Unmanned forklift man-machine collaborative optimization scheduling method and system

Abstract

The invention relates to the field of control and adjustment, in particular to a man-machine collaborative optimization scheduling method and system for an unmanned forklift. The method comprises the steps that tasks, unmanned forklift resource portraits and personnel authority skill labels are obtained, and the resource portraits comprise operation capacity and economic attributes; extracting task features to quantify priorities of the task features; predicting a personnel behavior intention based on environmental perception; calculating a collaborative adaptation degree by synthesizing task features, resource portraits, authority skills and behavior intentions so as to quantify the adaptation degree of the candidate objects and the tasks; distributing a target unmanned forklift and personnel and generating an instruction; and performing dynamic security risk assessment and triggering an adjustment strategy according to sensing and prediction changes during execution. According to the invention, multi-dimensional resource collaborative optimization scheduling of vehicle-personnel combination and tasks is realized in a complex man-machine mixed operation environment, and the comprehensive effects of reducing the comprehensive operation cost, greatly improving the collaborative efficiency and safety, and remarkably enhancing the intelligence and flexibility of the system are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of control adjustment, in particular to a man-machine collaborative optimization scheduling method and system of unmanned forklifts. BACKGROUND

[0002] In recent years, as the core equipment of intelligent logistics, unmanned forklifts have realized extensive application from simple line-side handling to complex warehouse management through deep integration of navigation technology, artificial intelligence and the Internet of Things. The industry is rapidly developing towards high flexibility and high intelligence, and the market prospect is broad. However, in the current and foreseeable future, a completely unmanned "black light factory" is not a realistic choice for all enterprises. In most industrial scenarios, such as unstacking, picking, and on-line and off-line feeding, unmanned forklifts still need to work closely with personnel to form a "man-machine mixed field" of the normal working mode.

[0003] The existing scheduling methods mainly focus on path optimization and obstacle avoidance of vehicles, which have the following significant limitations: first, the scheduling strategy is single, only aiming at "shortest distance" or "fastest time", ignoring the differences in operating costs of different vehicle types, resulting in waste of high-cost equipment resources; second, the system simply regards personnel as dynamic obstacles to be avoided, rather than schedulable collaborative resources, and cannot match tasks according to the skills, permissions and real-time intentions of personnel, resulting in long waiting time for human operation, interruption of collaborative processes, and safety hazards; third, the rigid scheduling rules cannot adapt to complex and variable mixed task demands (such as the coexistence of emergency distribution and regular replenishment), and the overall system efficiency and flexibility are insufficient. Therefore, there is an urgent need for an intelligent collaborative scheduling scheme that can comprehensively optimize economic, safety and efficiency factors, and realize deep integration and lean management of people, vehicles and tasks. SUMMARY

[0004] The purpose of the present application is to provide an intelligent man-machine coordination scheduling scheme for unmanned forklifts that takes into account economy, safety and work efficiency.

[0005] According to a first aspect of the present application, a man-machine collaborative optimization scheduling method for unmanned forklifts is provided, comprising the following steps: Step S1, obtaining resource portrait information of a task to be scheduled, a forklift formation and permission and skill label information of a staff group, the resource portrait information at least including forklift types, operating capacity attributes and economic attributes of a plurality of unmanned forklifts in the forklift formation, and the permission and skill label information at least including safety permissions and skill levels of a plurality of staff in the staff group; Step S2, extracting task features of the task to be scheduled, the task features including quantification of task priority; Step S3, predicting the behavior intention of the staff in the operating area based on real-time sensed environmental information; Step S4, based on the task characteristics, resource portrait information, permission skill label information and predicted behavior intention, the collaborative adaptation degree of all task-unmanned forklift-worker combinations is calculated; the collaborative adaptation degree is used to quantify the adaptation degree of the candidate unmanned forklift-worker combination for executing the task to be dispatched; Step S5, according to the collaborative adaptation degree, the target unmanned forklift and the target worker are assigned to the task to be dispatched, and the control instruction is generated.

[0006] According to some embodiments, in the method of the first aspect of the application, step S2 comprises: A task feature vector is constructed based on the task to be dispatched, including urgency, human-machine collaboration demand, material characteristics, operation area density and time window constraint; A quantitative score of task priority is generated by calculating multiple dimensions of the task feature vector through a weighted summation formula; wherein the weight of each task feature vector in the weighted summation formula is determined according to a work demand strategy template, and the work demand strategy template includes a production line guarantee template, a cost priority template and an efficiency smooth template; the production line guarantee template is determined based on the business goal of ensuring the uninterrupted production assembly line; the cost priority template is determined based on the business goal of completing daily tasks while minimizing operating costs; and the efficiency smooth template is determined based on the business goal of maximizing the overall throughput of the warehouse.

[0007] According to some embodiments, in the method of the first aspect of the application, the human-machine collaboration demand is used to represent the degree of collaboration between personnel and unmanned forklifts for the task to be dispatched, which is obtained according to the following steps: The task to be dispatched is preliminarily classified based on a preset collaboration rule library to obtain a first collaboration demand value corresponding to the task collaboration type; The attribute increment value of the material in the task to be dispatched is obtained based on a material master database, and the first collaboration demand value is adjusted for the first time based on the attribute increment value to obtain a second collaboration demand value; The historical average collaboration intensity is determined based on historical task data, and the second collaboration demand value is adjusted for the second time according to the historical average collaboration intensity to determine the final human-machine collaboration demand.

[0008] According to some embodiments, in the method of the first aspect of the application, for each task-unmanned forklift-worker combination, step S4 comprises: The task-forklift adaptation score is calculated based on the matching degree of the task feature vector and the operation capability attributes and economic attributes of the unmanned forklift, specifically by weighted summation of six dimension indexes of load matching degree, lifting matching degree, tool matching degree, channel adaptability, navigation precision matching degree and economic matching degree, and the quantification of each dimension index is obtained by matching degree definition rule; The forklift-personnel coordination score is calculated based on the matching degree of the work ability attribute of the unmanned forklift and the permission skill tag information of the worker, and the matching degree of the work ability attribute and the predicted behavior intention; The environment adaptation score is calculated based on the friendliness of the real-time perceived environment information to the execution of the task to be dispatched. The coordination adaptation degree is calculated based on the following formula: coordination adaptation degree = task-forklift adaptation score × A + forklift-personnel coordination score × B + environment adaptation score × C, wherein A, B, and C are preset weight coefficients.

[0009] According to some embodiments, in the method of the first aspect of the application, step S4 further comprises: The initial values of the weight coefficients A, B, and C are set based on the balanced starting principle, and the numerical values of the weight coefficients A, B, and C are dynamically adjusted according to the task characteristics of the task to be dispatched, so as to switch between different scheduling goals of cost priority, efficiency priority, or safety priority.

[0010] According to some embodiments, in the method of the first aspect of the application, step S3 comprises: The multi-modal data frame of the worker is obtained by the multi-modal sensor; Based on the multi-modal data frame, a feature sequence is extracted, which includes body joint angle, hand orientation, and motion speed information; The feature sequence is input into the trained time series prediction model, and a probability distribution for representing the worker's execution of picking, walking, or entering an emergency state in the future period of time is output.

