Intelligent scheduling method and system of mobile robot

Abstract

The invention discloses an intelligent scheduling method and system for a mobile robot, and relates to the technical field of robots. Extracting task basic attributes to form a static feature layer, establishing a dynamic feature layer formed by dynamic influence factors, and generating a double-layer task feature model; establishing a self-adaptive matching model of the robot and the task, performing feature fusion on the robot state data and the task feature matrix, and calculating a matching degree scoring matrix; a task allocation scheme is generated based on the matching degree scoring matrix, potential path conflict points are predicted, an avoidance strategy is determined, and conflict prediction and avoidance are executed; dynamically adjusting a task allocation and execution sequence according to the load state of the robot; and a performance closed-loop optimization framework is established, and scheduling strategy parameters are automatically adjusted. According to the method, the double-layer task feature model and the self-adaptive matching mechanism are constructed, and track prediction and conflict avoidance technologies are combined, so that intelligent matching and dynamic adjustment of the robot and the task are realized.

Description

Technical Field

[0001] The present invention relates to the field of robotics technology, and in particular to an intelligent scheduling method and system for mobile robots. Background Art

[0002] With the rapid development of industry and intelligent manufacturing, mobile robots are increasingly being used in factories, warehouses, and logistics centers. Existing mobile robot scheduling systems primarily rely on static task allocation and fixed path planning, with robots executing operations according to pre-set routes and task sequences. While these traditional scheduling methods operate reliably in simple environments, they exhibit significant limitations when faced with dynamically changing production environments and complex, ever-changing task requirements. In particular, in multi-robot collaborative operations, the lack of effective real-time coordination mechanisms often leads to path conflicts, resource competition, and uneven task allocation, resulting in overall system inefficiency.

[0003] Furthermore, existing robot scheduling technologies largely rely on manual intervention and empirical rules, making it difficult to adapt to environmental changes and task characteristics. When faced with emergencies or urgent task changes, the system responds slowly and cannot effectively replan task allocation. Furthermore, traditional scheduling systems lack real-time monitoring and dynamic balancing mechanisms for robot load status, resulting in some robots operating at high loads for extended periods while others remain idle. This not only reduces equipment efficiency but also accelerates wear and failure rates for high-load robots. Summary of the Invention

[0004] The present invention provides an intelligent scheduling method and system for mobile robots, which are used to solve the technical problems of intelligent task allocation, path planning and load balancing in a multi-robot collaborative operation environment.

[0005] In view of this, a first aspect of the present invention provides an intelligent scheduling method for a mobile robot, comprising:

[0006] Obtain scheduling scenario data, which includes robot status data, operating environment data, and scheduling task data;

[0007] Extract the basic attributes of the task to form a static feature layer, establish a dynamic feature layer composed of dynamic influencing factors, and generate a two-layer task feature model;

[0008] An adaptive matching model between robots and tasks is established based on machine learning algorithms. The robot status data is fused with the task feature matrix to calculate the matching score matrix.

[0009] Generate a task allocation plan based on the matching score matrix, use trajectory prediction methods to predict potential path conflict points and determine avoidance strategies, and perform conflict prediction and avoidance;

[0010] Dynamically adjust task allocation and execution order based on robot load status;

[0011] Establish a performance closed-loop optimization framework, regularly analyze system operation data, identify performance bottleneck factors, and automatically adjust scheduling strategy parameters.

[0012] Optionally, the robot status data includes location information, power information and task execution status; the working environment data includes obstacle distribution, personnel activities and temporary events in the working area; the scheduling task data includes task identification, task content description, expected completion time, and task resource requirements.

[0013] Optionally, the process of constructing a two-layer task feature model includes:

[0014] Processing scheduling task data to generate task description parameter sets, processing robot status data and working environment data to generate environment status parameter sets;

[0015] Extract the basic attributes of the task from the task description parameter set, perform multi-dimensional feature quantization encoding on the basic attributes of the task, and generate a static feature layer;

[0016] Analyze the dynamic influencing factors in the environmental state parameter set, calculate the timeliness evaluation index, environmental matching index and task coordination index, and build a dynamic feature layer;

[0017] The feature fusion algorithm of inter-layer weight transfer is used to establish the mapping relationship between the static feature layer and the dynamic feature layer, generate a two-layer task feature model, and output the task feature matrix.

[0018] Optionally, the process of building the adaptive matching model includes:

[0019] Based on the environmental state parameter set, the state evaluation score of the robot individual is calculated to generate the robot state feature representation;

[0020] Design a feature fusion module that inputs the robot state feature representation and the task feature vector matrix, performs cross-modal feature mapping, and outputs the interactive feature representation;

[0021] Design a multi-head attention module to calculate the importance weight of each dimension of the interactive feature representation, and generate a correlation feature matrix through weighted aggregation;

[0022] Construct a deep neural network consisting of graph convolutional networks and residual connections, take the correlation feature matrix as input, and extract high-order matching features through a multi-layer graph attention mechanism;

[0023] Based on the high-order matching features, the initial matching score of each robot-task combination is calculated;

[0024] Construct a global optimization function, calculate the global balance adjustment coefficient, use the global balance adjustment coefficient to correct the initial matching score, and obtain the matching score matrix.

[0025] Optionally, using a trajectory prediction method to predict potential path conflict points and determine an avoidance strategy includes:

[0026] An integer linear programming model is constructed based on the matching score matrix, resource constraints and task priority constraints are set, and the initial task allocation plan is obtained using the branch and bound algorithm.

