Multi-scene oriented robot cooperative path intelligent scheduling system and method

Abstract

The present application relates to the technical field of intelligent robots, in particular to a multi-scene-oriented robot collaborative path intelligent scheduling system and method, the method extracts a general mixed feature vector covering space-time structure, dynamic obstacles and the like through multi-dimensional environment data collection, generates adaptive parameters from a lightweight fully connected neural network after scene recognition and feature weighting, drives the execution of underlying algorithms such as speed obstacle method and model predictive control, can adapt to different scenes such as dynamic and structured scenes, realizes safe and efficient scheduling of robot collaborative paths, and improves multi-scene adaptability and scheduling flexibility; the system comprises a multi-scene environment data collection module, a general mixed feature vector extraction module, a scene type recognition and feature weighting module, an adaptive mapping model construction and parameter generation module, and an algorithm execution module.

Description

Technical Field

[0001] This invention relates to the field of intelligent robot technology, and more specifically, to a robot collaborative path intelligent scheduling system and method for multiple scenarios. Background Technology

[0002] Robot collaborative path scheduling technology is one of the core technologies in the field of intelligent robots. It is widely used in various scenarios such as warehousing and logistics, park inspection, and industrial production. Its performance directly affects the work efficiency and operational safety of robot teams.

[0003] Currently, existing robot collaborative path scheduling methods are mostly designed for single, specific scenarios, lacking universal adaptability. In dynamic scenarios, traditional methods struggle to accurately capture dynamic changes in obstacles, easily leading to collision risks; in structured scenarios, they cannot fully adapt to fixed topologies to optimize path efficiency. Furthermore, existing technologies lack a dynamic mapping relationship between features and algorithm parameters; parameters are mostly fixed settings and cannot be flexibly adjusted according to the core needs of the scenario. This results in a significant drop in scheduling performance when switching between complex multi-scenario environments, making it difficult to balance safety, timeliness, and collaboration.

[0004] To address this, the present invention proposes an intelligent scheduling system and method for robot collaborative paths in multiple scenarios. By using feature quantization fusion, scenario adaptive weighting, and dynamic parameter generation, it solves the problems of scenario adaptability limitations and insufficient scheduling performance in existing technologies. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a robot collaborative path intelligent scheduling system and method for multiple scenarios.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligent scheduling of robot collaborative paths in multiple scenarios includes the following steps: Step 1: Multi-scene environment data acquisition: Through static perception unit, dynamic perception unit and task and robot state unit, static topology information, dynamic obstacle information, task constraint information and robot state data of the scene are collected respectively; Step 2: Extraction of general hybrid feature vectors: Perform feature quantization and fusion processing on the collected multi-dimensional raw data to extract general hybrid feature vectors covering spatiotemporal structure, dynamic obstacles, task load and robot team features, and perform normalization processing. Step 3: Scene type identification and feature weighting: The current scene type is identified using a preset scene classifier, and the feature weight vector is determined based on the core scheduling requirements of different scenes. The general mixed feature vector is then weighted. Step 4: Construction and parameter generation of the adaptive mapping model: Construct a lightweight fully connected neural network as the adaptive mapping model. After offline training, input the weighted feature vector into the model to generate an algorithm parameter set adapted to the current scene. Step 5: Algorithm Execution: The generated parameter set drives the execution of the underlying scheduling algorithm to achieve intelligent scheduling of robot collaborative paths.

[0007] The present invention also provides a robot collaborative path intelligent scheduling system for multiple scenarios, including a multi-scenario environment data acquisition module, a general hybrid feature vector extraction module, a scenario type identification and feature weighting module, an adaptive mapping model construction and parameter generation module, and an algorithm execution module; The multi-scenario environment data acquisition module is used to collect multi-dimensional raw data; the general hybrid feature vector extraction module is used to generate normalized general hybrid feature vectors; the scene type identification and feature weighting module is used to identify scene types and perform weighted processing on feature vectors; the adaptive mapping model construction and parameter generation module is used to generate an algorithm parameter set adapted to the scene; and the algorithm execution module is used to drive the execution of the underlying scheduling algorithm and optimize scheduling performance.