[0011] According to some embodiments, the method of the first aspect of the application further comprises: During the task execution process, the continuously perceived environment information and the predicted behavior intention are stored at a first cycle value; The environment information and the behavior intention are subjected to safety risk assessment, and an adjustment strategy is triggered according to the assessment result.

[0012] According to some embodiments, in the method of the first aspect of the application, the safety risk assessment of the environment information and the behavior intention, and the triggering of the corresponding adjustment strategy according to the assessment result, comprise: A real-time risk index is derived from the environment information and the behavior intention; in the case that the real-time risk index meets the safety warning condition, a safety priority adjustment strategy is triggered; in the case that the real-time risk index meets the efficiency optimization condition, an efficiency priority adjustment strategy is triggered, and a new unmanned forklift is assigned to take over the task.

[0013] According to some embodiments, in the method of the first aspect of the application, the safety priority adjustment strategy comprises: The control-related unmanned forklifts immediately enter the emergency stop or low-speed slow running state; The unmanned forklifts and the staff in the surrounding area are broadcasted with early warning information, and the electronic fence area is dynamically updated.

[0014] According to a second aspect of the present application, an unmanned forklift man-machine collaborative optimization scheduling system is provided, comprising: An information acquisition module is configured to acquire a to-be-scheduled task, resource portrait information of an unmanned forklift formation, and permission skill label information of a staff group, wherein the resource portrait information at least includes forklift types, work capacity attributes and economic attributes of multiple unmanned forklifts in the unmanned forklift formation, and the permission skill label information at least includes safety permissions and skill levels of multiple staffs in the staff group; A task feature extraction module is configured to extract task features of the to-be-scheduled task, wherein the task features include quantification of task priorities; A perception prediction module is configured to predict behavior intentions of the staff in the work area based on real-time perceived environmental information; A collaborative adaptation degree calculation module is configured to calculate collaborative adaptation degrees of all task-unmanned forklift-staff combinations based on the task features, the resource portrait information, the permission skill label information and the predicted behavior intentions, wherein the collaborative adaptation degrees are used to quantify adaptation degrees of candidate unmanned forklift-staff combinations for executing the to-be-scheduled task; A scheduling decision module is configured to assign target unmanned forklifts and target staffs to the to-be-scheduled task according to the collaborative adaptation degrees, and generate control instructions; A safety closed-loop control module is configured to dynamically perform safety risk assessment based on continuously perceived environmental information and predicted behavior intentions during task execution, and trigger corresponding adjustment strategies according to the assessment results.

[0015] The scheme provided by the present application has the following beneficial effects: 1. In view of the problem that the existing scheduling system leads to resource mismatch and high operating cost due to neglecting equipment economy, the scheme of the present application constructs an economic portrait including "hourly operation cost" for different types of unmanned forklifts, and designs a cost priority type hierarchical scheduling strategy, thereby realizing the technical effects of preferentially scheduling low-cost forklifts under the premise of ensuring task completion, significantly reducing comprehensive operating cost, and maximizing forklift fleet resource utilization.

[0016] 2. In view of the problems of low efficiency and prominent safety hazards caused by regarding personnel as obstacles instead of resources in existing human-machine collaboration, the scheme of the present application establishes a permission skill tag system for workers, and integrates personnel behavior intention prediction and perception data as core factors into a collaborative adaptation quantitative calculation model, thereby realizing accurate allocation of suitable tasks to the most suitable person and vehicle, greatly improving collaborative operation efficiency, and fundamentally reducing collision risk.

[0017] 3. In view of the problem that a single scheduling strategy cannot adapt to complex and variable business scenarios of mixed warehouses, resulting in poor system flexibility and limited overall throughput, the scheme of the present application introduces a hierarchical strategy matching mechanism, dynamically selects the optimal scheduling target according to task characteristics, realizes intelligent self-adaptation of the system to complex environments with diversified tasks, multiple vehicle types, and multiple personnel coexistence, and significantly enhances overall throughput and flexibility. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art based on these drawings without exceeding the scope of the present application.

[0019] Figure 1 A flowchart of an embodiment 1000 of the unmanned forklift human-machine collaborative optimization scheduling method of the present application; Figure 2 A flowchart of step S2 in the embodiment 1000; Figure 1 A flowchart of the human-machine collaboration demand degree quantitative calculation process S21a in step S2; Figure 3 Figure 2 A flowchart of the human-machine collaboration demand degree quantitative calculation process S21a in step S2; Figure 4 A flowchart of step S3 in the embodiment 1000; Figure 1 A flowchart of the embodiment S4a of step S4 in the embodiment 1000; Figure 5 Figure 1 A flowchart of the embodiment S4b of step S4 in the embodiment 1000; Figure 6 A flowchart of the embodiment S4b of step S4 in the embodiment 1000; Figure 1 A flowchart of an embodiment 2000 of the unmanned forklift human-machine collaborative optimization scheduling method of the present application; Figure 7 A flowchart of the embodiment 2000 of the unmanned forklift human-machine collaborative optimization scheduling method of the present application; Figure 8 Figure 2 ​​​a flowchart of step S207 in embodiment 2000 of the application; Figure 9 a structural schematic diagram of an embodiment 3000 of a man-machine collaborative optimization scheduling system of a unmanned forklift truck of the application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are some but not all of the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the application.

[0021] Figure 1 a flowchart of an embodiment 1000 of a man-machine collaborative optimization scheduling method of a unmanned forklift truck of the application. As shown in Figure 1 embodiment 1000 includes steps S1-S5.

[0022] In step S1, a man-machine collaborative optimization scheduling method (for example, a processor, hereinafter referred to as a processor) acquires a to-be-scheduled task, resource portrait information of a unmanned forklift truck formation, and permission skill label information of a staff team, the resource portrait information at least including forklift truck types, work capacity attributes and economic attributes of multiple unmanned forklift trucks in the unmanned forklift truck formation, and the permission skill label information at least including safety permissions and skill levels of multiple staff in the staff team.

[0023] In some specific embodiments, in step S1, the processor acquires the to-be-scheduled task through an upper management system, and communicates between systems through a pre-defined application program interface. Optionally, the upper management system is a warehouse management system WMS or a manufacturing execution system MES. In step S1, the processor calls an interface provided by the WMS / MES to acquire a task queue in a polling or push-receiving manner, and acquires original instructions of the task.

[0024] In some specific embodiments, in step S1, the processor acquires the resource portrait information of the unmanned forklift truck formation from a local resource management system, and acquires all forklift truck records that are online and in an idle or about-to-be-idle state by querying a database table. Optionally, the resource portrait information acquired in step S1 includes the following information of each unmanned forklift truck: (1) forklift truck type: static attribute, determined based on a forklift truck configured in a current warehouse scene, for example, including a balanced heavy-duty unmanned forklift truck, a front-moving unmanned forklift truck, a narrow-channel three-way unmanned forklift truck, a light pallet unmanned forklift truck, and an unmanned forklift truck AMR; (2) Work ability attributes: static attributes, for example, including max_load (rated load), fork_type (fork type, such as standard fork, side shift fork), max_lift_height (maximum lifting height), navigation_type (navigation method); dynamic attributes, for example, including current_battery_level (current battery level), current_location (current location), current_status (current status, such as running, idle, charging); (3) Economic attributes: hourly operation cost, calculated by a financial model; energy consumption level, power consumption per unit distance. Alternatively, the hourly operation cost is calculated by a formula: hourly operation cost = (depreciation + energy consumption + estimated maintenance cost) / daily average effective working hours.