[0027] Analyze each robot's historical motion data, extract motion pattern features, and predict the robot's future motion trajectory based on the currently assigned task target position;

[0028] Build a multi-robot trajectory interaction model, taking the future motion trajectories of all robots as input, calculate the spatiotemporal similarity matrix between trajectory pairs, identify trajectory intersections and overlapping path segments, and determine potential path conflict points, their probability of occurrence, and their estimated time of occurrence;

[0029] For potential path conflict points, a set of avoidance solutions is generated based on the conflict point information and the motion status of the relevant robots;

[0030] The Monte Carlo tree search algorithm is used to select the global optimal avoidance strategy combination from the avoidance scheme set and generate conflict avoidance execution instructions.

[0031] Optionally, dynamically adjusting task allocation and execution order according to the robot load state includes:

[0032] Establish a real-time load monitoring module to collect robot operation status data and calculate comprehensive load indicators;

[0033] Build a load threshold adaptive adjustment mechanism to calculate the upper and lower limits of the load threshold, and trigger task adjustment when the comprehensive load index exceeds the load threshold range;

[0034] For robots in high-load conditions, a task migration strategy is used to build a critical path analysis model based on the task dependency graph, identify non-critical tasks, calculate migration candidate scores, and perform task reallocation.

[0035] Build a dynamic task sorting algorithm, calculate the task time sensitivity score, and reorder the conflict avoidance execution order according to the task time sensitivity score;

[0036] Design task splitting and merging strategies, and dynamically adjust task structure based on task characteristics.

[0037] Optionally, adopting a task migration strategy to construct a critical path analysis model based on a task dependency graph includes:

[0038] Build a task dependency graph;

[0039] Apply the critical path algorithm to analyze the task dependency graph and calculate the time slack of each task;

[0040] Classify tasks into critical tasks and non-critical tasks based on time slack. Combined with the task resource requirements and execution complexity, calculate the migration candidate score for each non-critical task.

[0041] Search for robots in low-load state, build a candidate set of migration targets, and calculate the receiving capability score of each candidate robot;

[0042] Establish a task-robot matching evaluation matrix and calculate the transfer benefit value for each pair of candidate tasks and candidate robots;

[0043] Select the task-robot pair with the highest migration benefit value and perform the task migration operation.

[0044] A second aspect of the present invention provides an intelligent scheduling system for a mobile robot, comprising:

[0045] The data acquisition module is used to obtain scheduling scenario data, which includes robot status data, operating environment data and scheduling task data;

[0046] The task feature determination module is used to extract the basic attributes of the task to form a static feature layer, establish a dynamic feature layer composed of dynamic influencing factors, and generate a two-layer task feature model;

[0047] The adaptive matching module is used to establish an adaptive matching model between the robot and the task based on the machine learning algorithm, fuse the robot state data with the task feature matrix, and calculate the matching score matrix;

[0048] The task allocation and conflict avoidance module is used to generate a task allocation plan based on the matching score matrix, predict potential path conflict points using trajectory prediction methods, determine avoidance strategies, and perform conflict prediction and avoidance.

[0049] Load balancing module, used to dynamically adjust task allocation and execution order according to the robot load status;

[0050] The performance optimization module is used to establish a performance closed-loop optimization framework, regularly analyze system operation data, identify performance bottleneck factors, and automatically adjust scheduling strategy parameters.

[0051] The beneficial effects of the present invention are: the present invention constructs a comprehensive intelligent scheduling information acquisition framework, integrates the three-dimensional data of "executing subject-environmental scenario-task goal", comprehensively grasps the core elements of the scheduling problem, realizes the global perception of the scheduling scenario, and reduces the decision-making error rate caused by incomplete information; the established two-layer task feature model and adaptive matching model realize the transformation of the intelligent scheduling system from "static rule-driven" to "dynamic environment adaptation", so that robots can flexibly adjust decision-making strategies according to environmental changes and task characteristics; through trajectory prediction and conflict avoidance mechanism, the safety and smoothness of multi-robot collaborative operation are improved, the path planning problem in high-density robot environment is solved, and a technological leap from passive conflict handling to active conflict prevention is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 Flowchart of the intelligent scheduling method for mobile robots.

[0054] Figure 2 Construct a flow chart for a two-layer task feature model for an intelligent scheduling approach for mobile robots.

[0055] Figure 3 Construct a flow chart for the adaptive matching model of the intelligent scheduling method for mobile robots. DETAILED DESCRIPTION

[0056] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0057] Reference Figures 1 to 3 , is an embodiment of the present invention, which provides an intelligent scheduling method for a mobile robot. The flow chart of the method is as follows Figure 1 As shown, the method includes:

[0058] S1: Obtain scheduling scenario data.

[0059] Scheduling scenario data includes robot status data, operating environment data, and scheduling task data. Robot status data includes location information, power information, and task execution status; operating environment data includes obstacle distribution, human activity, and temporary events in the operating area; and scheduling task data includes task identification, task description, expected completion time, and task resource requirements.

[0060] Preferably, the present invention selects robot status data, working environment data and scheduling task data as components of the scheduling scenario data, forming a comprehensive scheduling information acquisition framework. This data selection has obvious advantages: the position information, power information and task execution status in the robot status data can accurately reflect the current capabilities and constraints of the execution subject; the obstacle distribution, personnel activities and temporary events in the working environment data can reflect the real-time changes in the work scene, enabling the scheduling system to adapt to environmental changes; the task identification, content description, expected completion time and resource requirements in the scheduling task data clarify the specific needs of the task. The combination of these three types of data realizes the information integration of "execution subject-environmental scenario-task goal", which can more comprehensively grasp the key factors of the scheduling problem and help improve the scheduling efficiency and adaptability of the system in practical applications.

[0061] S2: Based on the scheduling scenario data, the basic attributes of the task are extracted to form a static feature layer, and a dynamic feature layer composed of dynamic influencing factors is established to generate a two-layer task feature model.

