A robot task planning method and system
By performing semantic parsing and spatial normalization on the tasks, combined with the improved DBSCAN algorithm, regional task clusters are obtained and execution sequences are optimized, solving the problem of poor task homogeneity in existing technologies and improving the efficiency and coherence of robot task planning.
Patent Information
- Application Number
- CN202611184587.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-06
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-08-06
AI Technical Summary
In multi-task planning scenarios, existing technologies fail to effectively integrate the multi-dimensional semantic attributes of tasks through spatial distance-based clustering algorithms, resulting in poor task homogeneity within the region and increased ineffective switching losses.
By performing basic semantic parsing and spatial semantic normalization on the tasks, spatial similarity and semantic similarity are calculated, and a comprehensive similarity is obtained by fusion. An improved DBSCAN density clustering algorithm is used to obtain regional task clusters, which are then sorted by importance score and relevance score to optimize the task execution sequence.
It achieves spatial proximity aggregation while ensuring task homogeneity, reduces cross-regional empty driving costs, and improves robot operation efficiency and continuity.
Smart Images

Figure CN122694147B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of robotics, and more specifically, to a robot task planning method and system. Background Technology
[0002] With the rapid development of intelligent manufacturing and smart logistics industries, industrial robots and mobile robots have been widely used in various operational scenarios such as workshop assembly, warehouse sorting, and factory inspection. In actual production scheduling, robots usually need to handle multiple tasks issued in batches simultaneously, and these tasks are scattered across different physical work areas. The quality of task planning methods directly determines the robot's operating efficiency, energy consumption level, and task response time, making it one of the core technologies of intelligent robot scheduling systems.
[0003] In existing technologies, for multi-task planning scenarios, clustering algorithms based on spatial distance are often used to divide tasks into regions, and then the execution order is determined by combining task priorities. For example, some solutions use density clustering algorithms to group regions based on the physical location coordinates of tasks to reduce the distance the robot travels across regions; other solutions use task priority ranking algorithms to prioritize the execution of tasks with high urgency to ensure the timely delivery of critical tasks.
[0004] However, existing technologies use spatial distance-based clustering algorithms to divide tasks into regions. These often rely on single spatial dimension features and do not integrate multi-dimensional semantic attributes such as task type and execution tool. They cannot guarantee the homogeneity of tasks within a region while spatially aggregating them, and are prone to generating a large amount of ineffective switching losses. Summary of the Invention
[0005] To reduce switching losses between different areas and improve the efficiency of robot task planning and scheduling, this invention provides a robot task planning method and system, the specific technical solution of which is as follows: A robot task planning method includes the following steps: Basic semantic parsing and spatial semantic normalization are performed on multiple tasks. The spatial similarity and semantic similarity between each pair of tasks after basic semantic parsing and spatial semantic normalization are calculated. Spatial similarity represents the proximity of physical locations, and semantic similarity represents the semantic matching degree. Spatial similarity and semantic similarity are fused to obtain a comprehensive similarity. Based on the comprehensive similarity, clustering is performed to obtain multiple regional task clusters. For each regional task cluster, the importance of regional tasks is quantified to obtain an importance score. The region task cluster with the highest importance score is taken as the first execution unit. Using the first execution unit as the sorting starting point, all remaining unsorted region task clusters are traversed. The correlation score between each region task cluster and the last task cluster in the current sorted sequence is obtained. The region task clusters are iteratively sorted according to the correlation score to obtain the region task execution sequence.
[0006] The robot task planning method described above obtains a comprehensive similarity by fusing spatial and semantic similarity. Based on this comprehensive similarity, it performs clustering to obtain multiple regional task clusters. For each regional task cluster, it quantifies the importance of regional tasks and obtains an importance score. Finally, it obtains the regional task execution sequence based on the importance score and the correlation score. This method achieves a dual-modal clustering approach that integrates spatial and semantic similarity. It can achieve spatial proximity aggregation of regional tasks while ensuring the homogeneity of tasks within the region, reducing cross-regional empty driving costs and minimizing job switching losses within the region. In addition, iteratively sorting the regional task cluster with the highest importance score as the starting anchor point and using the correlation score between regional task clusters as the sorting criterion helps to maximize the execution continuity between regions while ensuring the priority execution of highly important tasks, thereby improving the overall operational efficiency of the robot in multi-region, multi-task scenarios.