[0025] In some embodiments, at step S1, the processor obtains the work staff's permission skill tag information through a human resource system or a warehouse management system, specifically by an API interface or querying a local database to obtain all on-duty and available personnel information.

[0026] Alternatively, at step S1, the work staff's permission skill tag information includes: (1) Safety permissions: authorized areas, used to ensure that personnel are not sent to unauthorized dangerous areas; equipment operation qualifications, used to ensure the correct and compliant operation of personnel on equipment; (2) Skill level: a proficiency level corresponding to the distribution of core skills in a warehouse scenario; for example, skills include fragile product handling, electronic product picking, and code disc, and proficiency levels can be divided into 1-5 levels or 1-10 levels. Efficiency index, calculated based on the average time of historical completed tasks.

[0027] Step S2, the processor extracts the task feature vector of the task to be dispatched, and the task feature vector includes the quantification of the task priority.

[0028] In some embodiments, at step S2, the processor constructs a task feature vector based on the task to be dispatched, including urgency, human-machine cooperation demand, material characteristics, work area density, and time window constraints; and then calculates multiple dimensions of the task feature vector through a weighted summation formula to generate a quantification score of the task priority.

[0029] Optionally, the calculation of the human-machine collaboration demand in step S2 comprises: performing preliminary classification on the to-be-scheduled task based on the collaboration rule base to obtain a first collaboration demand value corresponding to the task collaboration type; obtaining an attribute increment value of the material in the to-be-scheduled task based on the material master database, and performing first fine-tuning on the first collaboration demand value based on the attribute increment value to obtain a second collaboration demand value; determining a historical average collaboration intensity based on historical task data, and performing second fine-tuning on the second collaboration demand value according to the historical average collaboration intensity to determine the final human-machine collaboration demand.

[0030] Step S3, the processor predicts the behavior intention of the personnel in the work area based on the real-time sensed environment information.

[0031] In some embodiments, in step S3, the processor completes the prediction of the behavior intention of the personnel by the following steps: obtaining an image sequence and motion data of the personnel through the multi-modal sensor; extracting a feature sequence containing a body joint angle, a hand orientation, and a motion speed based on the image sequence and the motion data; inputting the feature sequence into a trained time series prediction model, and outputting a probability distribution for representing the personnel to perform a picking, threading, or entering an emergency state in a future period of time.

[0032] Step S4, the processor calculates the collaborative adaptation degree of all task-unmanned forklift-personnel combinations based on the task features, the resource portrait information, the permission skill label information, and the predicted behavior intention, and the collaborative adaptation degree is used to quantify the adaptation degree of the candidate unmanned forklift-personnel combination for executing the to-be-scheduled task.

[0033] In some embodiments, in step S4, the processor calculates the collaborative adaptation degree of the task-unmanned forklift-personnel combination by the following steps: obtaining a task-forklift adaptation score based on the matching degree calculation of the task features and the work ability attributes and economic attributes of the unmanned forklift; obtaining a forklift-personnel collaboration score based on the matching degree calculation of the work ability attributes of the unmanned forklift and the permission skill label information of the personnel, and the predicted behavior intention; obtaining an environment adaptation score based on the friendliness calculation of the real-time sensed environment information for executing the to-be-scheduled task; calculating the collaborative adaptation degree based on the following formula: collaborative adaptation degree = task-forklift adaptation score x A + forklift-personnel collaboration score x B + environment adaptation score x C, wherein A, B, and C are preset weight coefficients.

[0034] Step S5, the processor assigns a target unmanned forklift and a target personnel to the to-be-scheduled task according to the collaborative adaptation degree, and generates a control instruction.

[0035] In some embodiments, the specific process of step S5 comprises: The processor first generates a list of candidate triplets, including all possible combinations of task-unmanned forklift-worker, and for each combination, the processor obtains the synergy fitness score of each combination calculated in step S4; Then, resource conflicts of these candidate combinations are detected and globally optimized. For example, the same forklift or the same worker can appear in multiple high-score candidate triplets. Alternatively, in the process of global optimization, the processor's goal is not to find the best allocation for a single task, but to find a globally optimal or near-optimal allocation scheme for all currently scheduled tasks; The processor selects an appropriate algorithm according to the size of the current work scene, including: for small and medium-sized scenes, the Hungarian algorithm or the minimum cost maximum flow algorithm can be used to find a globally optimal solution to ensure that the sum of the total synergy fitness scores of all tasks is the highest; for large-scale real-time scenes: a variant of the greedy algorithm is used. That is: all candidate triplets are sorted in descending order of synergy fitness score, and then allocated in order. Once a resource is allocated, all triplets containing that resource are removed from the candidate list.

[0036] In some embodiments, after the allocation scheme is determined in step S5, the processor sends structured control instructions to the target unmanned forklift and the target worker. Alternatively, the control instructions for the target unmanned forklift include the worker ID, the worker coordinates, and the cooperation mode. Alternatively, the control instructions for the target worker include the action to be performed, the forklift ID, the forklift estimated arrival time, and the prompt text / map.

[0037] According to the embodiment as shown in Figure 1 The scheme proposed by the present application achieves a paradigm shift from traditional rigid task-vehicle binary scheduling to flexible task-vehicle-worker-environment multi-element collaborative scheduling by constructing a complete decision-making data chain covering four-dimensional information of tasks, vehicles, workers, and environment, and performing quantitative calculation and dynamic optimization based on this. The scheme of the present application can achieve multi-objective optimization of significantly reducing operating costs, significantly improving collaborative work efficiency, and further enhancing system safety and flexibility in a complex and dynamic human-machine mixed operation environment.

[0038] Figure 2 For Figure 1 the flowchart of step S2 in embodiment 1000. As shown in Figure 2 step S2 includes steps S21-S22.

[0039] In step S21, the processor constructs a task feature vector based on the tasks to be scheduled, including urgency, human-machine collaboration demand, material characteristics, work area density, and time window constraints.

[0040] In some embodiments, in step S21, the processor obtains real-time state data through the warehouse management system WMS or the manufacturing execution system MES, queries the state of the external system in real time and makes logical judgments by using a rule engine or a rule matching module built in the processor, to obtain the urgency value. Alternatively, in step S21, a rule library is included in the rule engine or the rule matching module, and the task attributes and the pre-defined urgency rules are matched to obtain a normalized value.

[0041] In some embodiments, in step S21, the processor obtains material characteristics through the material master database, and the material characteristics represent the actual physical attributes of the material. Alternatively, the material characteristics are a composite dimension, which is usually obtained by weighted calculation of multiple sub-features to obtain a comprehensive score, and the dimensions involved include weight, size / volume, and special properties (for example, fragile goods, dangerous goods, and ordinary goods).