[0062] In a specific embodiment of the present invention, the double-layer task feature model construction flow chart is as follows: Figure 2 Shown, including:

[0063] S2.1: Process the scheduling task data to generate a task description parameter set, and process the robot state data and working environment data to generate an environment state parameter set.

[0064] First, the scheduling task data is processed, including: encoding the task identifier; extracting keywords from the task content description to identify the task type and operation object; converting the expected completion time into a standard time format; converting the task resource requirements into numerical values; and integrating the above processing results to generate a task description parameter set.

[0065] Secondly, the robot status data and working environment data are processed, including: converting the robot's position information into a unified coordinate system, converting the power information into the remaining work capacity value, and converting the task execution status into a status code; processing the obstacle distribution data to obtain environmental accessibility data; processing personnel activity information and temporary event data; and integrating the above processing results to generate an environmental status parameter set.

[0066] S2.2: Extract the basic attributes of the task from the task description parameter set, perform multi-dimensional feature quantization encoding on the basic attributes of the task, and generate a static feature layer.

[0067] Specifically, the basic attributes of tasks are extracted from task description parameters, including at least the task type, operation object, priority coefficient, and task resource requirements. A multi-dimensional feature quantization encoding operation is performed on the extracted basic attributes of the tasks, converting the task type into a feature vector and mapping the operation object into a classification code. Time-related features are calculated based on the expected completion time data and converted into standardized feature values. Task resource requirements are numerically processed to form a resource demand feature vector. The dependency tags in the task identifier are analyzed to construct task-related features that reflect the constraints on the execution sequence of tasks. The above quantitative features are integrated to form a basic attribute representation of the tasks, and normalized to unify the magnitude of the features in each dimension, ultimately generating a static feature layer that represents the inherent attributes of the tasks. By capturing the essential characteristics and inherent attributes of the tasks, the static feature layer provides a stable basic feature representation for subsequent fusion with the dynamic feature layer, ensuring that scheduling decisions can accurately understand the basic requirements and constraints of the tasks.

[0068] S2.3: Analyze the dynamic influencing factors in the environmental state parameter set and construct a dynamic feature layer to characterize the real-time execution status of the task.

[0069] Furthermore, the environmental state parameter set is analyzed to identify and extract key environmental variables that influence task execution. An environmental matching index is generated by calculating the distance function between the robot's current position and the task execution location, combined with an obstacle distribution density map. A resource availability index is evaluated and generated using the ratio function of the robot's current power information and the task's estimated energy consumption. A real-time interference factor, representing the degree of external interference, is calculated based on task execution status data and a heat map of personnel activity. A task timeliness evaluation index is calculated based on the expected completion time and the current system time. Resource competition and collaboration relationships between multiple concurrently executed tasks are analyzed to generate a task synergy index. These indicators are integrated and a time-series weighted algorithm is applied to emphasize the influence of the latest environmental state, constructing a dynamic feature layer that represents the real-time execution state of the task. This dynamic feature layer continuously reflects the impact of environmental state changes on task execution, ensuring that scheduling decisions can adapt to complex and changing operating environments.

[0070] S2.4: Use the feature fusion algorithm of inter-layer weight transfer to establish the mapping relationship between the static feature layer and the dynamic feature layer, generate a two-layer task feature model, and output the task feature matrix.

[0071] Furthermore, the static feature layer and the dynamic feature layer are represented as feature matrices respectively, and the weight connection relationship between the two layers is defined; a feature fusion weight matrix is ​​designed to regulate the influence ratio of static features and dynamic features; the degree of environmental change is calculated according to the temporal change amplitude of the environmental state parameters, and the weight parameters are dynamically adjusted. When the environment is stable, the static feature weight is increased, and when the environment is changeable, the dynamic feature weight is increased; the product of the static feature vector and the weight matrix is ​​calculated to obtain the static feature contribution component; the product of the dynamic feature vector and the weight matrix is ​​calculated to obtain the dynamic feature contribution component; the two contribution components are weighted and summed to form a comprehensive feature representation; a nonlinear activation function is applied to process the comprehensive features to enhance the model's expressiveness; a task association matrix is ​​constructed to capture the mutual influence between tasks; task features and association information are integrated to generate a two-layer task feature model, and a task feature matrix that adapts to environmental changes is output.

[0072] Optimally, the two-layer feature structure effectively distinguishes between inherent task attributes and environmental dependencies, preventing environmental fluctuations from interfering with the essential characteristics of the task. The inter-layer weight transfer mechanism dynamically adjusts feature weights based on the degree of environmental change, enhancing the system's environmental adaptability. Multi-dimensional feature quantization and normalization enable standardized representation of heterogeneous tasks. The introduction of an inter-task correlation matrix enables the scheduling system to consider the overall benefits of task groups. This solution overcomes the technical bottleneck of traditional task representation methods, which struggle to balance static description and dynamic adaptation, and provides technical support for the efficient operation of robotic task scheduling in complex environments.

[0073] S3: Based on the machine learning algorithm, an adaptive matching model between the robot and the task is established, the robot state data is fused with the task feature matrix, and the matching score matrix is ​​calculated.

[0074] In a specific embodiment of the present invention, the adaptive matching model construction flow chart is as follows: Figure 3 As shown, specifically including:

[0075] S3.1: Based on the environmental state parameter set, calculate the state evaluation score of the robot individual and generate the robot state feature representation.

[0076] The status assessment scores include task fitness, resource utilization, and scheduling priority. Specifically, the task fitness score is calculated by calculating the degree of match between the robot's capability parameters and the task's requirements using a weighted cosine similarity method. The resource utilization score is calculated by evaluating the robot's remaining battery life, computing power utilization, and sensor load status. The scheduling priority score is calculated by comprehensively considering factors such as the distance between the robot's position and the task, task timeliness, system load balance, and historical task quality scores.