[0008] Furthermore, the general hybrid feature vector includes spatiotemporal structure feature components, dynamic obstacle feature components, task load feature components, and robot team feature components. Each feature component respectively represents the matching relationship between the static topology of the scene and the time constraints of the task, the interference risk of obstacles to the robot's cooperative motion, the matching degree between the urgency of the task and the robot team's load, and the overall motion and load coordination capability of the robot team.

[0009] Furthermore, the feature weight vector is determined based on the core scheduling requirements of different scenarios, as detailed below: To address the core scheduling requirements of different scenarios, differentiated weights are assigned to each feature component in the general hybrid feature vector. Specifically, in dynamic scenarios, the weight of the dynamic obstacle feature component is increased; in structured scenarios, the weight of the spatiotemporal structure feature component is increased; and in other scenarios, the weight of each feature component is adjusted according to its core requirements. The weighted feature vector can accurately match the scheduling priority of the current scenario.

[0010] Furthermore, the output parameters of the adaptive mapping model include the safety radius, prediction time domain, and path update frequency. These parameters are converted into practically usable algorithm parameters after inverse normalization.

[0011] Furthermore, the underlying scheduling algorithm includes a speed obstacle method, a model predictive control, and a path planning algorithm, which respectively achieve obstacle avoidance, trajectory optimization, and real-time path adjustment based on the generated corresponding parameters.

[0012] Furthermore, the speed obstacle avoidance method uses a safety radius to achieve obstacle avoidance, as detailed below: The safe radius output by the adaptive mapping model is used as a constraint parameter of the velocity obstacle method. During the cooperative movement of the robot, a safe distance threshold between the robot and the obstacle is constructed based on this safe radius. When the real-time distance between the robot and the obstacle is less than this threshold, the velocity obstacle method automatically adjusts the robot's movement speed and direction, so that the robot can avoid the obstacle while maintaining cooperation and avoiding the risk of collision.

[0013] Furthermore, the aforementioned model predictive control uses the prediction time domain to achieve trajectory optimization, as detailed below: The predicted time domain output by the adaptive mapping model is used as the time window parameter of the model predictive control. Based on this predicted time domain, the model predictive control predicts the motion trajectory and task execution status of the robot team in the future period. Combined with task constraints and robot state data, the trajectory is optimized and adjusted so that the optimized trajectory meets the requirements of task timeliness and motion stability.

[0014] Furthermore, the aforementioned path planning utilizes path update frequency to achieve real-time path adjustment, as detailed below: The path update frequency output by the adaptive mapping model is used as a dynamic adjustment parameter for path planning. The path planning module periodically collects environmental data and task status of the current scene according to this frequency, compares the deviation between the current path and the actual requirements, and when the deviation exceeds the preset range, the path is replanned based on the real-time feature data of the general hybrid feature vector, and the new path is synchronized to the robot team to realize the dynamic adaptation and real-time adjustment of the path.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. Employing multi-dimensional data acquisition and general hybrid feature vector extraction technology, it comprehensively covers spatiotemporal structure, dynamic obstacles, task load, and robot team features, combined with a scene classifier to achieve accurate scene recognition. Through a differentiated feature weighting strategy, dynamic scenes strengthen the weight of obstacle interference features, and structured scenes increase the proportion of topology adaptation features, enabling feature vectors to accurately match the core needs of the scene. This solves the problem of single scene adaptation in traditional methods and significantly improves the scheduling targeting in multiple scenarios. 2. A lightweight fully connected neural network is constructed as an adaptive mapping model to establish a dynamic mapping relationship between features and algorithm parameters. This automatically generates the safety radius, prediction time domain, and path update frequency, driving the underlying scheduling algorithm to execute collaboratively. The velocity obstacle method avoids collisions based on the safety radius, model predictive control optimizes the trajectory through prediction time domain, and path planning is dynamically adjusted according to the update frequency. This solves the problem of poor flexibility in traditional fixed-parameter scheduling, achieving a balance between safety, timeliness, and collaboration. Attached Figure Description

[0016] Figure 1 A flowchart for a multi-scenario intelligent scheduling method for robot collaborative paths; Figure 2 This is a flowchart of the algorithm execution in step five of this invention; Figure 3 This is a structural block diagram of the intelligent scheduling system for robot collaborative paths in multiple scenarios according to the present invention. Detailed Implementation

[0017] Example 1, refer to Figure 1 The intelligent scheduling method for robot collaborative paths in this embodiment, which is designed for multiple scenarios, specifically includes the following steps: Step 1: Data collection in multiple scenarios.