[0042] In step S21, the weight in the material characteristics is normalized by Weight = (material weight - minimum weight) / (maximum weight - minimum weight), and the size / volume is also normalized in a similar manner. For the special properties of the material, the values are normalized by pre-defined rules, for example, a quantitative rule for a special property includes: {fragile goods: 1.0, dangerous goods: 0.9, ordinary goods: 0.1}.

[0043] In some embodiments, in step S21, the work area density is obtained by the real-time dynamic map of the processor itself. Specifically, in the actual working scene, the processor includes a real-time position table of all moving targets (personnel, unmanned forklifts, and other AGVs). When a new task arrives, the processor calculates the number of moving targets within a preset radius on the scheduled path of the task or around the target point, and then calculates the work area density value by the formula: area density = min((total number of targets in the area / area) / maximum density threshold, 1.0).

[0044] In some embodiments, in step S21, the time window constraint value is obtained by the expected completion time of the task itself and the current time obtained by the processor, and specifically includes: time difference calculation, time urgency = (task deadline - current time) / task standard time consumption; function mapping, using a decay function to map, the closer to the deadline, the exponential growth of the urgency.

[0045] In step S22, the processor calculates multiple dimensions of the task feature vector by a weighted summation formula to generate a quantitative score of the task priority. In step S22, the weighted summation is not a simple formula summation, but is calculated by a configurable strategy framework.

[0046] For example, in step S22, the processor defines the formula of the task priority score as follows: , wherein U, C, M, D, T represent the normalized values of the five dimensions of urgency, human-machine collaboration demand, material characteristics, regional density, and time window constraint, respectively; are weight coefficients corresponding to the five dimensions, respectively, and the sum of the weight coefficients is 1. Optionally, in step S22, the weight coefficients are stored in a weight configuration table of the processor database, allowing the administrator to dynamically adjust the strategy through the UI interface without the need to redeploy the code.

[0047] In some embodiments, in step S22, the processor presets multiple work demand strategy templates for determining the weights for weighted summation according to the work demand. Optionally, the strategy templates include a production line guarantee template, a cost priority template, and an efficiency smooth template, the production line guarantee template is determined based on the business goal of ensuring the uninterrupted production assembly line; the cost priority template is determined based on the business goal of completing daily tasks while minimizing operating costs; and the efficiency smooth template is determined based on the business goal of maximizing the overall throughput of the warehouse. For example, the production line guarantee template is: , the cost priority template is: , and the efficiency smooth template is: . Optionally, the implementation process of step S22 includes the following procedures: receiving a task and its five dimension values; querying the currently effective weight configuration from the database; performing weighted summation; and outputting a unique and comparable PriorityScore value.

[0048] According to the embodiment as shown in Figure 2 , the scheme proposed by the present application converts abstract and multi-aspect business demands into structured and computable quantitative indicators, providing precise and unambiguous digital inputs for subsequent automated decision-making.

[0049] Figure 3 The flowchart of the quantification calculation process S21a of the human-machine collaboration demand degree in step S2 of Figure 2 . As shown in Figure 3 , the quantification calculation process S21a of the human-machine collaboration demand degree includes steps S211-S212.

[0050] In step S211, the processor classifies the task to be dispatched based on the collaboration rule library to obtain a first collaboration demand degree value corresponding to the collaboration type of the task. Optionally, the collaboration rule library preset in step S211 is preset based on expert experience and is fixed into the configuration file of the entire system as the source of the initial value and the reference value of the human-machine collaboration demand degree value of the task.

[0051] Optionally, one specific embodiment of the collaboration rule library includes: Rule 1, the task to be scheduled requires personnel to pick, determine the task collaboration type as A class → the first collaboration demand value is 0.7, this type of task requires personnel to participate deeply, and the collaboration demand is the highest; Rule 2, the task to be scheduled requires personnel to handle goods, determine the task collaboration type as B class → the first collaboration demand value is 0.4, this type of task requires personnel to cooperate for a short time, and the collaboration demand is medium; Rule 3, the task to be scheduled is pure forklift handling, determine the task collaboration type as C class → the first collaboration demand value is 0, this type of task basically does not require personnel to operate, and the collaboration demand is very low.

[0052] In some specific embodiments, the preliminary classification and initial value determination of step S211 ensure that the subsequent fine mapping will not produce unreasonable results, and ensure the logical rationality and business relevance of the value assignment.

[0053] Step S212, the processor obtains the attribute increment value of the material in the task to be scheduled based on the material master database, and performs the first fine adjustment on the first collaboration demand value based on the attribute increment value to obtain the second collaboration demand value.

[0054] In some specific embodiments, in step S212, the processor automatically queries the material master database in the enterprise resource planning (ERP) or warehouse management system (WMS), and calculates the attribute increment value of the first fine adjustment on the first collaboration demand value through mapping rules, which specifically includes: querying the attributes of the material through the material ID in the task to be scheduled, the key attributes including storage unit, weight, whether fragile, and whether dangerous goods; mapping the material attributes into values, for example, the storage unit is piece, the mapping value is 0.2; the weight is greater than 20 kg, the mapping value is 0.1; the material is fragile, the mapping value is 0.15; finally, the attribute increment value is obtained by adding the material attribute mapping values, and the sum of the first collaboration demand value obtained in step S211 and the attribute increment value is calculated as the second collaboration demand value.

[0055] Step S213, the processor determines the historical average collaboration intensity based on the historical task data, and performs the second fine adjustment on the second collaboration demand value according to the historical average collaboration intensity to determine the final human-machine collaboration demand.

[0056] In some specific embodiments, in step S213, the processor performs statistics and learning based on the historical task data, and performs the second fine adjustment on the second collaboration demand value to determine the final human-machine collaboration demand, which realizes the continuous self-optimization of the method, so that the value is more in line with the actual demand.

[0057] In some specific embodiments, the historical task data in step S213 is recorded by the processor when completing the task, and the key fields include: Task basic information: task ID, material ID, start point, end point, planned priority; Timestamp information: task start time, task end time, personnel start operation time, personnel end operation time, time consumption for waiting for personnel arrival; Resource information: executed forklift ID, assisted personnel ID.

[0058] Among them, the core index related to the human-machine collaboration demand degree is the personnel operation time length ratio, which is calculated by the following formula:

[0059] In some specific embodiments, at step S213, the processor fine-tunes the second collaboration demand degree by calculating the historical average collaboration intensity, specifically including: querying similar historical tasks, determined by keyword query; calculating the average collaboration intensity of the historical tasks, that is, calculating the average value of the personnel operation time length ratio of the historical characters; and obtaining the final human-machine collaboration demand degree value based on the following formula: , wherein, is the confidence weight, determined by min(1, similar historical task number / N). N is a confidence threshold, N is used to measure the influence degree of historical data. Optionally, N is set to 10.

[0060] Optionally, one specific embodiment of steps S211-S213 includes: Task content: transport material A to picking area B.

[0061] Rule base output: because the destination is a picking area, the rule base gives a benchmark value, that is, the first collaboration demand degree value 0.7; First fine-tuning: the cargo weight is 20kg, the attribute increment value is 0.1, and the second collaboration demand degree value is 0.7+0.1=0.8; Data query: the processor queries the historical data and finds that the number of similar tasks is 8, the average value of the collaboration intensity of the 10 similar tasks is 0.5, the confidence threshold N=10, =min(1,8 / 10)=0.8; Second fine-tuning: the final human-machine collaboration demand degree value is: .