[0007] Preferably, clustering based on comprehensive similarity to obtain multiple regional task clusters specifically involves: Based on comprehensive similarity, an improved DBSCAN density clustering algorithm is used for clustering to obtain multiple regional task clusters.
[0008] Preferably, the distance threshold of the improved DBSCAN density clustering algorithm is decoupled into a spatial granularity threshold component and a semantic granularity threshold component, wherein the spatial granularity threshold component and the semantic granularity threshold component correspond to the clustering granularity benchmarks of spatial distance and semantic distance, respectively.
[0009] Preferably, the robot task planning method further includes the following steps: Obtain real-time network heat data, and based on the real-time network heat data, obtain a spatial heat index to reflect the density of task requests and the intensity of scheduling, and a semantic heat index to reflect the fluctuation and urgency of various task demands. The spatial granularity threshold component is modulated according to the spatial heat index, so that the modulated spatial granularity threshold component decreases as the spatial heat index increases. At the same time, the semantic granularity threshold component is modulated according to the semantic heat index, so that the modulated semantic granularity threshold component decreases as the semantic heat index increases. The modulated spatial granularity threshold components and semantic granularity threshold components are weighted and fused to obtain the comprehensive threshold distance; Among them, clustering is performed based on comprehensive threshold distance to obtain multiple regional task clusters.
[0010] Preferably, clustering based on a comprehensive threshold distance to obtain multiple regional task clusters specifically includes the following steps: Based on comprehensive similarity and comprehensive threshold distance, all regional task clusters that meet the comprehensive threshold distance are retrieved to obtain the initial neighborhood set; After performing neighborhood pruning on the initial neighborhood set, weighted neighborhoods are calculated and core points are determined. Then, cluster expansion and clustering are performed until all tasks are traversed, resulting in multiple regional task clusters.
[0011] Preferably, decoupling the distance threshold of the improved DBSCAN density clustering algorithm into a spatial granularity threshold component and a semantic granularity threshold component specifically includes the following steps: Obtain the robot's historical task data, where each historical task data includes scene feature attributes, clustering parameter attributes, and performance label attributes; Multiple efficiency tags of efficiency tag attributes are weighted and fused to obtain a normalized comprehensive operation efficiency index. Historical operation task data are divided according to scene feature attributes to obtain multiple homogeneous scene subsets with consistent scene features. For each homogeneous scenario subset, using spatial granularity threshold components and semantic granularity threshold components as input variables and the comprehensive operational efficiency index as the optimization objective, we obtain the Pareto optimal solution set and then obtain the spatial granularity threshold components and semantic granularity threshold components based on the Pareto optimal solution set.
[0012] A robot task planning system, comprising: The similarity acquisition module is used to perform basic semantic parsing and spatial semantic normalization on multiple tasks, and to calculate the spatial similarity and semantic similarity between each pair of multiple tasks after basic semantic parsing and spatial semantic normalization. Spatial similarity represents the degree of proximity of physical locations, and semantic similarity represents the degree of semantic matching. The task cluster acquisition module is used to fuse spatial similarity and semantic similarity to obtain a comprehensive similarity, perform clustering based on the comprehensive similarity to obtain multiple regional task clusters, and quantify the importance of regional tasks for each regional task cluster to obtain an importance score. The execution sequence acquisition module is used to take the regional task cluster with the highest importance score as the first execution unit, and use the first execution unit as the sorting starting anchor point to traverse all the remaining unsorted regional task clusters, obtain the correlation score between each regional task cluster and the last task cluster in the current sorted sequence, and complete the iterative sorting of regional task clusters according to the correlation score to obtain the regional task execution sequence.