[0018] To ensure the comprehensiveness and accuracy of subsequent feature extraction, a standardized environmental data acquisition system must first be established, clarifying the selection and operating parameters of each sensing unit, thus providing raw data support for generating high-quality, general-purpose hybrid feature vectors. Specific configurations include: Static perception unit: It uses LiDAR and visual camera to collect static topological information of the scene (such as path distribution and region boundaries) at a sampling frequency of 10Hz; Dynamic sensing unit: Employs millimeter-wave radar to collect position and speed information of dynamic obstacles (such as people and mobile devices), with a sampling frequency of 20Hz; Task and Robot Status Unit: Acquires task time constraints, workload, and robot speed, load capacity, and other status data via a communication bus (such as CAN bus). The data update cycle is 500ms.

[0019] Step 2: Extraction of general hybrid feature vectors.

[0020] Based on the collected multi-dimensional raw data, targeted feature quantization and fusion processing are used to extract a general hybrid feature vector that can uniformly represent the characteristics of different scenes, laying the foundation for subsequent scene recognition and parameter adaptation. Specifically, the general hybrid feature vector is calculated... (There are 6 components in total), and the calculation method is as follows: S21. Spatiotemporal structure feature components: used to characterize the matching relationship between the static topological complexity of the scene and the time constraints of the task, quantified by selecting two core indicators: Path effectiveness percentage: ,in, Obtained by fitting the passable area from the lidar point cloud. It is the sum of the total lengths of all passable paths in the scene, obtained by accumulating the lengths of all passable path segments fitted by the LiDAR point cloud; Time-topology constraint factor: ,in, The maximum allowable delay time for the task (in seconds). (E is the number of edges in the scene topology graph, and V is the number of nodes), and the value range is limited to [0,1] by normalization.

[0021] S22. Dynamic obstacle feature components: Focusing on the interference risk of obstacles to the robot's cooperative motion, a comprehensive risk coefficient is calculated by integrating multi-dimensional obstacle information. The formula is as follows: ; in, The number of dynamic obstacles (statistics based on target detection results from millimeter-wave radar); Let be the speed of the i-th obstacle (unit: m / s); The obstacle volume coefficient is obtained by normalizing the area of ​​the target's bounding rectangle, and its range is [0.1, 1]. The minimum Euclidean distance (in meters) between the obstacle and the robot team, when Use 0.5 to avoid the denominator being 0.

[0022] S23. Task Load Characteristic Components: Reflect the matching degree between the urgency of the task and the workload of the robot team, quantified through two key indicators: Task urgency: , Remaining time for the task Specify a completion time for the task, with a value range of [0,1]. Load density: ,in, Total workload of the task (unit: kg∙m). The maximum workload of a single robot (based on the preset value set by the robot at the factory). The total number of robots participating in this collaborative path scheduling is preset with parameters that can be configured according to task requirements; S24. Robot Team Feature Components: Characterizing the team's overall motion and load coordination capabilities, calculated through a normalized average performance score, using the following formula: ; in, Let be the maximum speed of the j-th robot (in m / s). Let be the load capacity (in kg) of the j-th robot. , The maximum speed and maximum load capacity of the robot supported by the system (based on the preset values ​​set by the robot at the factory). To eliminate the dimensional differences among the feature vectors and ensure the effectiveness of subsequent model processing, a min-max normalization operation is performed on the above six components to obtain a normalized general mixed feature vector. Each component takes the value [0,1].

[0023] Step 3: Scene type identification and feature weighting.

