index. For preset magnification factor and >0, , , , For scene-adaptive dynamic weights and satisfying + + + = 1; The four-stage negotiation state machine includes: the proposal stage, in which each involved robot broadcasts its own... In the proposed passage time window phase, the robot with lower priority adjusts its proposed passage time window based on the highest priority in its neighborhood and rebroadcasts it. In the commitment phase, when the proposed passage time windows of all robots do not overlap in two consecutive rounds of interaction, each robot sends a commitment message. In the confirmation phase, after receiving commitment messages from all involved robots, a unified passage sequence is generated. A fixed number of negotiation iteration rounds is preset. ,achieve If the commitment stage has not yet been reached, it is determined to be a negotiation deadlock, and an arbitration request will be automatically initiated to the central arbitration module. Specific example: Three robots, R1 (urgent materials, high task urgency, medium battery, fully loaded, short path duration), R2 (regular transportation, medium urgency, high battery, empty, medium duration), and R3 (returning to charge, low urgency, low battery, empty, long duration), drive towards an intersection.

[0040] When calculating priority, R1 It is significantly amplified by the exponential amplification term. R1 is the highest, followed by R2, and then R3 is the lowest. Proposal phase: R1 proposes. - R2 proposal - R3 proposal - Passed; R3 adjusted to the counter-proposal stage. - R2 adjusted to - R1 remains unchanged; the final time windows after iterations have no overlap. After two consecutive rounds of non-overlapping events, the commitment phase begins, and the confirmation phase generates the passage sequence R1→R2→R3. If... If no agreement can be reached after the rounds, an arbitration request will be initiated to the central arbitration module; Priority quantization algorithm with nonlinear amplification of urgency ensures priority passage for critical task robots, improving overall task execution efficiency; four-stage negotiation state machine provides a structured and highly convergent negotiation mechanism, effectively avoiding negotiation deadlock. The iteration limit and the automatic deadlock reporting mechanism form a conflict resolution system that combines distributed and centralized approaches, improving system robustness.

[0041] In another implementation, to adapt the priority weights to the dynamically changing industrial logistics scenario, the scenario-adaptive dynamic weights have a real-time iterative update mechanism, constructing a dynamic weight adjustment model with saturation constraints: ; in For the updated version Item weight, For the weights before the update, Adjusting the step size for fixed weights and , For the first The scenario adjustment coefficients corresponding to each indicator are obtained by linearly fusing three types of scenario features: factory congestion density, proportion of urgent task batches, and proportion of low-battery robots. ; in To account for the density of traffic congestion in the factory area, The percentage of urgent batches. For the proportion of low-power robots, , , These are the preset scene feature mapping coefficients; the path occupancy time corresponding to high congestion scenarios. Increase thus enlarge In emergency mission scenarios, the urgency level corresponds to... Increase thus enlarge After the update, the system is normalized to meet the requirements again. .

[0042] Specific example: Detected =0.8, =0.1, When =0.05, the path occupancy time , combined =0.05、 After the update ; Task urgency , After the update After updating, normalization ensures that the sum is 1; The scenario-adaptive dynamic weighting enables the priority evaluation model to dynamically adjust the weights of each indicator based on real-time scenario characteristics such as factory congestion density, the proportion of urgent task batches, and the proportion of low-battery robots. In the case of high congestion, the focus is on path efficiency, and in the case of urgent tasks, the focus is on task timeliness. This overcomes the limitations of rigid decision-making with fixed weights and improves the scheduling flexibility and decision rationality of the cluster in complex and ever-changing environments.