[0013] Preferably, the task cluster acquisition module includes: The granularity threshold decoupling unit is used to decouple the distance threshold of the improved DBSCAN density clustering algorithm into spatial granularity threshold components and semantic granularity threshold components, wherein the spatial granularity threshold components and semantic granularity threshold components correspond to the clustering granularity benchmarks of spatial distance and semantic distance, respectively. The heat index acquisition unit is used to acquire real-time network heat data, and to acquire spatial heat index reflecting task request density and scheduling intensity, and semantic heat index reflecting the fluctuation and urgency of various task demands based on the real-time network heat data. The granularity threshold modulation unit is used to modulate the spatial granularity threshold component according to the spatial heat index, so that the modulated spatial granularity threshold component decreases as the spatial heat index increases. At the same time, it modulates the semantic granularity threshold component according to the semantic heat index, so that the modulated semantic granularity threshold component decreases as the semantic heat index increases. The distance threshold acquisition unit is used to perform weighted fusion of the modulated spatial granularity threshold component and the semantic granularity threshold component to obtain the comprehensive threshold distance; The task cluster acquisition unit is used to perform DBSCAN density clustering based on the comprehensive threshold distance to acquire multiple regional task clusters.
[0014] Preferably, the granularity threshold decoupling unit includes: The historical data acquisition subunit is used to acquire the robot's historical task data. Each historical task data includes scene feature attributes, clustering parameter attributes, and performance label attributes. The scene subset acquisition subunit is used to perform weighted fusion of multiple efficiency tags of efficiency tag attributes to obtain a normalized comprehensive operation efficiency index, and to divide historical operation task data according to scene feature attributes to obtain multiple homogeneous scene subsets with consistent scene features. The optimal solution set acquisition sub-unit is used to obtain the Pareto optimal solution set for each homogeneous scenario subset, with spatial granularity threshold components and semantic granularity threshold components as input variables and the comprehensive operation efficiency index as the optimization objective. The spatial granularity threshold components and semantic granularity threshold components are then obtained based on the Pareto optimal solution set.
[0015] Preferably, the scene feature attributes include the total number of tasks, spatial distribution density, and job scene type; the clustering parameter attributes include the number of clusters used when executing historical job tasks, the average number of tasks within a cluster, the number of tool switches within a cluster, and the total cross-regional travel distance; and the performance label attributes include the total task completion time, the total time spent switching tools, the average job time per unit task, and the task delay rate. Attached Figure Description
[0016] The invention will be further understood from the following description taken in conjunction with the accompanying drawings. The components in the drawings are not necessarily drawn to scale, but rather the emphasis is on illustrating the principles of the embodiments. In different views, the same reference numerals designate corresponding parts.
[0017] Figure 1 This is a schematic diagram of the overall process of a robot task planning method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the process of decoupling the distance threshold into spatial granularity threshold components and semantic granularity threshold components in one embodiment of the present invention. Figure 3 This is a schematic diagram of the overall process of a robot task planning method in another embodiment of the present invention. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to its embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of the invention.
[0019] It should be noted that when an element is referred to as being "fixed to" another element, it can be directly attached to the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.
[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.
[0021] In this invention, "first" and "second" do not represent a specific quantity or order, but are merely used to distinguish names.
[0022] like Figure 1 As shown, an embodiment of the present invention provides a robot task planning method, including the following steps: S1 performs basic semantic parsing and spatial semantic normalization on multiple tasks, and calculates the spatial similarity and semantic similarity between each pair of tasks after basic semantic parsing and spatial semantic normalization. Spatial similarity represents the proximity of physical locations, and semantic similarity represents the semantic matching degree.
[0023] Basic semantic parsing is performed on the tasks to extract the core objectives, scenario constraints, and implicit preconditions for each task, generating a structured task semantic description. The task semantic description includes, but is not limited to, job types such as handling, welding, inspection, and assembly; the end effectors and auxiliary tools required for task execution; the core action unit set mapped to the skill atom library; and job constraint patterns such as force control mode, visual positioning mode, and high-speed motion mode. The multiple semantic attributes included in each task are concatenated into a complete semantic text and embedded encoding is performed to generate a semantic feature vector.