[0077] S3.2: Design a feature fusion module that inputs the robot state feature representation and the task feature vector matrix, performs cross-modal feature mapping, and outputs the interaction feature representation.

[0078] Specifically, the feature fusion module employs a two-stream network structure. First, the robot state features and task features are mapped to a feature space of the same dimension via two fully connected layers, achieving initial feature alignment. A tensor product operation is then used to capture the high-order cross-relationships between the two features. Specifically, a three-dimensional interaction tensor is generated by calculating the outer product, and residual connections are used to preserve the original feature information. Finally, normalization is applied to generate a dimensional unified interaction feature representation, effectively addressing the heterogeneity between robot and task features.

[0079] It should be noted that after the tensor product operation, Tucker decomposition is used to reduce the high-dimensional interaction tensor to an appropriate dimension. Simultaneously, the original features are linearly mapped and concatenated with the interaction features. Feature normalization is then performed to output a feature matrix of uniform dimension. This method effectively prevents the curse of dimensionality while preserving critical cross-feature information.

[0080] S3.3: Design a multi-head attention module to calculate the importance weights of features in each dimension in the interactive feature representation, and perform weighted aggregation to generate a correlation feature matrix.

[0081] It should be noted that the multi-head attention module first projects the input interactive feature representations into multiple feature subspaces using linear transformations, encompassing different semantic dimensions such as task type, resource requirements, and time constraints. Then, in each subspace, a query vector is generated using the task type encoding. An attention score is calculated with the features of that subspace to obtain feature importance weights. The weighted features are then nonlinearly transformed and normalized, and the feature representations of each subspace are merged and integrated into a unified correlation feature matrix through a fully connected layer. This matrix contains the core information required to predict the probability of task completion.

[0082] Optimally, the design of the multi-head attention module achieves comprehensive capture of feature relevance at different semantic levels, improving the discriminability and specificity of feature representation compared to traditional single-feature weighting methods. By processing multiple feature subspaces in parallel, each focusing on a different combination of semantic features, and dynamically adjusting attention distribution using task type encoding, the system can adaptively identify key matching factors for different task types, enhancing the model's adaptability to complex and changing task environments and resolving the existing problem of feature importance assessment being disconnected from specific task types.

[0083] S3.4: Construct a deep neural network consisting of a graph convolutional network and residual connections, take the correlation feature matrix as input, and extract high-order matching features through a multi-layer graph attention mechanism.

[0084] Furthermore, the correlation feature matrix is ​​first constructed as a bipartite graph structure, with robots and tasks set as two different types of nodes, and initial edge weights are set based on the strength of the correlation features. When the correlation feature strength exceeds a set threshold, a connection is established between the corresponding nodes to ensure the sparsity of the graph structure. The heterogeneous graph is then processed using a multi-layer graph convolutional network. Each layer includes operations such as feature aggregation, nonlinear transformation, residual connection, and layer normalization to extract deep correlation features between nodes layer by layer. Simultaneously, during the graph convolution process, edge weights are dynamically updated based on the feature representation of each layer, enabling the network to adaptively adjust the connection strength between nodes. Finally, node features are aggregated through an attention pooling layer to generate a high-order matching feature representation that embodies the complex robot-task interaction relationship.

[0085] It's important to note that in the node attribute design of the graph structure, robot nodes contain information about their state characteristics, and task nodes contain information about their required characteristics. Edge weights between nodes are initialized through feature similarity calculations, forming the basic structure for information transfer. This design ensures computational efficiency while providing a complete information foundation for subsequent feature extraction.

[0086] Preferably, this solution captures the multi-hop association relationship between nodes through graph neural networks, making up for the limitation of the S3.3 attention mechanism that mainly focuses on direct feature associations; by adding residual connections after each layer of graph convolution, the original feature information is retained, and the dynamic edge weight update mechanism is used to achieve adaptive adjustment of the feature extraction process, effectively improving the model's ability to express complex task scenarios.

[0087] S3.5: Based on the high-order matching features, calculate the initial matching score for each robot-task combination.

[0088] Furthermore, a multi-stage scoring model was designed, comprising a scenario adaptation layer, a feature evaluation layer, and a scoring output layer. First, in the scenario adaptation layer, the importance of high-order matching features is adjusted based on the current job scenario type. Then, in the feature evaluation layer, a decision tree structure is constructed to perform hierarchical evaluation of the adjusted features. Historical matching data is also introduced for comparative learning to enhance scoring accuracy. Finally, in the scoring output layer, a softmax function is used to generate a standardized matching score and calculate a confidence index.

[0089] Furthermore, the scoring results are organized into a matching score matrix, where each element contains two indicators: matching score and confidence, which represent the matching degree and scoring reliability of a specific robot-task combination, respectively.

[0090] Preferably, the scoring model realizes dynamic adjustment of feature weights through scene perception, provides fine-grained evaluation through hierarchical decision-making, improves scoring accuracy through comparative learning, and ensures scoring reliability through confidence assessment, thus forming a complete scoring framework.

[0091] S3.6: Construct a global optimization function based on the Lagrangian relaxation method, calculate the global balance adjustment coefficient, and use the global balance adjustment coefficient to correct the initial matching score to obtain the matching score matrix.

[0092] Specifically, an optimization objective function is first established that takes into account constraints such as robot resource capacity and task time window. The constraints are then transformed into a relaxed optimization problem through the Lagrange multiplier method. The optimization problem is then solved to obtain the optimal allocation solution and the corresponding Lagrange multiplier. Finally, the balance adjustment coefficient is calculated based on the solution result and multiplied with the initial matching score to obtain the final scoring matrix.