[0043] In another implementation, to address the discrepancy between the robot path tracking timing and the negotiated preset planning time, the negotiation commitment execution module is configured with a timing constraint closed-loop verification algorithm. This algorithm embeds the negotiated passage order and node time window constraints as hard constraint parameters into the local path planner, iteratively generating a smooth tracking path that satisfies the timing constraints. The timing deviation calculation formula is defined as follows: ; in This is the timing deviation value. The actual arrival time of the robot. Set a timeframe for negotiation; set a tolerance threshold for deviations. ,when If the timing of path tracking fails, the local conflict detection module is restarted to conduct a second conflict detection. If a medium-risk or higher conflict still exists after the second detection, the distributed right-of-way negotiation module is triggered to conduct a second negotiation to achieve dynamic closed-loop error correction. When the timing deviation is caused by communication interruption, the communication fault tolerance and completion mechanism is triggered first before the second conflict detection is performed. Specific example: Robot A negotiates with Robot B, giving A priority. It is expected that A will... The scheduled arrival and departure times were 1.5 seconds delayed due to slippery ground (A) during actual operation. , preset The timing sequence was determined to be faulty, and a secondary conflict detection was restarted. It was found that the delay of A caused a mismatch with the original plan. Robot C, which approved the request in seconds, generated a medium-risk conflict, triggering a second round of negotiation between A and C. A then proposed... If the process is completed within seconds, C adjusts accordingly, and once consensus is reached, the new constraints are reinserted into the path planner. If the delay is caused by a communication interruption, the current intention of A is first inferred from historical data using the communication fault tolerance and completion mechanism, and then a second check is performed. The timing constraint closed-loop verification algorithm monitors the deviation between the robot's actual operation and the plan in real time. It forms a dynamic closed-loop error correction through secondary conflict detection and secondary negotiation, which effectively avoids potential conflicts caused by execution errors, environmental changes or communication problems. Prioritizing the handling of deviations caused by communication interruptions further ensures the accuracy of decision-making and significantly improves the robustness and security of cluster scheduling.

[0044] In another implementation, to balance cluster scheduling efficiency, high-priority task timeliness, and individual robot fairness, the central arbitration module is configured with a globally optimal arbitration decision algorithm containing fairness constraints. Upon receiving an arbitration request, it simultaneously collects the priority score of the robot involved, the global road network congestion occupancy rate, the cluster task time slack, and the number of times each robot has been suppressed by arbitration in the past, and constructs a multi-objective optimized arbitration decision model: ; in The single-robot negotiation priority loss function is defined as the ratio of the sum of priority scores of the robots suppressed by arbitration to the total priority scores of all involved robots. The cluster global scheduling latency loss function is defined as the ratio of the sum of the additional waiting time of all involved robots under the arbitration scheme to the baseline scheduling time. The arbitration fairness loss function is defined as the range of the number of times each robot has been suppressed by arbitration in its history: ; in For the first The number of times Taiwan's robot history has been suppressed by arbitration; , , To balance the weights and satisfy By solving the minimization problem The globally optimal passage sequence scheme is obtained, and a standardized arbitration passage instruction is generated and synchronously sent to all involved robots, forcibly covering the results of local negotiation disagreements, as a rigid constraint for unified execution of the cluster; In its implementation, the central arbitration module can be configured with an arbitration decision algorithm based on the Model Predictive Control (MPC) framework. When an arbitration request occurs, the priority score is extracted from the negotiation message. The global road network congestion occupancy rate is calculated in real time by analyzing the positions, speeds, and preset paths of all robots in conjunction with the factory map. The cluster task timing margin is obtained from the upper-level task management system. The number of times historical tasks were suppressed by arbitration is stored in the central database and updated in real time. The multi-objective optimization model is deployed on a high-performance computing server, and the iterative optimizer calculates the optimal performance at each time step. Minimize traffic flow timing scheme. Balancing weights are preset based on operational strategy, increasing during peak production periods. Prioritize weighting to ensure efficiency, and increase efficiency during off-peak periods. The weights take into account fairness. After the optimal solution is obtained, it is encapsulated into a unified arbitration passage instruction (including path segment, speed limit and precise time window), which is synchronously sent out through the wireless communication network. The robot uses it as the highest priority constraint of the local path planner and immediately adjusts its motion trajectory. The multi-objective optimization arbitration decision model with fairness constraints comprehensively weighs priority loss, scheduling delay loss and fairness loss from a global perspective. It ensures the smooth execution of high-priority tasks, minimizes overall scheduling delay, and avoids the long-term suppression of a certain robot. It solves the problems of low efficiency or insufficient fairness of traditional arbitration mechanisms and significantly improves the overall operating efficiency, stability and robustness of the cluster.