[0024] To ensure that feature vectors more accurately match the core requirements of a scene, the current scene type must first be identified using a pre-defined scene classifier. Then, feature weight vectors are determined based on the scheduling priorities of different scenes. This weighting strengthens the influence of key features on subsequent parameter generation, thereby improving scene adaptation accuracy. The scene classifier is implemented using a Support Vector Machine (SVM), and its training process, classification logic, and weight vector determination method are disclosed below: Classifier training data preparation: The training sample set covers three typical scenarios: dynamic scenarios, structured scenarios, and transitional scenarios. Dynamic scenarios include hospitals, industrial parks, and shopping malls; structured scenarios include warehouses, workshops, and logistics sorting centers; and transitional scenarios are the areas connecting two types of scenarios. A total of 1500 samples are included, each containing a general hybrid feature vector. And corresponding scene tags. Among them: 600 sets of dynamic scene samples (200 sets of hospital scenes, 200 sets of park scenes, 200 sets of shopping mall scenes), 600 sets of structured scene samples (200 sets of warehouse scenes, 200 sets of workshop scenes, 200 sets of logistics sorting center scenes), and 300 sets of transition scene samples (such as the area connecting the workshop and the park). The sample data was obtained through a combination of on-site collection and simulation generation to ensure the diversity and representativeness of the samples. Classifier training process: Kernel function parameters of SVM classifier The regularization parameter C' is determined using a grid search-cross-validation method. The specific process is as follows: 1) Setting the parameter search range: The search range is [0.1, 2.0], with a step size of 0.1; C''s search range is [0.1, 5.0], with a step size of 0.1; 2) Grid search and cross-validation: Five-fold cross-validation is used to calculate the classification accuracy of the validation set for each parameter combination; 3) Determining the optimal parameters: when When C' = 0.8 and C' = 1.2, the classification accuracy on the validation set reaches its maximum, which is the highest value among all parameter combinations. Therefore, this set of parameters is determined to be the optimal training parameters. During the training process, the sample set is divided into a training set and a validation set in a 4:1 ratio. Iterative training is performed until the difference in accuracy between two adjacent iterations is less than 0.001, at which point training is stopped and the classifier parameters are saved. Classification execution logic: The normalized general mixed feature vector... Input the trained SVM classifier, and the classifier calculation... The distance to the decision boundary of each scene is used to output the scene type with the largest distance as the recognition result; if the distance difference of each scene is less than 0.02, it is judged as a transition scene, and the average of the weight vectors of dynamic scene and structured scene is used as the transition weight vector. After classification, a weight vector is determined based on the core requirements of scene scheduling, and a general hybrid feature vector is applied. The weighted processing is performed, and the specific weight settings and basis are as follows: Dynamic Scene Weighting Strategy: The core scheduling requirement of dynamic scenes is to avoid interference from dynamic obstacles and ensure timely task completion. The weight vector is determined through a requirement priority quantification-sensitivity analysis verification method. The specific process includes: 1) Requirement Priority Quantification: The weights of core scene requirements are determined using the Analytic Hierarchy Process (AHP): dynamic obstacle avoidance requirement weight 0.45, task timeliness requirement weight 0.35, basic topology adaptation requirement weight 0.15, and team collaboration requirement weight 0.05; 2) Feature Association Mapping: The requirement weights are mapped to corresponding feature components, such as dynamic obstacle features... The requirement for obstacle avoidance is assigned a weight of 0.4; the task urgency feature is also considered. The timeliness requirement of related tasks is weighted at 0.2; spatiotemporal structural features ( , The associated basic topology adaptation requirements are each weighted at 0.15; task load density. Characteristics of robot teams Each was assigned a weight of 0.05; 3) Sensitivity verification: Through multiple sets of simulation experiments, it was verified that when the weight vector was... At that time, the obstacle collision rate of the scheduling system was less than 0.03 times / s and the on-time task completion rate was higher than 98%, verifying the effectiveness of the weight vector. The weighted feature vector is obtained by element-wise multiplication, i.e. ,in This represents the element-wise multiplication operation; Structured Scenario Weighting Strategy: The core requirement of structured scenarios is to adapt to fixed topologies and improve path planning efficiency. The weight vector is determined through a correlation analysis method between topology adaptability and planning efficiency. Specific process: 1) Definition of correlation indicators: Topology adaptability (coverage of passable paths) and planning efficiency (path length reduction rate) are defined as core evaluation indicators; 2) Measurement of feature influence: Through controlled variable experiments, the influence coefficients of each feature component on the evaluation indicators are statistically analyzed. The influence coefficient of spatiotemporal structure features on topology adaptability is 0.6, while the influence coefficients of dynamic obstacle features, task load features, and robot team features are 0.1, 0.2, and 0.1, respectively; 3) Weight determination and verification: Weights are assigned based on the influence coefficients to obtain the weight vector. Simulation results show that path planning efficiency is improved by more than 35% under this weighting, and topology fit reaches 96%, meeting the requirements. Weighted feature vector ; Transition Scenario Weighting Strategy: Transition scenarios possess characteristics of both types of scenarios, and the weight vector... Weighted feature vectors ; The core purpose of weighted processing is to enable subsequent adaptive mapping models to focus on the key influencing factors of the current scenario and generate an algorithm parameter set that better fits the needs of the scenario by allocating feature weights.