[0062] According to the embodiments as shown in Figure 3 , the scheme provided by the present application converts the fuzzy collaboration concept into an accurate continuous value, can realize clear distinction of different levels of tasks, and realizes intelligent adjustment of the human-machine collaboration demand degree through combination of historical data, thereby providing a key basis for subsequent resource matching, and fundamentally avoids resource mismatch and efficiency loss caused by task type misjudgment.

[0063] Figure 4 For Figure 1 a flowchart of step S3 in embodiment 1000. As Figure 4 shown, step S3 includes steps S31-S33.

[0064] Step S31, the processor acquires a multi-modal data frame of the worker through the multi-modal sensor.

[0065] Optionally, the multi-modal sensor in step S31 includes an RGB camera, a depth camera or a ToF camera, a millimeter wave radar, and an inertial measurement unit. Among them, the RGB camera is deployed on the forklift body or the key area of the warehouse, providing high-resolution 2D color images for accurate visual feature extraction; the depth camera or the ToF camera is registered with the RGB camera, providing pixel-level depth information for calculating accurate distance and spatial relationship; the millimeter wave radar is not sensitive to environmental light and can directly provide radial velocity information of the target, which is used to stably track the motion speed of the worker in low-visibility working conditions such as dust and haze; the inertial measurement unit is deployed on the worker's smart bracelet or badge, providing more accurate acceleration and angular velocity data.

[0066] In some specific embodiments, in step S31, the multi-modal sensor meets the following conditions: the timestamps are aligned through hardware synchronization or software synchronization algorithm to ensure that the data frames at the same time correspond to the same scene; through sensor calibration, the RGB image, the depth point cloud and the radar data are spatially aligned to form a multi-modal data frame that integrates color, depth and speed information.

[0067] Step S32, the processor extracts a feature sequence containing body joint angle, hand orientation and motion speed based on the multi-modal data frame.

[0068] In some specific embodiments, in step S32, the processor uses a pre-trained deep learning model to process each frame of RGB image, which will output the 2D pixel coordinates of several key points (usually 17-25, such as nose, neck, shoulder, elbow, wrist, hip, knee, ankle, etc.) of the human body. Optionally, the deep learning model includes any one of OpenPose, AlphaPose, and MediaPipe.

[0069] Optionally, in step S32, the processor converts these 2D pixel coordinates into 3D world coordinates in combination with the information of the depth camera to obtain the skeletal joint positions of the worker in the real space; then performs feature calculation, mainly including calculating the body joint angle, the hand orientation and the motion speed, which specifically includes: Body joint angle: Calculate the angle between two bone lines formed by connecting two joints; for example: elbow joint angle = angle between vector(shoulder->elbow) and vector(elbow->wrist). If the angle is less than 90 degrees, it can be inferred that the arm is bending, which may be taking goods.

[0070] Hand orientation: Calculate the direction of the normal vector formed by the hand key points (such as wrist, palm, and fingertips), and determine whether the hand is facing down (possibly relaxed), forward (possibly pushing the cart), or towards the shelf (possibly taking goods).

[0071] Motion speed: Overall speed, obtained directly from radar data or calculated by dividing the displacement of the pelvis joint between consecutive frames by time; local speed, calculate the speed of joints such as hands and feet to determine the degree of urgency of the action.

[0072] In some embodiments, at step S32, the processor constructs a time sequence of features, arranges the above-mentioned features of multiple time steps in chronological order to form a multi-dimensional time sequence, which contains the dynamic change trend of personnel behavior and is the key to predicting future intentions.

[0073] Step S33, the processor inputs the feature sequence into the trained time sequence prediction model, and outputs a probability distribution representing the probability of the worker performing taking goods, walking, or entering an emergency state in the future. Optionally, at step S33, the processor selects a long short-term memory network (LSTM) or a Transformer model as the time sequence prediction model.

[0074] In some embodiments, at step S33, the processor inputs the above-constructed time sequence of features into the model, and the output of the model is a probability distribution vector representing the probability of the worker performing various predefined intentions in the future. In practical applications, the time sequence prediction model in step S33 is trained and deployed through the following process: training data, collect a large amount of warehouse worker operation video data, and manually label the behavior intentions; training process, train the LSTM or Transformer model on a GPU server using the labeled data to minimize the difference between the prediction results and the true labels; deployment and reasoning, after the trained model is lightened, it is deployed on the edge computing device of the forklift or the edge server of the warehouse to meet the low latency requirement of real-time reasoning.

[0075] According to the above method, the forklift can accurately predict the worker's behavior intention in real time, and the forklift can be controlled to avoid collisions and improve work efficiency. Figure 4In the illustrated embodiment, the scheme proposed by the present application increases the proactive perception capability, jumps from passive perception of the current position of personnel to active prediction of the future behavior of personnel, so that the dispatch system can plan in advance, actively avoid or prepare for cooperation, change the safety control from emergency stop after collision to avoidance before predicting the risk, change the efficiency optimization from reducing waiting to pre-coordination, and realize the leap from passive response to active adaptation.

[0076] Figure 5 For Figure 1 the flowchart of an embodiment S4a of step S4 in the embodiment 1000. As Figure 5 indicated, the embodiment S4a includes steps S41-S44.

[0077] Step S41, the processor calculates the task-forklift adaptation score based on the matching degree of the task feature vector and the work ability attributes and economic attributes of the unmanned forklift. Optionally, in step S41, the task-forklift adaptation score is a quantitative indicator of the inherent ability and economy of the candidate unmanned forklift to independently complete the task.

[0078] In some specific embodiments, the processor calculates a matching degree score for each combination of task requirements and forklift capabilities, establishes a matching degree matrix including all combinations, and then calculates the task-forklift adaptation score based on the matching degree matrix. Specifically, the task-forklift adaptation score is obtained by weighted summation of six dimension indicators of load matching degree, lifting matching degree, tool matching degree, channel adaptability, navigation accuracy matching degree, and economic matching degree. The quantification of each dimension indicator is obtained by matching degree definition rules.

[0079] Optionally, an embodiment of a set of matching degree definition rules includes: the load matching degree is calculated by , perfect match is 1, overload is 0; the lifting matching degree is 1 if the forklift lifting height is greater than or equal to the task requirement height, otherwise it is 0; the tool matching degree is 1.0 if there is the required tool, otherwise it is 0.0; the channel adaptability, ; the navigation accuracy matching degree is matched according to the task requirement docking accuracy and the forklift navigation accuracy, completely satisfied is 1.0, partially satisfied is 0.5, completely unsatisfied is 0.0; the economic matching degree is calculated by , the cost of the forklift closest to 1, the highest to 0.

[0080] Step S42, the processor calculates the forklift-personnel coordination score based on the matching degree of the work ability attributes of the unmanned forklift and the permission skill label information of the staff, and the matching degree of the work ability attributes and the predicted behavior intention.