[0045] In another implementation, to ensure the balancing weights of the arbitration decision-making model continuously adapt to complex dynamic environments, the central arbitration module possesses an iterative archiving and strategy optimization mechanism for arbitration results based on Bayesian optimization. After a single arbitration execution, it automatically collects and stores data on conflict type, reasons for negotiation deadlock, arbitration instruction parameters, robot execution time, actual scheduling delay, and fairness indicators. This constructs a system based on balancing weights. , , To optimize variables, the average loss of arbitration events within a sliding window is used. For the Bayesian optimization problem of the objective function, a Gaussian process is used to model the mapping relationship between the weight vector and the average loss, and the acquisition function is improved based on the expected value: ; in For the weight vector, This is the current optimal weight vector; each accumulated... After each arbitration event, a weight update iteration is performed, and the weight vector that maximizes the expected improvement is output as the balancing weight of the arbitration decision model in the next stage. Specific example: After the system processes 50 arbitrations, it finds that the current weights are causing high scheduling latency or that some robots are being frequently forced to avoid obstacles. The Bayesian optimization mechanism is activated, using historical data and a Gaussian process to predict the average loss for different weight combinations. If the model predicts a decrease... And improve It can significantly reduce overall latency. It is expected that the improved acquisition function will guide the system to evaluate the combination, and after iteration, it will output better weights. After the central arbitration module loads the new weights, subsequent arbitration will be more inclined to choose the solution that resolves conflicts quickly and reduces the overall waiting time. The arbitration strategy will continue to adapt to the actual operation. The iterative archiving and strategy optimization mechanism based on Bayesian optimization enables the central arbitration module to dynamically adjust the balance weights according to actual operating data and historical performance, overcoming the limitation that static weights cannot adapt to dynamic environments.

[0046] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A multi-device collaborative industrial logistics robot cluster scheduling and path optimization control system, comprising an onboard subsystem deployed on each robot body and a central arbitration module deployed at the central scheduling layer; characterized in that: The vehicle-mounted subsystem includes: a motion intent broadcasting module, an intent receiving and parsing module, a local conflict autonomous detection module, a distributed right-of-way negotiation module, and a negotiation commitment execution module; The motion intent broadcasting module includes broadcasting its own motion intent information to other robots within the communication range at a preset frequency via a vehicle-to-vehicle communication link; The intent receiving and parsing module interacts with the motion intent broadcasting module, including receiving motion intent information broadcast by neighboring robots, parsing and spatiotemporally aligning multi-source intent information, and generating a local intent perception map. The intent receiving and parsing module interacts with the motion intent broadcasting module, receives motion intent information broadcast by the neighboring robot, parses and performs spatiotemporal alignment processing on the multi-source intent information, generates a local intent perception map and outputs it to the local conflict autonomous detection module. The local conflict autonomous detection module is connected to the intent receiving and parsing module. It retrieves the local motion intent information, combines it with the local intent perception map to perform spatiotemporal overlap detection, calculates the conflict severity index and performs risk classification, and outputs the conflict identification result, which includes the conflict location, conflict duration, the robot involved, and the conflict risk level, to the distributed right-of-way negotiation module. The distributed right-of-way negotiation module is connected to the local conflict autonomous detection module. It executes a differentiated negotiation triggering strategy based on the conflict risk level: when the conflict risk level is low, it generates a local deceleration and avoidance command and sends it directly to the negotiation commitment execution module; when the conflict risk level is medium, it establishes a point-to-point negotiation session with all involved robots, negotiating the passage order based on the robot's task urgency, remaining battery power, load status, and path occupancy time. If the negotiation reaches an agreement, the passage order and time window constraints are output to the negotiation commitment execution module; if the negotiation is deadlocked, an arbitration request is initiated to the central arbitration module; when the conflict risk level is high, it directly initiates an arbitration request to the central arbitration module. The negotiation commitment execution module is connected to the distributed right-of-way negotiation module. It receives deceleration and avoidance instructions or negotiated passage order and time window constraints, and embeds the constraints into the local path planner to drive the robot to perform path tracking under time constraints. The central arbitration module communicates with the distributed right-of-way negotiation module of each robot, receives arbitration requests sent by the distributed right-of-way negotiation module, and issues arbitration passage instructions to the robot involved when point-to-point negotiation fails to reach a consensus or high-risk conflicts are directly reported.