[0025] Step 4: Construction and parameter generation of the adaptive mapping model.

[0026] The adaptive mapping model establishes a mapping relationship between weighted feature vectors and the parameter set of the underlying algorithm by constructing a lightweight fully connected neural network, thereby achieving automatic parameter optimization and generation. Its structure and training process are as follows: Model Structure: Balancing inference efficiency and mapping accuracy, a lightweight fully connected neural network with two hidden layers is employed. The specific structure is as follows: Input layer: 6 neurons (corresponding to the weighted feature vector) (dimensions) Hidden layer 1: 12 neurons, activation function is ReLU; Hidden layer 2: 8 neurons, activation function is ReLU; Output layer: 3 neurons, activation function is sigmoid, output is the normalized parameter set. .

[0027] Model training: Offline training allows the model to learn the mapping relationship between features and optimal parameters across multiple scenarios, clarifying the training data, loss function, and optimization strategy. Training dataset: Contains 1000 sets of weighted feature vectors from multiple scenes. Samples, and the corresponding optimal parameter set (historical data, parameters obtained by manual debugging that meet scheduling performance); Loss function (objective minimization): The weighting coefficients of 0.6 and 0.4 were determined through performance index balance analysis: the impact weights of task completion time deviation and conflict rate were statistically analyzed in multiple scenarios. When the task timeliness weight was 0.6 and the security performance weight was 0.4, the system's overall performance score was ( The highest value is determined by this weight combination; where... The actual time to complete the task. The task requires a completion time. Number of robot collisions; Training optimization parameters: The Adam optimizer learning rate of 0.001 and the number of training epochs of 200 were determined through a convergence speed-generalization ability balance analysis: The learning rate was tested stepwise from 0.0001 to 0.01 using a trial-and-error method. When the learning rate was 0.001, the model converged the fastest and there was no overfitting. The number of training epochs was determined by monitoring the change in the loss function on the validation set. When the number of training epochs reached 200, the change in the loss function was less than 0.0001, and it tended to converge. Therefore, the number of training epochs was determined to be 200. Parameter generation process: The weighted feature vector is input into the trained model, and normalized parameters are obtained through forward propagation. Then, inverse normalization is performed to convert them into usable algorithm parameters. Safety radius Among them, the safety radius boundary value , Through multi-scenario collision risk verification, it was determined that: the minimum safe distance requirement for robots in different scenarios is 0.3m, which is the minimum safety redundancy when there are no dynamic obstacles, and 1.5m is the maximum safety requirement in dynamic scenarios with dense crowds. This range can cover the safety protection requirements of 99% of scenarios. Prediction Time Domain Among them, predicting the time domain boundary values , Through experiments balancing planning efficiency and real-time performance, it was determined that: when At the same time, it can meet the needs of rapid response in structured scenarios; At this time, it can cover the prediction needs of obstacle motion trajectory in dynamic scenes, and the overall performance of model predictive control is optimal within this range; Path update frequency Among them, the path update frequency boundary value , System resource utilization tests determined that 1Hz is the minimum path update frequency for structured scenarios (resource utilization <10%), and 5Hz is the maximum update frequency for dynamic scenarios (resource utilization <50%), ensuring real-time path updates while avoiding system overload.

[0028] Step 5: Algorithm execution.