[0081] In some embodiments, at step S42, the processor calculates the forklift- worker collaboration score based on the work capacity attributes of the unmanned forklift, the permission skill label of the worker, and the predicted behavior intention, through the following processes: permission safety verification, checking whether the worker has permission to enter the area where the task is located, if not, the collaboration score is directly 0 to avoid safety hazards; skill-capability matching degree, calculating the matching degree of the worker's skill level and the required skill of the task; intention-function matching score, according to the predicted worker behavior intention, evaluating whether the function of the forklift supports optimal collaboration; and finally obtaining the forklift-worker collaboration score by synthesizing the permission verification, skill-capability matching degree, and intention-function matching score.

[0082] At step S43, the processor calculates the friendliness degree of the environment based on the real-time perceived environmental information to obtain an environmental adaptation score. Alternatively, at step S43, the processor calculates the dynamic friendliness degree of the environment according to the real-time perceived environmental information.

[0083] In some embodiments, at step S43, the processor obtains the environmental adaptation score through the following processes: congestion index, calculating the current average moving obstacle density on the task path congestion = min(1, current density / maximum acceptable density); environmental risk index, checking whether there are temporary static obstacles, slippery ground, etc. on the path, and if so, the risk value increases; and synthesizing the congestion index and the environmental risk index to calculate the environmental adaptation score.

[0084] At step S44, the processor calculates the collaboration adaptation degree based on the following formula: collaboration adaptation degree = task-forklift adaptation score × A + forklift-worker collaboration score × B + environmental adaptation score × C, where A, B, and C are preset weight coefficients. Alternatively, at step S44, the processor considers the multi-dimensional indicators obtained in steps S41-S43 to obtain a collaboration adaptation degree value that can comprehensively reflect the adaptation degree.

[0085] According to the embodiment as shown in Figure 5 , the present application integrates heterogeneous information of four dimensions of task, forklift, worker, and environment into a unified, comparable quantitative score, providing a core decision basis for objectively selecting the globally optimal solution from numerous candidate schemes.

[0086] Figure 6 For the embodiment 1000 as shown in Figure 1 , the flowchart of the embodiment S4b of step S4 is shown. As shown in Figure 6 , the embodiment S4b includes steps S45-S49, wherein steps S45-S48 and steps S41-S44 in the embodiment S4a are the same and will not be repeated here. Figure 5

[0087] ​Step S49, the initial values of the weight coefficients A, B and C are set based on the equalization starting principle, and the processor dynamically adjusts the numerical values of the weight coefficients A, B and C according to the task characteristic vector of the task to be scheduled, so as to realize switching between different scheduling targets of cost priority, efficiency priority or safety priority. Alternatively, in step S49, the processor focuses on the urgency and cooperation demand degree in the task characteristics to realize dynamic adjustment of the weight coefficient values.

[0088] Alternatively, one embodiment of the adjustment strategy in step S49 includes: The scheduling target is the efficiency priority mode, at this time the task urgency is high, and the weight distribution is A = 0.6, B = 0.3 and C = 0.1; the allocation logic under this target is that the task-forklift adaptation is the most important, the forklift itself needs to be concerned about whether it can complete the task quickly, the cooperation efficiency is less important, and the environmental impact is the least important; The scheduling target is the cost priority mode, at this time the task urgency is low and the cooperation demand degree is high, the weight distribution is A = 0.3, B = 0.6 and C = 0.1; the allocation logic under this target is that the resource utilization rate is optimized, the total cost is reduced, the cooperation efficiency is the most important, the forklift performance is less important, and the environmental impact is the least important; The scheduling target is the safety priority mode, at this time the environmental risk is high or the intention of the personnel is unstable, the weight distribution is A = 0.2, B = 0.2 and C = 0.6; the allocation logic under this target is that the environment is extremely safe, and the other two factors are equally important.

[0089] According to the embodiment as shown in Figure 6 , the scheme of the present application dynamically adjusts the priority of the optimization target according to the task characteristics, so that a set of algorithms can be flexibly adapted to various different business scenarios, realizes the task-adapted scheduling strategy, and greatly enhances the practicality and flexibility of the scheme.

[0090] Figure 7 A flowchart of an embodiment 2000 of the unmanned forklift human-machine cooperative optimization scheduling method of the present application is shown in Figure 7 , the embodiment 2000 includes steps S201-S207, wherein steps S201-S205 and Figure 1 the steps S1-S5 in the embodiment 1000 are the same, and will not be described here.

[0091] Step S206, during the task execution process, the processor stores the continuously perceived environmental information and the predicted behavior intention with a first period value. Optionally, the first period value in step S206 defines the frequency of the safety monitoring cycle, ensuring that the processor can respond to sudden conditions at a fast enough speed to meet the response time requirements in functional safety. For example, the first period value is 100 ms.

[0092] In some embodiments, in step S206, the processor aggregates the perceived environmental information and the predicted behavior intention in each period, and caches them in the form of a sliding time window in the memory to form a short-term storage for trend analysis and accident traceability, providing time series data for the safety risk assessment in step S207.

[0093] Step S207, the processor performs safety risk assessment on the environmental information and the behavior intention, and triggers adjustment strategies according to the assessment results. Optionally, in step S07, the processor triggers a safety priority adjustment strategy when the environmental information or the behavior intention meets the safety warning condition. Optionally, the processor triggers an efficiency priority adjustment strategy when the environmental information or the behavior intention meets the efficiency optimization condition, recalculates the collaborative adaptation degree and assigns new unmanned forklift takeover tasks.

[0094] According to the embodiment as shown in Figure 7 , the present application converts static, pre-set safety rules into dynamic, real-time situation-based risk assessment by continuously monitoring changes in the environment and intention, and triggers a hierarchical response strategy, ensuring that not only is it safe in the planning stage, but also in the execution process. It can also continuously maintain safety, and the response to unexpected events is more accurate and flexible, thereby maximizing job continuity while ensuring safety.

[0095] Figure 8 For Figure 2 the flowchart of step S207 in embodiment 2000. As shown in Figure 8 , step S207 includes step S2071-step S2072.

[0096] Step S2071, the processor derives a real-time risk index according to the environmental information and the behavior intention.

[0097] In some embodiments, in step S2071, the processor continuously calculates a real-time risk index according to the environmental information and the behavior intention, which is determined by the maximum value of the behavior intention risk score, the regional intrusion risk score, and the device state risk score.

[0098] Optionally, the behavior intention risk score in step S2071 is determined by the behavior intention predicted in step S203, a basic risk value is assigned to each predicted intention based on the change of the behavior intention, and then the behavior intention risk score is calculated by the product of the basic risk value and the behavior intention confidence. For example, the predicted behavior intention is running, the corresponding basic risk value is 0.8, and the confidence of the current intention is 0.85, and the obtained behavior intention risk score is the product of the two, which is 0.68.

[0099] Optionally, the area intrusion risk score in step S2071 is obtained from the relative position of the personnel / equipment in the environment information and the electronic fence. In some specific embodiments, the map of the working area is marked with different risk levels respectively, for example, divided into high-risk area, medium-risk area and low-risk area, and the area intrusion risk is triggered when the working personnel or equipment enters the high-risk or non-permission area. Optionally, the final area intrusion risk score is obtained by the product of the area inherent risk level and the intrusion depth coefficient, wherein the intrusion depth coefficient is calculated according to the distance of the safety boundary, and the farther the distance, the deeper the intrusion, and the higher the coefficient value.