[0093] Preferably, the optimization method transforms hard constraints into soft constraints through Lagrangian relaxation technology, and achieves the unity of local optimality and overall optimality through global balance adjustment, providing an evaluation basis for task allocation that takes into account both matching effect and resource balance.

[0094] S4: Generate a task allocation plan based on the matching score matrix, use trajectory prediction methods to predict potential path conflict points and determine avoidance strategies, and perform conflict prediction and avoidance.

[0095] In a specific embodiment of the present invention, step S4 specifically includes:

[0096] S4.1: Construct an integer linear programming model based on the matching score matrix, set resource constraints and task priority constraints, and use the branch and bound algorithm to solve the initial task allocation plan.

[0097] Specifically, the matching score matrix is ​​converted into the objective function coefficients of the integer linear programming problem, and a decision variable matrix is ​​constructed, in which each element represents whether a specific robot is assigned to a specific task; resource constraints are established, including robot load capacity constraints, energy constraints and time window constraints, to ensure that the allocation plan meets the physical resource limitations; task priority constraints are set, and priority relationship constraint equations between tasks are established based on task urgency and importance indicators; task mutual exclusivity and dependency constraints are constructed to handle the prerequisite relationship and mutually exclusive execution requirements between tasks; the branch and bound algorithm is used to solve the integer linear programming model, and the optimal or approximately optimal task allocation plan is obtained through iterative search; the feasibility of the execution plan is verified, the executableness of the obtained allocation plan under resource constraints and time windows is evaluated, and an initial task allocation plan containing robot-task mapping relationship and execution time window is generated.

[0098] Advantageously, this invention addresses the inefficient multi-robot task allocation problem in existing technologies by employing a task allocation mechanism based on integer linear programming and a branch-and-bound algorithm. This mechanism achieves a globally optimal task allocation solution while satisfying multi-dimensional constraints such as resource constraints and task priority constraints. Compared to the greedy algorithms commonly used in existing technologies, this improves task completion efficiency and system resource utilization, enabling the entire system to maintain efficient operation even in complex environments.

[0099] S4.2: Analyze the historical motion data of each robot, extract the motion pattern features, and combine them with the currently assigned task target position to predict the robot's future motion trajectory.

[0100] Specifically, the historical motion data is first preprocessed and standardized to extract time series features such as position coordinate sequences and velocity vectors. Then, a prediction model based on the LSTM network is constructed to learn the robot's dynamic behavior characteristics such as acceleration, deceleration, and steering through an encoder-decoder structure. Finally, combined with the current task target position, the position coordinate sequence and velocity vector sequence within the specified time window are predicted to form a complete future trajectory representation.

[0101] S4.3: Construct a multi-robot trajectory interaction model based on a spatiotemporal graph. Take the future motion trajectories of all robots as input, calculate the spatiotemporal similarity matrix between trajectory pairs, identify trajectory intersections and overlapping path segments, and determine potential path conflict points, their occurrence probability, and estimated occurrence time.

[0102] Furthermore, the predicted trajectories of multiple robots are mapped to a unified spatiotemporal coordinate system to construct a spatiotemporal graph structure, in which nodes represent the position states of robots at specific time points, and edges represent the connection relationships in time and space; a spatiotemporal distance measurement function is designed, combining the spatial Euclidean distance and time difference to calculate the spatiotemporal distance between any two trajectory points; based on the spatiotemporal distance, the spatiotemporal similarity matrix between trajectory pairs is calculated to quantify the proximity of any two trajectories at different time points; cluster analysis and spatial indexing techniques are applied to identify high-similarity regions from the spatiotemporal similarity matrix to determine potential trajectory intersections and overlapping path segments; for each identified intersection and overlapping region, the actual conflict risk score is calculated based on the robot's physical size and safety distance parameters; a probabilistic model is established to calculate the probability of occurrence and expected time of occurrence of each potential conflict point based on the confidence interval and uncertainty estimation of the trajectory prediction.

[0103] S4.4: For potential path conflict points, generate a set of avoidance solutions based on the conflict point information and the motion states of the relevant robots.

[0104] Furthermore, for each potential conflict point, the current state parameters of the relevant robots are extracted, including speed, acceleration capability, energy level and task priority; based on the time window adjustment strategy, the arrival time offset of the relevant robots is calculated, and a time adjustment plan is generated to enable the robots to pass through the conflict point at off-peak times; a speed adjustment optimization model is constructed to calculate the optimal deceleration or acceleration curve of the robot before approaching the conflict point, and a speed adjustment plan is generated; using the heuristic path planning algorithm, alternative paths are generated for the relevant robots to bypass the conflict point, and a path replanning plan is constructed; a comprehensive evaluation is conducted on the generated avoidance plans, and the energy consumption, delay cost and safety margin of each plan are calculated. Taking into account the robot constraints and task time requirements, the optimal feasible plan set is screened; the spatiotemporal correlation between multiple conflict points is analyzed to ensure that the avoidance plans for different conflict points are coordinated with each other to form a globally consistent avoidance strategy.

[0105] S4.5: Use the Monte Carlo tree search algorithm to select the globally optimal avoidance strategy combination from the avoidance solution set, comprehensively consider the avoidance cost, task delay impact and overall system efficiency, and generate conflict avoidance execution instructions.

[0106] Furthermore, a decision tree structure is constructed, in which each node represents the avoidance strategy selection of a conflict point, and each path represents a complete avoidance strategy combination; a comprehensive evaluation function is defined, integrating multi-dimensional indicators such as avoidance cost, energy consumption, task delay time and system throughput, to quantify the global benefits of the avoidance strategy combination; a Monte Carlo tree search process is performed to explore the avoidance strategy combination space through four stages: selection, expansion, simulation and backtracking; in the simulation stage, the system dynamics model is used to predict the system state evolution after the execution of the avoidance strategy and evaluate the long-term impact; based on the search results, the globally optimal avoidance strategy combination is identified, and detailed avoidance instructions are generated including execution timing, parameter configuration and trigger conditions; the avoidance execution instructions are encoded into a command format recognizable by the robot control system and prepared for distribution to relevant robots for execution.