[0024] The task undergoes semantic space normalization, identifying all spatial entities in the task's semantic description, such as region names, landmarks, workstation numbers, and location descriptions. This maps the fuzzy spatial information expressed in natural language to a pre-defined unified spatial coordinate system and region ID system, thereby labeling each task with standardized spatial coordinates, candidate region labels, and operational space range. This resolves the ambiguity and heterogeneity issues in natural language spatial representations, enabling automatic conversion of unstructured task instructions to standardized spatial labels.
[0025] For spatial similarity, it can be obtained by calculating the Euclidean distance or shortest path distance between tasks based on standardized spatial coordinates and then normalizing the results. Semantic similarity can be obtained by generating a semantic embedding vector for each task and calculating the semantic matching degree using cosine similarity.
[0026] S2 integrates spatial similarity and semantic similarity to obtain a comprehensive similarity score. Based on the comprehensive similarity score, clustering is performed to obtain multiple regional task clusters. For each regional task cluster, the importance of regional tasks is quantified to obtain an importance score.
[0027] A comprehensive similarity score is obtained by weighting and fusing normalized spatial and semantic similarities based on preset weights. To adapt to DBSCAN's distance-based clustering logic, the comprehensive similarity score is converted into a comprehensive distance, and then an improved DBSCAN density clustering algorithm is used for clustering to obtain multiple regional task clusters. The smaller the comprehensive distance, the more suitable the two tasks are to be assigned to the same regional task cluster. Thus, the resulting multiple regional task clusters each correspond to a physical work area. Tasks within a cluster can simultaneously satisfy the dual constraints of spatial proximity and semantic similarity, ensuring that the robot can continuously perform similar tasks within a single area, reducing the overhead of tool switching and action mode switching.
[0028] It can acquire and call the corresponding environmental raster map data, verify whether there are physical barriers such as walls, access control, and isolation fences in the spatial boundaries of each regional task cluster. If there are barriers, the clusters are automatically split, and finally the spatially continuous and semantically aggregated regional task cluster set is output.
[0029] As a preferred technical solution, clustering based on a comprehensive threshold distance to obtain multiple regional task clusters specifically includes the following steps: First, based on comprehensive similarity and comprehensive threshold distance, all regional task clusters that meet the comprehensive threshold distance are retrieved to obtain an initial neighborhood set; then, after neighborhood pruning of the initial neighborhood set, weighted neighborhoods are calculated and core points are determined, and then cluster expansion clustering is performed until all tasks are traversed to obtain multiple regional task clusters.
[0030] Specifically, iterate through each unvisited task t in the task set. i And mark task t i Tasks are marked as "visited," and then, based on the comprehensive distance, all tasks satisfying a comprehensive distance ≤ a comprehensive threshold distance are added to the initial neighborhood set. Next, neighborhood pruning is performed: traversing each task t in the initial neighborhood set... j Call the raster map connectivity interface to verify task t i With t j Whether the workspaces are physically connected; if there are insurmountable barriers, then t j Remove from the initial neighborhood set. Where t i Let t represent the i-th unvisited task in the task set. j The initial neighborhood set represents the j-th task. The task set is obtained by performing basic semantic parsing and spatial semantic normalization on multiple tasks. It represents the structured set of all tasks to be assigned to the robot for execution within the same scheduling cycle.
[0031] Finally, weighted neighborhoods are calculated, core points are determined, and cluster expansion is performed until all tasks are traversed, resulting in multiple regional task clusters. Since the specific process of clustering based on the improved DBSCAN density clustering algorithm is a standard technique in this field, it will not be elaborated upon here.
[0032] The importance score is obtained by weighting and integrating the task's urgency, the impact of failure, and the task's value. The task's urgency, the impact of failure, and the task's value can be scored by technical personnel according to preset rules and used as attribute tags for the task.
[0033] S3: Using the regional task cluster with the highest importance score as the first execution unit, and using the first execution unit as the sorting starting anchor point, traverse all the remaining unsorted regional task clusters, obtain the correlation score between each regional task cluster and the last task cluster in the current arrangement sequence, and complete the iterative sorting of regional task clusters according to the correlation score to obtain the regional task execution sequence.