[0100] Optionally, the equipment state risk score in step S2071 is obtained from the state code uploaded by the self-checking system of the unmanned forklift, and the processor maps the specific fault type to the risk value according to the preset equipment fault risk mapping table.

[0101] Step S2072, in the case that the real-time risk index meets the safety warning condition, the processor triggers the safety priority adjustment strategy. Step S2073, in the case that the real-time risk index meets the efficiency optimization condition, the processor triggers the efficiency priority adjustment strategy, recalculates the cooperative adaptation degree and assigns a new unmanned forklift to take over the task.

[0102] Optionally, in steps S2072 and S2073, the processor performs threshold triggering judgment on the real-time risk index in step S2071. For example, if the value of the real-time risk index is greater than or equal to the first risk threshold, the safety priority adjustment strategy is triggered; if the value of the real-time risk index is less than the first risk threshold and the equipment state risk score is higher than the second risk threshold, the efficiency priority adjustment strategy is triggered.

[0103] Optionally, the safety priority adjustment strategy in step S2072 includes: controlling the relevant unmanned forklifts to immediately enter the emergency stop or low-speed slow running state; broadcasting warning information to the unmanned forklifts and working personnel in the surrounding area, and dynamically updating the electronic fence area.

[0104] Optionally, the efficiency priority adjustment strategy in step S2073 includes: marking the equipment with an equipment state risk score higher than the second risk threshold as an unreliable equipment, recalculating the cooperative adaptation degree and assigning a new unmanned forklift to take over the task.

[0105] Figure 9 FIG. 3 is a structural schematic diagram of an embodiment 3000 of the human-robot collaborative optimization scheduling system of the present application. As shown in FIG. 3, the embodiment 3000 includes an information acquisition module 301, a task feature extraction module 302, a perception prediction module 303, a collaborative adaptation degree calculation module 304, a scheduling decision module 305, and a safety closed-loop control module 306. Figure 9

[0106] In some embodiments, the information acquisition module 301 acquires the to-be-scheduled task through an upper management system and communicates between systems through a pre-defined application program interface. Optionally, the upper management system is a warehouse management system (WMS) or a manufacturing execution system (MES). The information acquisition module 301 calls the interface provided by the WMS / MES to acquire the task queue in a polling or push manner and acquires the original instruction of the task.

[0107] In some embodiments, the information acquisition module 301 acquires the resource portrait information of the unmanned forklift from a local resource management system, and acquires all forklift records that are online and in an idle or about-to-be-idle state by querying a database table.

[0108] In some embodiments, the information acquisition module 301 acquires the permission and skill tag information of the staff through a human resource system or a warehouse management system, and achieves this in a manner of acquiring all on-duty and available staff information through an API interface or querying a local database.

[0109] In some embodiments, the task feature extraction module 302 extracts the task features of the to-be-scheduled task, which include the quantification of the task priority.

[0110] In some embodiments, the task feature extraction module 302 constructs a task feature vector based on the to-be-scheduled task, which includes the urgency, the human-robot collaboration demand, the material features, the operation area density, and the time window constraint; and then calculates multiple dimensions of the task feature vector through a weighted summation formula to generate a quantification score of the task priority.

[0111] Optionally, the calculation of the human-robot collaboration demand by the task feature extraction module 302 includes: preliminarily classifying the to-be-scheduled task based on a pre-set collaboration rule library to obtain a first collaboration demand value corresponding to the task collaboration type; first fine-tuning the first collaboration demand value based on a material master database to obtain a second collaboration demand value; and second fine-tuning the second collaboration demand value based on historical task data to determine the final human-robot collaboration demand.

[0112] In some embodiments, the perception prediction module 303 predicts the behavior intention of the staff in the operation area based on real-time perceived environmental information.​

[0113] In some embodiments, the perception prediction module 303 completes the prediction of the personnel behavior intention by the following steps: obtaining the image sequence and motion data of the personnel through the multi-modal sensor; based on the image sequence and motion data, extracting a feature sequence containing body joint angle, hand orientation and motion speed; inputting the feature sequence into the trained time series prediction model, and outputting a probability distribution representing the personnel executing the picking, threading or entering the emergency state in the future period of time.

[0114] In some embodiments, the collaborative adaptation degree calculation module 304 calculates the collaborative adaptation degree based on the task characteristics, resource portrait information, permission skill label information and predicted behavior intention, to quantify the degree of adaptation of the candidate unmanned forklift and the worker to execute the to-be-scheduled task.

[0115] In some embodiments, the collaborative adaptation degree calculation module 304 calculates the collaborative adaptation degree by the following steps: based on the matching degree of the task characteristics and the work ability attributes and economic attributes of the unmanned forklift, a task-forklift adaptation score is calculated; based on the matching degree of the work ability attributes of the unmanned forklift and the permission skill label information of the worker, as well as the predicted behavior intention, a forklift-personnel collaboration score is calculated; based on the real-time perceived environmental information, the friendliness of executing the to-be-scheduled task is calculated, and an environmental adaptation score is calculated; the collaborative adaptation degree is calculated based on the following formula: collaborative adaptation degree = task-forklift adaptation score x A + forklift-personnel collaboration score x B + environmental adaptation score x C, wherein A, B and C are preset weight coefficients.

[0116] The scheduling decision module 305 assigns target unmanned forklifts and target workers to the to-be-scheduled tasks according to the collaborative adaptation degree, and generates control instructions.

[0117] In some embodiments, the specific process executed by the scheduling decision module 305 includes: First, a candidate triple list is generated, including all possible task-unmanned forklift-worker triple combinations, and for each triple combination, the collaborative adaptation degree score of each combination is calculated; then, resource conflicts of these candidate combinations are detected and globally optimized and sorted. For example, the same forklift or the same worker may appear in multiple high-score candidate triples at the same time. Alternatively, in the global optimization and sorting process, the goal of the scheduling decision module 305 is not to find the best allocation for a single task, but to find a globally optimal or near-optimal allocation scheme for all to-be-scheduled tasks at present.

[0118] The scheduling decision module 305 selects a suitable algorithm according to the size of the current working scene, including: for small and medium-sized scenes, the Hungarian algorithm or the minimum cost maximum flow algorithm can be used to find the global optimal solution, and to ensure that the sum of the total collaborative adaptation degree scores of all tasks is the highest; for large-scale real-time scenes: a variant of the greedy algorithm is used. That is: all candidate triplets are sorted in descending order of collaborative adaptation degree scores, and then assigned in order. Once a resource is assigned, all triplets containing the resource are removed from the candidate list.

[0119] In some embodiments, after the allocation scheme is determined, the scheduling decision module 305 sends structured control instructions to the target unmanned forklift and the target worker. Optionally, the control instructions for the target unmanned forklift include the cooperative worker ID, the personnel coordinates, and the cooperation mode. Optionally, the control instructions for the target worker include the action to be performed, the cooperative forklift ID, the forklift expected arrival time, and the prompt text / map.

[0120] In some embodiments, the safety closed-loop control module 306 stores the continuously perceived environmental information and the predicted behavior intention with a first cycle value, then performs safety risk assessment on the environmental information and the behavior intention, and triggers adjustment strategies according to the assessment results to complete closed-loop control.