[0029] The underlying scheduling algorithm is driven by an adaptively generated parameter set. For example... Figure 2 As shown, specifically, the parameter set Input underlying algorithms (velocity barrier method and model predictive control): S51, Speed ​​Obstacle Method As a safety radius, collisions between the robot and obstacles are avoided; specifically as follows: S511, Obstacle Area Modeling: Based on the real-time position and velocity data of dynamic obstacles, combined with... Construct a safe obstacle zone for each obstacle, that is, a circle centered on the center of the obstacle, (obstacle volume coefficient × ... A circular region with radius (+equivalent radius of the obstacle) where the obstacle volume coefficient is... The classification is based on the actual volume level and motion characteristics of obstacles: the equivalent volume is calculated using the circumscribed rectangle dimensions of obstacles detected by millimeter-wave radar, and small obstacles (such as pedestrians) are classified as such. =1.0), medium-sized (such as small delivery vehicles, =1.2), large (such as freight forklifts, =1.5) Three levels; at the same time, combined with obstacle movement speed correction, the coefficient is increased by 0.1 when the speed is ≥1m / s and decreased by 0.1 when the speed is ≤0.2m / s, to ensure that the safe zone is adapted to the actual threat level of the obstacle; the equivalent radius of the obstacle is obtained by fitting the outer rectangle of the target detected by millimeter wave radar; S512. Velocity Space Partitioning: Based on the robot's current motion state, a comprehensive and controllable set of candidate velocities is generated across multiple dimensions, and the predicted trajectory corresponding to each velocity is accurately calculated. Specific execution details are as follows: (1) Velocity dimension division: Combining robot dynamics constraints (maximum speed, maximum acceleration), linear discretization is performed in the interval [0, robot maximum speed], with the velocity step size set to 0.1m / s (to ensure fine velocity coverage and adapt to the velocity adjustment requirements in different scenarios). (2) Angle dimension division: Based on the current direction of the robot, the direction angle is divided in 5° steps within the range of [-45°, +45°]. This angle range covers the robot's regular turning requirements and avoids the trajectory from deviating from the task target due to excessively large direction angles, forming a two-dimensional candidate velocity matrix of velocity-angle, i.e., the candidate velocity set; (3) Trajectory calculation method: Based on the robot kinematics model, this embodiment adopts a differential driving model: , , ,in The angular velocity is derived from the rate of change of velocity and direction angle. The Euler integral method is used to calculate the discrete trajectory points of each candidate velocity in the next 0.5s (with an integration step size of 0.01s to ensure the accuracy of trajectory calculation). Finally, the complete motion trajectory corresponding to each candidate velocity is output. S513, Collision Risk Assessment: Traverse candidate velocity trajectories and determine whether the trajectory will intersect with any safe obstacle area within the next 0.5 seconds (i.e., within the time domain window to ensure rapid response). If they intersect: mark the speed as a dangerous speed and remove it from the candidate set; if they do not intersect: keep the speed as a safe speed. S514, Safe speed output: Select the speed with the smallest angle to the current task target direction from the remaining safe speeds and pass it as the output of the speed obstacle method to the model prediction control module; S52, Model Predictive Control Usage As part of the prediction time domain, the robot's motion trajectory is optimized; specifically as follows: S521. Trajectory Prediction Input Construction: Integrate the following data to construct the prediction input vector: Robot's current state: position ,speed Load status; Environmental constraints: the safe speed range output by the speed barrier method, and the traversable area boundary of the static topology; Task objectives: target location, remaining task time (calculated from task status data); S522, Multi-target trajectory optimization: In Within the time window, with the dual objectives of minimizing trajectory time and maximizing the distance between the trajectory and obstacles, an optimization objective function is constructed (the objective is to minimize this function): ; in, The total time taken to predict the trajectory, Remaining time for the task The minimum distance between the predicted trajectory and all obstacles; 0.6 and 0.4 are weighting coefficients determined through multiple trials and optimizations; constraints: the trajectory does not exceed the passable area and the speed does not exceed the safe speed range. S523, Optimal Trajectory Solution: The objective function is solved using a quadratic programming algorithm to obtain... Discretized trajectory points within a time window (one trajectory point every 0.02s), outputting the trajectory's position, velocity, and acceleration commands; S524, Trajectory Tracking Control: Converts discretized trajectory points into motion control commands (linear velocity, angular velocity) for the robot, and adjusts the robot's actuators through a PID controller; S53, Path Planning Module Usage The path is adjusted in real time according to the update frequency; details are as follows: S531, Path Validity Detection: (According to...) The frequency of this validation is used to verify the validity of the current global path. Static verification: Check whether the path exceeds the passable area updated by the LiDAR (e.g., the path is occupied by temporary obstacles). Dynamic verification: Predicting future obstacle movement trends based on millimeter-wave radar. Will obstacles occupy critical nodes of the path within a given time? S532, Path Update Decision: Regular update: When it reaches During the update cycle, regardless of whether the path is valid, a path recalculation is triggered to ensure that the path is synchronized with environmental changes; Urgent Update: If path validity detection indicates that an obstacle is about to occupy the path (distance ≤ If the path (×1.2) is already in use, an emergency update will be triggered immediately, interrupting the regular cycle; S533, New Path Generation: A new global path is generated using the A* algorithm combined with spatiotemporal structure features from a general mixed feature vector. Path cost function: taking into account path length ( Related), obstacle avoidance difficulty ( Related), task urgency ( (Related) Path smoothing: The generated path is fitted with a B-spline curve to reduce the number of turns during robot movement and improve motion stability; S534, Path handover and transition: The newly generated global path is passed to the model prediction and control module as a global constraint for the next round of trajectory optimization; at the same time, a path transition segment (length = robot's current speed × 0.1s) is generated to ensure that the robot smoothly switches from the original path to the new path and avoids sudden acceleration and deceleration.