[0034] The spatial adjacency correlation, logical dependency correlation, and resource reuse correlation between two regional task clusters can be calculated, and these three values can be normalized and weighted summed. Spatial adjacency correlation can be based on the reciprocal of the shortest travel distance between the centroids of the two regional clusters, adjusted for path difficulty (e.g., ramps, access control). Generally, the closer the distance and the smoother the travel, the higher the value; it is mainly used to measure the time cost of robot movement across regions. Logical dependency correlation can be obtained by inferring the business logic relationships between two regional task clusters using a large language model, such as strong pre-dependencies, weak continuation relationships, and parallel non-dependencies. For example, logical dependency correlation with strong dependencies is assigned the highest score, while no dependency is assigned the base score. Resource reuse correlation can be obtained by statistically analyzing the proportion of shared skills, shared tools, and shared sensors between two regional task clusters. Generally, the higher the sharing ratio, the higher the value; it is used to measure the cost of robot resource switching.
[0035] After obtaining the regional task execution sequence, the robot is driven to perform tasks based on the order of the sequence. For subtasks within each regional task cluster, path planning can be performed according to the importance of each subtask.
[0036] In summary, the robot task planning method of this invention obtains a comprehensive similarity by fusing spatial and semantic similarity, clusters based on the comprehensive similarity to obtain multiple regional task clusters, quantifies the importance of regional tasks for each regional task cluster to obtain an importance score, and finally obtains the regional task execution sequence based on the importance score and the correlation score. It realizes a dual-modal clustering method that integrates spatial and semantic similarity, which can achieve spatial proximity aggregation of regional tasks while ensuring the homogeneity of tasks within the region, reducing cross-regional empty driving costs and reducing job switching losses within the region. In addition, iterative sorting using the regional task cluster with the highest importance score as the starting anchor point and the correlation score between regional task clusters as the sorting criterion is conducive to maximizing the execution continuity between regions while ensuring the priority execution of high-importance tasks, thereby improving the overall operational efficiency of robots in multi-region, multi-task scenarios.
[0037] In one embodiment, the present invention further includes the following step: decoupling the distance threshold of the improved DBSCAN density clustering algorithm into a spatial granularity threshold component and a semantic granularity threshold component, wherein the spatial granularity threshold component and the semantic granularity threshold component correspond to the clustering granularity benchmarks of spatial distance and semantic distance, respectively.
[0038] The decoupled spatial granularity threshold component serves as the normalized upper limit of spatial distance, corresponding to the maximum spatial coverage granularity of a single-region task cluster. It limits the robot's cross-task movement distance and empty-running costs. Its physical quantity corresponds to the actual spatial threshold, meaning the shortest passable path distance between any two task points within the same cluster must not exceed this value. The decoupled semantic granularity threshold component serves as the normalized upper limit of semantic distance, corresponding to the minimum attribute homogeneity requirement of a single-region task cluster. It limits the time cost of robot tool switching and action mode switching. Its physical quantity corresponds to the actual semantic distance threshold, meaning the semantic feature difference between any two tasks within the same cluster must not exceed this value.
[0039] As a preferred technical solution, such as Figure 2 As shown, decoupling the distance threshold of the improved DBSCAN density clustering algorithm into spatial granularity threshold components and semantic granularity threshold components specifically includes the following steps: S4. Obtain the robot's historical task data. Each historical task data includes scene feature attributes, clustering parameter attributes, and performance label attributes.
[0040] Specifically, scene feature attributes include, but are not limited to, total number of tasks, spatial distribution density, semantic type ratio, operation scene type (warehouse handling / factory inspection / workshop assembly, etc.), robot body parameters, and environmental map size; including but not limited to, spatial similarity threshold, semantic similarity threshold, number of clusters, average number of tasks within a cluster, number of tool switches within a cluster, and total cross-area travel distance; performance label attributes include total task completion time, robot idle rate, total tool switching time, average operation time per unit task, and task delay rate, which serve as true labels for comprehensive operation performance evaluation.
[0041] S5 performs weighted fusion of multiple efficiency tags of efficiency tag attributes to obtain a normalized comprehensive operation efficiency index, and divides historical operation task data according to scene feature attributes to obtain multiple homogeneous scene subsets with consistent scene features.