2. The industrial logistics robot cluster scheduling and path optimization management system for multi-device collaboration as described in claim 1, characterized in that: The motion intent information includes the planned path, speed curve, task urgency, load status, and the expected time window to reach each path node; the motion intent broadcasting module broadcasts a broadcast message carrying the motion intent information to other robots within the communication range via the vehicle-to-vehicle communication link at a preset frequency. The header of the broadcast message includes the robot's unique identifier, message sequence number, and global timestamp.

3. The industrial logistics robot cluster scheduling and path optimization management system for multi-device collaboration as described in claim 1, characterized in that: The vehicle-to-vehicle communication link is configured with a communication fault tolerance and message completion mechanism, and a formula for determining the effective communication status is defined: ； in For message transmission delay, To preset the maximum allowable communication delay, For indicator functions, This serves as a valid communication identifier; when communication latency exceeds limits or packet loss is detected, a local historical intent data completion and short-term communication reconnection mechanism is automatically triggered, based on the most recent... The historical intent data of the frames is supplemented by linear extrapolation to complete the path nodes and time windows of the missing frames, and the supplemented data is marked as low confidence and input into the intent receiving and parsing module to avoid the loss of intent perception and negotiation interruption caused by communication fluctuations.

4. The industrial logistics robot cluster scheduling and path optimization management system for multi-device collaboration as described in claim 1, characterized in that: The intent receiving and parsing module is configured with a multi-source intent spatiotemporal alignment algorithm, uniformly using the system's global clock as a reference to complete timestamp calibration, and mapping the path data of each robot's local coordinate system to the global coordinate system through a coordinate transformation matrix, satisfying the following: ； in For global coordinate vectors, This is the robot's local coordinate vector. Let be a rotation matrix. The translation vector is used. After eliminating the local clock deviation and coordinate reference difference of multiple robots, the path trajectory, running speed and node time window data of all neighboring robots are normalized and aligned. Using the robot's unique identifier as an index, the path space interval and time sequence window parameters are associated and matched to construct a structured local intent perception map. Each intent record in the local intent perception map carries a confidence label. The supplementary data from the communication fault tolerance module is marked as low confidence, and the data from real-time broadcast is marked as high confidence.

5. The industrial logistics robot cluster scheduling and path optimization management system for multi-device collaboration as described in claim 1, characterized in that: The local conflict autonomous detection module is equipped with a spatiotemporally coupled conflict severity grading judgment model, which quantifies the path space overlap ratio, time window overlap ratio, and robot relative speed factor, and defines the conflict severity index calculation formula: ； in As a conflict severity index, This represents the area of ​​the overlapping region in the path space. This refers to the footprint of a single machine. The duration of time window overlap. This represents the width of the time window for the corresponding path node. The relative speed of the robot involved. The maximum permissible speed of the system. , , The weighted coefficients are the spatial, temporal, and velocity dimensions and satisfy the following conditions: ; Preset two-level risk thresholds , and Risk classification is performed according to the following rules: When When it is judged as low risk, When it is determined to be medium risk, The path is identified as high-risk; CSI calculation is initiated only when there is an intersection in the path space and the time window overlap duration is greater than zero, eliminating pseudo-collision interference where there is no spatial intersection or the time window is completely misaligned.

6. The industrial logistics robot cluster scheduling and path optimization management system for multi-device collaboration as described in claim 1, characterized in that: The distributed right-of-way negotiation module constructs a multi-dimensional priority quantification algorithm with nonlinear amplification of urgency and a four-stage negotiation state machine. The four core indicators—task urgency, remaining battery power, load status, and path occupancy time—are normalized to a range of 0-1. A nonlinear weighted fusion model is then used to calculate the robot's passage priority score. ； in For the first Priority score for each robot , , , These are the normalized task urgency, remaining battery power, load status, and path occupancy time metrics, respectively. This is an amplified item for the task urgency index. For preset magnification factor and >0, , , , For scene-adaptive dynamic weights and satisfying + + + = 1; The four-stage negotiation state machine includes: a proposal stage, in which each involved robot broadcasts its own... In the proposed passage time window phase, the robot with lower priority adjusts its proposed passage time window based on the highest priority in its neighborhood and rebroadcasts it. In the commitment phase, when the proposed passage time windows of all robots do not overlap in two consecutive rounds of interaction, each robot sends a commitment message. In the confirmation phase, after receiving commitment messages from all involved robots, a unified passage sequence is generated. A fixed number of negotiation iteration rounds is preset. ,achieve If the negotiation has not yet reached the commitment stage, it is considered a deadlock and an arbitration request will be automatically initiated to the central arbitration module.