[0030] Example 2, refer to Figure 3 The robot collaborative path intelligent scheduling system for multiple scenarios in this embodiment includes a multi-scenario environment data acquisition module, a general hybrid feature vector extraction module, a scenario type identification and feature weighting module, an adaptive mapping model construction and parameter generation module, and an algorithm execution module. Multi-scene environment data acquisition module: A standardized perception system is built. Through static perception unit, dynamic perception unit and task and robot state unit, multi-dimensional raw data such as static topology of scene, dynamic obstacles, task constraints and robot state are collected respectively, providing high-quality data support for subsequent feature extraction. The general hybrid feature vector extraction module quantifies and fuses the collected multi-dimensional raw data, calculates the spatiotemporal structure, dynamic obstacles, task load and core feature components of the robot team, and generates a general hybrid feature vector that uniformly represents the characteristics of different scenarios after normalization, laying the foundation for scene recognition and parameter adaptation. Scene type identification and feature weighting module: The trained SVM classifier is used to identify the current scene type (dynamic, structured, transitional scene). Based on the core scheduling requirements of different scenes, the feature weight vector is determined. The general mixed feature vector is weighted by element-wise multiplication to strengthen the influence of key features on the generation of subsequent parameters. The module for constructing and generating parameters of the adaptive mapping model: Constructs a lightweight fully connected neural network, learns the mapping rules between features and optimal parameters through offline training, inputs the weighted feature vector into the model, and generates practically usable algorithm parameters such as safe radius, prediction time domain, and path update frequency that are adapted to the current scene through forward propagation and inverse normalization. Algorithm execution module: Drives the execution of underlying scheduling algorithms such as speed barrier method and model predictive control with adaptively generated parameter sets.

[0031] Through the detailed description of the above embodiments, the intelligent scheduling system and method for robot collaborative paths in multiple scenarios of the present invention lays the foundation through standardized multi-dimensional data collection, extracts key features through general hybrid feature vectors and scene-adaptive weighted enhancement, and then dynamically generates adaptation parameters by an adaptive mapping model, ultimately driving the collaborative execution of the underlying algorithm. This solution breaks through the limitations of traditional single-scenario scheduling, achieving comprehensive feature extraction, accurate scene recognition, adaptive parameter generation, and efficient scheduling execution. It can flexibly adapt to various scenarios such as dynamic and structured scenarios, effectively improving the safety, timeliness, and collaboration of robot collaborative path scheduling, and providing reliable technical support for multi-scenario robot operations.

[0032] The above formulas are all dimensionless calculations, and the preset parameters in the formulas should be set by those skilled in the art according to the actual situation.

[0033] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0034] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0035] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0036] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0037] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0038] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0039] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A robot collaborative path intelligent scheduling method for multiple scenarios, characterized in that, The method flow is as follows: Step 1: Multi-scene environment data acquisition: Through static perception unit, dynamic perception unit and task and robot state unit, static topology information, dynamic obstacle information, task constraint information and robot state data of the scene are collected respectively; Step 2: Extraction of general hybrid feature vectors: Perform feature quantization and fusion processing on the collected multi-dimensional raw data to extract general hybrid feature vectors covering spatiotemporal structure, dynamic obstacles, task load and robot team features, and perform normalization processing. Step 3: Scene Type Identification and Feature Weighting: A pre-defined scene classifier is used to identify the current scene type. Based on the core scheduling requirements of different scenes, a feature weight vector is determined, and the general hybrid feature vector is weighted. Specifically, for the core scheduling requirements of different scenes, differentiated weights are assigned to each feature component in the general hybrid feature vector. In dynamic scenes, the weight of the dynamic obstacle feature component is increased; in structured scenes, the weight of the spatiotemporal structure feature component is increased; and for other scenes, the weight of each feature component is adjusted according to its core requirements. The weighted feature vector can accurately match the scheduling priority of the current scene. Step 4: Construction and parameter generation of the adaptive mapping model: A lightweight fully connected neural network is constructed as the adaptive mapping model. After offline training, the weighted feature vector is input into the model to generate an algorithm parameter set adapted to the current scene. The output parameters of the adaptive mapping model include the safety radius, prediction time domain, and path update frequency. The parameters are converted into practically usable algorithm parameters after inverse normalization. Step 5: Algorithm Execution: The generated parameter set drives the execution of the underlying scheduling algorithm to achieve intelligent scheduling of robot collaborative paths.