[0042] First, normalize multiple performance tags of the performance tag attribute. Then, perform a weighted summation of the normalized performance tags to obtain the normalized comprehensive work performance index.
[0043] Historical task data is divided according to scene feature attributes to obtain multiple homogeneous scene subsets with consistent scene features. For example, according to task load level, it is divided into three homogeneous scene subsets: low load, medium load, and high load. According to scene type, it is divided into three homogeneous scene subsets: handling, inspection, and assembly.
[0044] S6. For each homogeneous scenario subset, with spatial granularity threshold components and semantic granularity threshold components as input variables and comprehensive operation efficiency index as the optimization objective, obtain the Pareto optimal solution set, and obtain the spatial granularity threshold components and semantic granularity threshold components based on the Pareto optimal solution set.
[0045] Specifically, for spatial granularity threshold components, within a preset upper and lower limit range of spatial granularity, based on historical real-world operational data, the optimal spatial granularity threshold component baseline is searched with the goal of minimizing overall movement cost. This means using the minimum total movement distance and the shortest task completion time as dual objectives to select the corresponding Pareto optimal solution set. Then, based on scenario preferences such as efficiency priority or cost priority, a compromise optimal solution is selected to obtain the spatial granularity threshold component baseline for that scenario. Finally, the spatial granularity threshold component baselines are classified and solidified according to the characteristics of the operational scenario to obtain the final spatial granularity threshold components.
[0046] For semantic granularity thresholds, based on historical real-world job data, candidate semantic thresholds can be traversed within a preset upper and lower limit range. The optimal granularity threshold component baseline is searched with the goal of achieving the best combination of switching cost and clustering efficiency, and the corresponding Pareto optimal solution set is selected. Then, the granularity threshold component baseline is differentiated and modified according to the differences in switching costs for different job types. For example, the granularity threshold component baseline is lowered by 10%-20% for high switching cost scenarios to tighten the semantic granularity, prioritize homogeneity within clusters, reduce switching losses, and obtain the final semantic granularity threshold component.
[0047] By filtering the Pareto optimal solution set, optimal solutions with different preferences, such as efficiency priority and switching loss priority, can be obtained. Based on this, for each homogeneous scenario subset, a gradient boosting regression tree model is trained with scenario features as input and the optimal solution's two components as output, outputting the default optimal spatial granularity threshold component and semantic granularity threshold component for the corresponding scenario. During the cold start phase, when sufficient historical data is unavailable, technical personnel can set corresponding initial default values as the spatial granularity threshold component and semantic granularity threshold component.
[0048] Traditional improved DBSCAN density clustering algorithms mostly employ a single comprehensive threshold, which is adjusted in tandem with spatial and semantic dimensions, making it impossible to optimize for a single cost. For example, when the spatial density of a task surges, it is generally only necessary to reduce the spatial granularity, but a single threshold will simultaneously reduce the semantic granularity, causing unnecessary cluster splitting and increasing cross-region switching losses. In this embodiment, after decoupling the two components, independent parameter tuning of each dimension can be achieved, significantly improving the granularity control accuracy and scene adaptability. Thus, by decoupling the distance threshold into spatial granularity threshold components and semantic granularity threshold components, which correspond to the optimization objectives of robot movement cost and tool switching cost respectively, it breaks through the granularity control limitations of a single comprehensive threshold and can differentiate the modulation based on the priority of the two types of costs, thereby improving the granularity control accuracy and scene adaptability.
[0049] In one embodiment, such as Figure 3 As shown, the robot task planning method further includes the following steps: S7, obtain real-time network heat data, and obtain spatial heat index to reflect task request density and scheduling intensity, and semantic heat index to reflect the fluctuation and urgency of various job task demands based on the real-time network heat data.