7. The industrial logistics robot cluster scheduling and path optimization management system for multi-device collaboration as described in claim 1, characterized in that: The scenario-adaptive dynamic weights have a real-time iterative update mechanism, constructing a dynamic weight adjustment model with saturation constraints: ； in For the updated version Item weight, For the weights before the update, Adjusting the step size for fixed weights and , For the first The scenario adjustment coefficients corresponding to each indicator are obtained by linearly fusing three types of scenario features: factory congestion density, proportion of urgent task batches, and proportion of low-battery robots. ； in To account for the density of traffic congestion in the factory area, The percentage of urgent batches. For the proportion of low-power robots, , , These are the preset scene feature mapping coefficients; the path occupancy time corresponding to high congestion scenarios. Increase thus enlarge In emergency mission scenarios, the urgency level corresponds to... Increase thus enlarge After the update, the system is normalized to meet the requirements again. .

8. The industrial logistics robot cluster scheduling and path optimization management system for multi-device collaboration as described in claim 5, characterized in that: The negotiation commitment execution module is configured with a time-constraint closed-loop verification algorithm. The negotiated passage order and node time window constraints are implanted as hard constraint parameters into the local path planner, iteratively generating a smooth tracking path that satisfies the time constraints. The time-series deviation calculation formula is defined as follows: ； in This is the timing deviation value. The actual arrival time of the robot. A timeframe is planned for the negotiation; Set deviation tolerance threshold ,when If the timing of path tracking fails, the local conflict detection module is restarted to conduct a second conflict detection. If a medium-risk or higher conflict still exists after the second detection, the distributed right-of-way negotiation module is triggered to conduct a second negotiation to achieve dynamic closed-loop error correction. When the timing deviation is caused by communication interruption, the communication fault tolerance and completion mechanism is triggered first before the second conflict detection is performed.

9. The industrial logistics robot cluster scheduling and path optimization management system for multi-device collaboration as described in claim 1, characterized in that: The central arbitration module is configured with a globally optimal arbitration decision algorithm containing fairness constraints. After receiving an arbitration request, it simultaneously collects the priority score of the robot involved, the global road network congestion occupancy rate, the cluster task time slack, and the number of times each robot has been suppressed by arbitration in the past, and constructs a multi-objective optimized arbitration decision model: ； in The single-robot negotiation priority loss function is defined as the ratio of the sum of priority scores of the robots suppressed by arbitration to the total priority scores of all involved robots. The cluster global scheduling latency loss function is defined as the ratio of the sum of the additional waiting time of all involved robots under the arbitration scheme to the baseline scheduling time. The arbitration fairness loss function is defined as the range of the number of times each robot has been suppressed by arbitration in its history: ； in For the first The number of times Taiwan's robot history has been suppressed by arbitration; , , To balance the weights and satisfy By solving the minimization problem The globally optimal passage sequence scheme is obtained, and a standardized arbitration passage instruction is generated and synchronously sent to all involved robots, forcibly covering the results of local negotiation disagreements, as a rigid constraint for unified execution of the cluster.

10. The industrial logistics robot cluster scheduling and path optimization management system for multi-device collaboration as described in claim 7, characterized in that: The central arbitration module possesses a Bayesian-optimized iterative archiving and strategy optimization mechanism for arbitration results. After a single arbitration execution, it automatically collects and stores data on conflict type, reasons for negotiation deadlock, arbitration instruction parameters, robot execution time, actual scheduling latency, and fairness indicators; and constructs a system based on balanced weights. , , To optimize variables, the average loss of arbitration events within the sliding window is used. For the Bayesian optimization problem of the objective function, a Gaussian process is used to model the mapping relationship between the weight vector and the average loss, and the acquisition function is improved based on the expected value: ； in For the weight vector, This is the current optimal weight vector; each accumulated... After each arbitration event, a weight update iteration is performed, and the weight vector that maximizes the expected improvement is output as the balancing weight of the arbitration decision model in the next stage.

Images