2. The intelligent scheduling method for robot cooperative paths in multiple scenarios according to claim 1, characterized in that, The general hybrid feature vector includes spatiotemporal structure feature components, dynamic obstacle feature components, task load feature components, and robot team feature components. Each feature component represents the matching relationship between the static topology of the scene and the time constraints of the task, the risk of interference from obstacles to the robot's cooperative motion, the matching degree between the urgency of the task and the robot team's load, and the overall motion and load coordination capability of the robot team.

3. The intelligent scheduling method for robot cooperative paths in multiple scenarios according to claim 1, characterized in that, The underlying scheduling algorithms include the speed obstacle method, model predictive control, and path planning algorithms, which respectively achieve obstacle avoidance, trajectory optimization, and real-time path adjustment based on the generated corresponding parameters.

4. The intelligent scheduling method for robot cooperative paths in multiple scenarios according to claim 3, characterized in that, The speed obstacle avoidance method described above uses a safety radius to achieve obstacle avoidance, as detailed below: The safe radius output by the adaptive mapping model is used as a constraint parameter of the velocity obstacle method. During the cooperative movement of the robot, a safe distance threshold between the robot and the obstacle is constructed based on this safe radius. When the real-time distance between the robot and the obstacle is less than this threshold, the velocity obstacle method automatically adjusts the robot's movement speed and direction, so that the robot can avoid the obstacle while maintaining cooperation and avoiding the risk of collision.

5. The intelligent scheduling method for robot cooperative paths in multiple scenarios according to claim 3, characterized in that, The aforementioned model predictive control uses the prediction time domain to achieve trajectory optimization, as detailed below: The predicted time domain output by the adaptive mapping model is used as the time window parameter of the model predictive control. Based on this predicted time domain, the model predictive control predicts the motion trajectory and task execution status of the robot team in the future period. Combined with task constraints and robot state data, the trajectory is optimized and adjusted so that the optimized trajectory meets the requirements of task timeliness and motion stability.

6. The intelligent scheduling method for robot cooperative paths in multiple scenarios according to claim 3, characterized in that, The route planning described uses a route update frequency to achieve real-time route adjustment, as detailed below: The path update frequency output by the adaptive mapping model is used as a dynamic adjustment parameter for path planning. The path planning module periodically collects environmental data and task status of the current scene according to this frequency, compares the deviation between the current path and the actual requirements, and when the deviation exceeds the preset range, the path is replanned based on the real-time feature data of the general hybrid feature vector, and the new path is synchronized to the robot team to realize the dynamic adaptation and real-time adjustment of the path.

7. A multi-scenario intelligent scheduling system for robot cooperative paths, used to implement the multi-scenario intelligent scheduling method for robot cooperative paths as described in any one of claims 1-6, characterized in that, The system includes a multi-scene environment data acquisition module, a general hybrid feature vector extraction module, a scene type identification and feature weighting module, an adaptive mapping model construction and parameter generation module, and an algorithm execution module. The multi-scenario environment data acquisition module is used to collect multi-dimensional raw data; the general hybrid feature vector extraction module is used to generate normalized general hybrid feature vectors; the scene type identification and feature weighting module is used to identify scene types and perform weighted processing on feature vectors; the adaptive mapping model construction and parameter generation module is used to generate an algorithm parameter set adapted to the scene; and the algorithm execution module is used to drive the execution of the underlying scheduling algorithm and optimize scheduling performance.

Images