[0107] Optionally, set an upper limit on the search time to ensure real-time requirements and dynamically adjust the search depth and breadth; build a policy execution monitoring mechanism to evaluate the execution effect in real time and trigger re-planning when necessary; design emergency avoidance strategies for emergencies to ensure system security and robustness.

[0108] The Monte Carlo tree search algorithm implemented in this invention for avoidance strategy optimization fundamentally addresses the technical challenge of avoidance strategies frequently falling into local optima in existing technologies. This algorithm, through an exploration-exploitation balancing mechanism and forward-looking decision-making planning, achieves the global optimal selection of avoidance strategies. This method balances current avoidance costs with the efficiency of subsequent tasks, reduces system deadlock, and improves overall task completion efficiency and system throughput, making it particularly suitable for high-density robot environments. This method exhibits strong environmental adaptability and scalability, enabling rapid response to dynamic changes and maintaining stable system operation.

[0109] S4.6: Establish a hierarchical feedback control mechanism to monitor the deviation between the robot's actual motion state and the predicted trajectory in real time. When the deviation exceeds the preset threshold, trajectory re-prediction and conflict re-evaluation are triggered to ensure the dynamic adaptability of the conflict avoidance strategy.

[0110] S5: Dynamically adjust task allocation and execution order according to the robot load status to achieve adaptive load balancing.

[0111] In a specific embodiment of the present invention, step S5 specifically includes:

[0112] S5.1: Establish a real-time load monitoring module to collect robot operation status data and calculate comprehensive load indicators.

[0113] Among them, the comprehensive load index is calculated by the weighted sum of the robot's current CPU utilization, memory occupancy, power consumption rate and task queue depth. The weight of each parameter is dynamically adjusted according to the system operation optimization goal.

[0114] S5.2: Build a load threshold adaptive adjustment mechanism to calculate the upper and lower limits of the load threshold based on historical load distribution and task execution performance data. When the comprehensive load index exceeds the load threshold range, the task adjustment is triggered.

[0115] It should be noted that the adaptive adjustment of the load threshold adopts the sliding time window analysis method. By evaluating the load fluctuation pattern and task completion quality in the last N time windows, the threshold is dynamically updated to improve the system's adaptability in different working environments.

[0116] S5.3: For robots in high-load states, a task migration strategy is used to build a critical path analysis model based on the task dependency graph, identify non-critical tasks, calculate migration candidate scores, and perform task reallocation.

[0117] Specifically, based on the task dependencies in the current task queue, a task dependency graph is constructed, where nodes represent tasks and edges represent the execution order dependencies between tasks. The dependencies are determined by analyzing task input and output, spatial location associations, and timing constraints. The critical path algorithm is applied to analyze the task dependency graph, and the earliest start time (EST), latest start time (LST), and time slack of each task are calculated. Slack is calculated by subtracting EST from LST, which represents the maximum time that the task can be delayed. Tasks are classified into critical tasks (time slack is 0) and non-critical tasks (time slack is greater than 0) based on the time slack value. Combined with the task resource demand characteristics and execution complexity, a migration candidate score is calculated for each non-critical task. The migration candidate score is calculated according to the following formula:

[0118]

[0119] Among them, M j is the migration candidate score of task j, S j is the time slack value of task j, R j is the resource requirement of task j, R max is the maximum value of task resource requirements in the system, P j is the priority of task j, P max is the maximum value of the task priority in the system, α1, α2, and α3 are weight coefficients and satisfy α1+α2+α3=1, which represent the weights of time slack, resource demand, and task priority, respectively.

[0120] Furthermore, we search for robots in a low-load state, construct a set of migration target candidates, and calculate the receiving capacity score of each candidate robot. The receiving capacity score comprehensively considers the robot's remaining resources, the length of the current task queue, and the distance to the target location of the task to be migrated. The score calculation adopts a standardized weighted method, converting each factor into a comparable indicator and then performing a weighted combination. The weight configuration can be adjusted according to the actual application scenario to optimize the overall performance of the system. A task-robot matching evaluation matrix is ​​established, and for each pair of candidate tasks and candidate robots, the migration benefit value is calculated. The specific formula is as follows:

[0121] G ij =M j ·A i -C ij

[0122] Among them, G ij is the benefit of migrating task j to robot i, M j is the migration candidate score of task j, A i Score the receiving ability of robot i, C ij is the migration cost (including communication overhead and delay loss).

[0123] Furthermore, a greedy algorithm is used to select the task-robot pair with the highest benefit value from the migration evaluation matrix, perform task migration operations, and update the task allocation plan and robot load status. The migration operation includes removing the target task from the source robot task queue, sending task information to the target robot, confirming the receipt status, and updating the system resource allocation table.

[0124] In addition, it should be noted that performing task reallocation means reallocating the task with the highest migration candidate score to the robot in low load state. The robot in high load state refers to the robot whose comprehensive load index exceeds the upper limit of the load threshold, and the robot in low load state refers to the robot whose comprehensive load index is lower than the lower limit of the load threshold.

[0125] Preferably, the present invention realizes the intelligent task migration of robots in high-load state by constructing a critical path analysis model based on the task dependency graph. First, the solution breaks through the limitations of traditional task allocation strategies, introduces task dependency relationships to construct a directed acyclic graph and applies the critical path algorithm, realizes the accurate calculation of task time slack, effectively distinguishes between critical tasks and non-critical tasks, and provides a theoretical basis for task migration. Secondly, an innovative migration candidate score calculation method that comprehensively considers time slack, resource requirements and task priority is proposed, realizing a multi-dimensional task evaluation system and ensuring the scientific selection of migration tasks. At the same time, the greedy algorithm is used to select the task-robot pair with the highest benefit to perform migration, which not only effectively alleviates the problem of load imbalance of the robot system, but also improves the overall task processing efficiency, enhances the stability and reliability of the system, and provides strong technical support for the collaborative operation of multi-robot systems in complex environments.