[0050] Real-time network heat data includes the intensity and fluctuation of real-time task demand reflected in networked channels such as industrial internet platforms and business order systems. It includes spatial heat index and semantic heat index. The relevant data of spatial heat index comes from network-side scheduling data such as the frequency of real-time task requests in the park, regional order density, equipment scheduling heat, and production workstation load rate. It can be calculated based on the historical average task density of the same period to calculate the relative multiple of the current regional task density, and combined with the regional accessibility weighting, and finally normalized to the [0,1] interval. The relevant data of semantic heat index comes from network-side business data such as the fluctuation of task type proportion in business systems, the number of special task triggers, the frequency of production plan changes, and the surge of industry event-driven tasks. It can be calculated based on the historical proportion of similar tasks to calculate the relative multiple of the current hot operation type task volume, and combined with the task urgency weighting, and normalized to the [0,1] interval.
[0051] S8, modulate the decoupled spatial granularity threshold component according to the spatial heat index, so that the modulated spatial granularity threshold component decreases as the spatial heat index increases. At the same time, modulate the decoupled semantic granularity threshold component according to the semantic heat index, so that the modulated semantic granularity threshold component decreases as the semantic heat index increases.
[0052] First, obtain the spatial heat index and semantic heat index corresponding to the spatial heat index and semantic heat index respectively. The two can be adjusted and set by technical personnel based on experience, such as setting them to 0.5. Then, combine the spatial heat index and spatial heat index to modulate the spatial granularity threshold component, and combine the semantic heat index and semantic heat index to modulate the semantic granularity threshold component.
[0053] For example, the modulated spatial granularity threshold component = decoupled spatial granularity threshold component / (1 + exp(ks × (spatial heat index - spatial heat inflection point value))), and the modulated semantic granularity threshold component = decoupled semantic granularity threshold component / (1 + exp(kv × (semantic heat index - semantic heat inflection point value))), where ks and kv are heat sensitivity coefficients, used to control the rate at which the threshold changes with heat, and their values are generally in the range of 0.5-3. The higher the value of the heat sensitivity coefficient, the more significant the influence of network heat on the modulation of the granularity threshold component. It can be understood that the lower the network heat index, the closer the modulated corresponding granularity threshold component is to the baseline value, i.e., the decoupled granularity threshold airflow, thus ensuring a larger clustering granularity and lower losses in cross-regional movement and tool switching. Conversely, the higher the network heat index, the smaller the modulated corresponding granularity threshold component, and the finer the clustering granularity, thereby improving scheduling flexibility and multi-robot parallel capabilities to adapt to high-concurrency business scenarios. Here, exp() represents the natural exponential function.
[0054] Specifically, the spatial granularity threshold component is modulated using a combination of the spatial heat index and the spatial heat inflection point value. When the spatial heat index rises, such as with a surge in regional task density, the spatial granularity threshold component is reduced using an S-shaped modulation function, thus refining the spatial granularity and splitting large clusters into smaller ones to accommodate parallel scheduling of multiple robots and alleviate regional congestion. Conversely, it is amplified when the spatial heat index decreases. Similarly, the semantic granularity threshold component is modulated using a combination of the semantic heat index and the semantic heat inflection point value. When the semantic heat index rises, such as with an explosion of a certain type of task, the semantic granularity threshold component is reduced using an S-shaped modulation function, tightening the semantic granularity and allowing explosive task clusters to form purer task clusters to maximize continuous task benefits. Conversely, it is relaxed when the semantic granularity threshold component is stable.
[0055] During cluster expansion, the maximum spatial span and maximum semantic dispersion of the current cluster are statistically analyzed in real time. If the maximum spatial span exceeds the corresponding spatial granularity threshold component or the maximum semantic dispersion exceeds the corresponding semantic granularity threshold component, expansion in that direction is stopped to avoid breaking cost constraints in a single dimension.
[0056] S9, weighted fusion of the modulated spatial granularity threshold component and semantic granularity threshold component to obtain the comprehensive threshold distance; wherein, DBSCAN density clustering is performed based on the comprehensive threshold distance to obtain multiple regional task clusters.