[0121] The above describes the embodiments of the present application in detail, and the specific examples are applied to explain the principles and implementation modes of the present application. The above descriptions of the embodiments are only used to help understand the method of the present application and its core idea. Meanwhile, the changes or deformations made by the skilled in the art according to the idea of the present application, based on the specific implementation modes and application scope of the present application, all belong to the scope of protection of the present application. In summary, the content of the present description should not be understood as a limitation of the present application.

Claims

1. A human-machine collaborative optimization scheduling method for unmanned forklifts, characterized in that, Includes the following steps: Step S1: Obtain the resource profile information of the task to be scheduled, the unmanned forklift fleet, and the permission and skill tag information of the staff group. The resource profile information includes at least the forklift type, operation capability attribute, and economic attribute of multiple unmanned forklifts in the unmanned forklift fleet. The permission and skill tag information includes at least the security permissions and skill levels of multiple staff members in the staff group. Step S2: Extract the task feature vector of the task to be scheduled, wherein the task feature vector includes a quantification of task priority; Step S3: Based on real-time perceived environmental information, predict the behavioral intentions of the workers in the work area; Step S4: Based on the task characteristics, resource profile information, permission skill tag information, and predicted behavioral intent, calculate the collaborative adaptability of all task-unmanned forklift-worker combinations; the collaborative adaptability is used to quantify the degree of adaptability of candidate unmanned forklift-worker combinations to execute the scheduled task. Step S5: Assign target unmanned forklifts and target workers to the task to be scheduled based on the cooperative adaptation degree, and generate control commands.

2. The method according to claim 1, characterized in that, Step S2 includes: Based on the tasks to be scheduled, a task feature vector is constructed, which includes urgency, human-machine collaboration requirements, material characteristics, work area density, and time window constraints. The task feature vector is calculated using a weighted summation formula across multiple dimensions to generate a quantitative score for the task priority. The weights of each task feature vector in the weighted summation formula are determined based on a work requirement strategy template, which includes a production line assurance template, a cost priority template, and an efficiency optimization template. The production line support template is determined based on the business objective of ensuring uninterrupted production and assembly lines. The cost-priority template is determined based on the business objective of minimizing operating costs while completing daily tasks; The efficiency flow template is determined based on the business objective of maximizing the overall throughput of the warehouse.

3. The method according to claim 2, characterized in that, The human-machine collaboration demand level is used to characterize the degree to which the task to be scheduled requires collaborative operation between personnel and unmanned forklifts, and is obtained according to the following steps: The tasks to be scheduled are initially classified based on the collaboration rule base to obtain a first collaboration requirement value corresponding to the task collaboration type. Based on the material master database, the attribute increment values ​​of the materials in the task to be scheduled are obtained, and the first collaboration demand value is fine-tuned based on the attribute increment values ​​to obtain the second collaboration demand value. Based on historical task data, the historical average collaboration intensity is determined, and the second collaboration demand value is fine-tuned a second time according to the historical average collaboration intensity to determine the final human-machine collaboration demand.

4. The method according to claim 1, characterized in that, Step S4 includes: The matching degree between the task feature vector and the unmanned forklift's operational capability and economic attributes is calculated. Specifically, the task-forklift matching score is obtained by weighted summation of six dimensions: load matching degree, lifting matching degree, attachment matching degree, channel adaptability, navigation accuracy matching degree, and economic matching degree. The quantification of each dimension is obtained through matching degree definition rules. Based on the matching degree between the unmanned forklift's operational capability attributes and the worker's permission skill tag information, as well as the matching degree between the operational capability attributes and the predicted behavioral intentions, the forklift-personnel collaboration score is calculated. Based on the real-time perceived environmental information, the friendliness of executing the scheduled task is calculated, and an environmental adaptation score is obtained. The collaboration adaptability is calculated based on the following formula: Collaboration adaptability = Task - Forklift adaptability score × A + Forklift - Personnel collaboration score × B + Environment adaptability score × C, where A, B, and C are preset weighting coefficients.

5. The method according to claim 4, characterized in that, Step S4 further includes: The initial values ​​of the weight coefficients A, B, and C are set based on the principle of balanced start. The values ​​of the weight coefficients A, B, and C are dynamically adjusted according to the task feature vector of the task to be scheduled, so as to achieve switching between different scheduling objectives such as cost priority, efficiency priority, or safety priority.

6. The method according to claim 1, characterized in that, Step S3 includes: The multimodal data frames of the staff were acquired using a multimodal sensor; Based on the multimodal data frames, feature sequences are extracted, which include information on body joint angles, hand orientation, and movement speed. The feature sequence is input into a trained time-series prediction model, which outputs a probability distribution representing the staff member's actions of retrieving goods, traversing, or entering an emergency state within a future period.

7. The method according to claim 1, characterized in that, Also includes: During task execution, the continuously perceived environmental information and the predicted behavioral intent are stored in the first period value. A security risk assessment is performed on the environmental information and the behavioral intent, and an adjustment strategy is triggered based on the assessment results.

8. The method according to claim 7, characterized in that, The process of conducting a security risk assessment of the environmental information and the behavioral intent, and triggering corresponding adjustment strategies based on the assessment results, includes: A real-time risk index is derived based on the environmental information and the behavioral intent. If the real-time risk index meets the safety warning conditions, a safety priority adjustment strategy will be triggered. If the real-time risk index meets the efficiency optimization conditions, the efficiency priority adjustment strategy is triggered, the collaborative adaptability is recalculated, and a new unmanned forklift takeover task is assigned.

9. The method according to claim 8, characterized in that, The security priority adjustment strategy includes: The control system immediately puts the unmanned forklift into an emergency stop or low-speed slow-moving state. The system broadcasts warning information to the unmanned forklifts and staff in the surrounding area and dynamically updates the electronic fence area.

10. A human-machine collaborative optimization scheduling system for unmanned forklifts, characterized in that, include: The information acquisition module is used to acquire resource profile information of the task to be scheduled and the unmanned forklift fleet, as well as the permission and skill tag information of the staff group. The resource profile information includes at least the forklift type, operation capability attribute and economic attribute of multiple unmanned forklifts in the unmanned forklift fleet, and the permission and skill tag information includes at least the security permissions and skill levels of multiple staff members in the staff group. The task feature extraction module is used to extract the task feature vector of the task to be scheduled, wherein the task feature vector includes a quantification of the task priority; The perception and prediction module is used to predict the behavioral intentions of workers in the work area based on real-time perceived environmental information. The collaborative adaptation calculation module is used to calculate the collaborative adaptation of all task-unmanned forklift-worker combinations based on the task characteristics, resource profile information, permission skill tag information, and predicted behavioral intent; the collaborative adaptation is used to quantify the degree of adaptation of candidate unmanned forklift-worker combinations to execute the scheduled task. The scheduling decision module is used to allocate target unmanned forklifts and target workers to the task to be scheduled based on the cooperative adaptation degree, and generate control commands; The safety closed-loop control module is used to dynamically assess safety risks during task execution based on continuously perceived environmental information and predicted behavioral intentions, and trigger corresponding adjustment strategies based on the assessment results.

Images