[0126] S5.4: Construct a dynamic task sorting algorithm, calculate the task time sensitivity score, and reorder the conflict avoidance execution order based on the task time sensitivity score.

[0127] Specifically, a comprehensive analysis is conducted on the task deadline urgency, task priority, task dependency, and current system load status to calculate the time sensitivity score of each task; for the conflict avoidance execution instruction set generated by S4.5, the time sensitivity scores of the instruction-related tasks are extracted, an instruction-score mapping table is established, and the avoidance instructions are sorted in descending order according to the time sensitivity scores of the related tasks; a conflict avoidance instruction priority adjustment rule is designed. When there is an execution conflict between the avoidance instruction of a high-time-sensitive task and the avoidance instruction of a low-time-sensitive task, the avoidance instruction of the high-sensitivity task is executed first, and the avoidance strategy of the low-sensitivity task is adjusted if necessary; an avoidance instruction execution feedback mechanism is established to monitor the execution effect of the avoidance instruction in real time. When it is detected that the avoidance effect is poor or the system status has changed significantly, the Monte Carlo tree search algorithm in S4.5 is triggered to re-evaluate the avoidance strategy and update the execution instruction queue.

[0128] S5.5: Design task splitting and merging strategies, and dynamically adjust task structure based on task characteristics.

[0129] Among them, for large and complex tasks, tasks are split into subtasks that can be executed in parallel based on resource requirement similarity and execution stage division; for multiple small similar tasks, tasks are merged into batch task groups based on execution location proximity and resource complementarity.

[0130] It should be noted that large, complex tasks are those whose execution time is expected to exceed the system's preset threshold and whose resource requirements are above average. They typically consist of multiple identifiable execution phases. Small, similar tasks are a collection of tasks with shorter execution times, similar resource requirements, and execution locations within a predetermined spatial range. The size and complexity of a task are determined through a comprehensive assessment of its resource requirements, estimated execution time, and the number of operational steps involved.

[0131] S5.6: Establish a load balancing feedback regulator to regularly evaluate the load adjustment effect and automatically optimize the load balancing strategy parameters.

[0132] S6: Establish a closed-loop performance optimization framework, regularly analyze system operation data, identify performance bottleneck factors, and automatically adjust scheduling strategy parameters to achieve continuous improvement in system performance.

[0133] Furthermore, a performance indicator monitoring system is constructed, focusing on collecting key operational data such as task completion efficiency, resource utilization, and conflict handling quality; a data analysis module is designed, and pattern recognition and anomaly detection methods are applied to identify performance fluctuations and bottlenecks in system operation; a performance bottleneck attribution model is established, and performance problems are mapped to scheduling strategy parameters through causal analysis; a parameter optimization model is constructed to automatically search for the optimal configuration of scheduling strategy parameters to balance short-term performance and long-term stability; a progressive parameter adjustment mechanism is adopted to optimize the scheduling strategy while ensuring system stability, and the optimization effect is verified through experimental evaluation to form a complete closed-loop optimization system.

[0134] Furthermore, this embodiment also provides an intelligent scheduling system for mobile robots, including: a data acquisition module for acquiring scheduling scenario data, the scheduling scenario data including robot status data, working environment data and scheduling task data; a task feature determination module for extracting basic task attributes to form a static feature layer, establishing a dynamic feature layer composed of dynamic influencing factors, and generating a two-layer task feature model; an adaptive matching module for establishing an adaptive matching model between robots and tasks based on a machine learning algorithm, performing feature fusion of robot status data and task feature matrix, and calculating a matching score matrix; a task allocation and conflict avoidance module for generating a task allocation plan based on the matching score matrix, predicting potential path conflict points and determining avoidance strategies using trajectory prediction methods, and performing conflict prediction and avoidance; a load balancing module for dynamically adjusting task allocation and execution order according to the robot load status; a performance optimization module for establishing a performance closed-loop optimization framework, regularly analyzing system operation data, identifying performance bottleneck factors, and automatically adjusting scheduling strategy parameters.

[0135] In summary, the present invention constructs a comprehensive intelligent scheduling information acquisition framework, integrates the three-dimensional data of "executing subject-environmental scenario-task goal", comprehensively grasps the core elements of the scheduling problem, realizes the global perception of the scheduling scenario, and reduces the decision-making error rate caused by incomplete information; the established two-layer task feature model and adaptive matching model realize the transformation of the intelligent scheduling system from "static rule-driven" to "dynamic environment adaptation", enabling robots to flexibly adjust decision-making strategies according to environmental changes and task characteristics; through trajectory prediction and conflict avoidance mechanism, the safety and smoothness of multi-robot collaborative operations are improved, the path planning problem in high-density robot environments is solved, and a technological leap from passive conflict handling to active conflict prevention is achieved.

[0136] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An intelligent scheduling method for mobile robots, characterized in that: include: Acquiring scheduling scenario data, wherein the scheduling scenario data includes robot status data, operating environment data, and scheduling task data; Extract the basic attributes of the task to form a static feature layer, establish a dynamic feature layer composed of dynamic influencing factors, and generate a two-layer task feature model; An adaptive matching model between robots and tasks is established based on machine learning algorithms. The robot status data is fused with the task feature matrix to calculate the matching score matrix. Generate a task allocation plan based on the matching score matrix, use a trajectory prediction method to predict potential path conflict points and determine an avoidance strategy, and perform conflict prediction and avoidance; Dynamically adjust task allocation and execution order based on robot load status; Establish a performance closed-loop optimization framework, regularly analyze system operation data, identify performance bottleneck factors, and automatically adjust scheduling strategy parameters.

2. The intelligent scheduling method for mobile robots according to claim 1, characterized in that: The robot status data includes location information, power information and task execution status; the working environment data includes obstacle distribution, personnel activities and temporary events in the working area; the scheduling task data includes task identification, task content description, expected completion time and task resource requirements.

3. The intelligent scheduling method for mobile robots according to claim 1, characterized in that: The construction process of the two-layer task feature model includes: Processing the scheduling task data to generate a task description parameter set, and processing the robot state data and the working environment data to generate an environment state parameter set; Extracting basic task attributes from the task description parameter set, performing multi-dimensional feature quantization encoding on the basic task attributes, and generating a static feature layer; Analyze the dynamic influencing factors in the environmental state parameter set, calculate the timeliness evaluation index, environmental matching index and task coordination index, and construct a dynamic feature layer; A feature fusion algorithm based on inter-layer weight transfer is used to establish a mapping relationship between the static feature layer and the dynamic feature layer, generate a two-layer task feature model, and output a task feature matrix.

4. The intelligent scheduling method for mobile robots according to claim 1, characterized in that: The process of constructing the adaptive matching model includes: Based on the environmental state parameter set, the state evaluation score of the robot individual is calculated to generate the robot state feature representation; Design a feature fusion module that inputs the robot state feature representation and the task feature vector matrix, performs cross-modal feature mapping, and outputs an interactive feature representation; Design a multi-head attention module to calculate the importance weight of each dimension feature in the interactive feature representation, and generate a correlation feature matrix through weighted aggregation; Construct a deep neural network including a graph convolutional network and residual connections, take the correlation feature matrix as input, and extract high-order matching features through a multi-layer graph attention mechanism; Calculating an initial matching score for each robot-task combination based on the high-order matching features; A global optimization function is constructed, a global balance adjustment coefficient is calculated, and the initial matching score is corrected using the global balance adjustment coefficient to obtain a matching score matrix.

5. The intelligent scheduling method for mobile robots according to claim 1, characterized in that: The method of using the trajectory prediction method to predict potential path conflict points and determine the avoidance strategy includes: An integer linear programming model is constructed based on the matching score matrix, resource constraints and task priority constraints are set, and the initial task allocation plan is obtained using the branch and bound algorithm. Analyze each robot's historical motion data, extract motion pattern features, and predict the robot's future motion trajectory based on the currently assigned task target position; Build a multi-robot trajectory interaction model, taking the future motion trajectories of all robots as input, calculate the spatiotemporal similarity matrix between trajectory pairs, identify trajectory intersections and overlapping path segments, and determine potential path conflict points, their probability of occurrence, and their estimated time of occurrence; For potential path conflict points, a set of avoidance solutions is generated based on the conflict point information and the motion status of the relevant robots; The Monte Carlo tree search algorithm is used to select the global optimal avoidance strategy combination from the avoidance scheme set and generate conflict avoidance execution instructions.

6. The intelligent scheduling method for mobile robots according to claim 1, characterized in that: The dynamically adjusting task allocation and execution order according to the robot load state includes: Establish a real-time load monitoring module to collect robot operation status data and calculate comprehensive load indicators; Constructing a load threshold adaptive adjustment mechanism to calculate the upper and lower limits of the load threshold, and triggering task adjustment when the comprehensive load indicator exceeds the load threshold range; For robots in high-load conditions, a task migration strategy is used to build a critical path analysis model based on the task dependency graph, identify non-critical tasks, calculate migration candidate scores, and perform task reallocation. Constructing a dynamic task sorting algorithm, calculating task time sensitivity scores, and reordering the conflict avoidance execution order based on the task time sensitivity scores; Design task splitting and merging strategies, and dynamically adjust task structure based on task characteristics.

7. The intelligent scheduling method for mobile robots according to claim 6, characterized in that: The task migration strategy is used to construct a critical path analysis model based on the task dependency graph, which includes: Build a task dependency graph; Apply the critical path algorithm to analyze the task dependency graph and calculate the time slack of each task; Classifying tasks into critical tasks and non-critical tasks according to the time slack, and calculating a migration candidate score for each non-critical task in combination with the task resource requirement characteristics and execution complexity; Search for robots in low-load state, build a candidate set of migration targets, and calculate the receiving capability score of each candidate robot; Establish a task-robot matching evaluation matrix and calculate the transfer benefit value for each pair of candidate tasks and candidate robots; Select the task-robot pair with the highest migration benefit value and perform the task migration operation.

8. The intelligent dispatching system of mobile robots is characterized by: include: A data acquisition module is used to acquire scheduling scenario data, wherein the scheduling scenario data includes robot status data, operating environment data and scheduling task data; The task feature determination module is used to extract the basic attributes of the task to form a static feature layer, establish a dynamic feature layer composed of dynamic influencing factors, and generate a two-layer task feature model; The adaptive matching module is used to establish an adaptive matching model between the robot and the task based on the machine learning algorithm, fuse the robot state data with the task feature matrix, and calculate the matching score matrix; A task allocation and conflict avoidance module is used to generate a task allocation plan based on the matching score matrix, predict potential path conflict points using a trajectory prediction method and determine an avoidance strategy, and perform conflict prediction and avoidance; Load balancing module, used to dynamically adjust task allocation and execution order according to the robot load status; The performance optimization module is used to establish a performance closed-loop optimization framework, regularly analyze system operation data, identify performance bottleneck factors, and automatically adjust scheduling strategy parameters.

Images