[0057] After the current batch of tasks is clustered and executed, the overall operational efficiency index for this batch can be calculated in real time and compared with the preset expected efficiency index to obtain the efficiency deviation rate. If the efficiency deviation rate is greater than the preset deviation rate threshold, the following strategies are implemented: if it is due to excessive intra-cluster load or excessively long task queuing time, the overall threshold distance is reduced by an adaptive step size to refine the clustering granularity; if it is due to frequent cross-regional movement or excessively high empty running rate, the overall threshold distance is increased by an adaptive step size to expand the coverage of a single cluster. At the same time, upper and lower limits of the overall threshold distance can be set to ensure that the overall threshold distance is always within a reasonable range and to avoid clustering failure caused by over-correction.
[0058] In this way, by decoupling the distance threshold into two independent components, spatial granularity threshold and semantic granularity threshold, the optimal baseline under different scenarios is first mined by using historical real-world operational efficiency data as input. Then, real-time network heat data is fused to perform differentiated elastic modulation on the spatial granularity threshold component and the semantic granularity threshold component, and finally fused into a comprehensive threshold distance. This allows the comprehensive threshold distance to simultaneously match data distribution patterns, historical best operational efficiency, and real-time business demand intensity, solving the problem that static baselines cannot match business peaks and the lag of sudden tasks.
[0059] One embodiment of the present invention also provides a robot task planning system, which includes a similarity acquisition module, a task cluster acquisition module, and an execution sequence acquisition module.
[0060] The similarity acquisition module performs basic semantic parsing and spatial semantic normalization on multiple tasks, and calculates the spatial similarity and semantic similarity between each pair of tasks after basic semantic parsing and spatial semantic normalization. Spatial similarity represents the proximity of physical locations, and semantic similarity represents the semantic matching degree. The task cluster acquisition module fuses spatial similarity and semantic similarity to obtain a comprehensive similarity. Based on the comprehensive similarity, it performs clustering to obtain multiple regional task clusters. For each regional task cluster, it quantifies the importance of regional tasks and obtains an importance score.
[0061] The execution sequence acquisition module uses the regional task cluster with the highest importance score as the first execution unit, and uses the first execution unit as the sorting starting anchor point to traverse all the remaining unsorted regional task clusters. It obtains the correlation score between each regional task cluster and the last task cluster in the current sorted sequence, and completes the iterative sorting of regional task clusters according to the correlation score to obtain the regional task execution sequence.
[0062] As a preferred technical solution, the task cluster acquisition module includes a granularity threshold decoupling unit, a heat index acquisition unit, a granularity threshold modulation unit, a distance threshold acquisition unit, and a task cluster acquisition unit.
[0063] The granularity threshold decoupling unit is used to decouple the distance threshold of the improved DBSCAN density clustering algorithm into spatial granularity threshold components and semantic granularity threshold components, where the spatial granularity threshold components and semantic granularity threshold components correspond to the clustering granularity benchmarks of spatial distance and semantic distance, respectively. The heat index acquisition unit is used to acquire real-time network heat data, and based on the real-time network heat data, acquire a spatial heat index that reflects the density of task requests and the intensity of scheduling, and a semantic heat index that reflects the fluctuation and urgency of various job task demands.
[0064] The granularity threshold modulation unit modulates the spatial granularity threshold component according to the spatial heat index, so that the modulated spatial granularity threshold component decreases as the spatial heat index increases. Simultaneously, it modulates the semantic granularity threshold component according to the semantic heat index, so that the modulated semantic granularity threshold component decreases as the semantic heat index increases. The distance threshold acquisition unit performs weighted fusion of the modulated spatial granularity threshold component and the semantic granularity threshold component to obtain a comprehensive threshold distance. The task cluster acquisition unit performs DBSCAN density clustering based on the comprehensive threshold distance to obtain multiple regional task clusters.
[0065] The granularity threshold decoupling unit includes a historical data acquisition subunit, a scene subset acquisition subunit, and an optimal solution set acquisition subunit.
[0066] The historical data acquisition subunit is used to acquire historical task data of the robot. Each historical task data includes scene feature attributes, clustering parameter attributes, and performance label attributes. The scene subset acquisition subunit is used to perform weighted fusion of multiple performance labels of the performance label attributes to obtain a normalized comprehensive task performance index. It also divides the historical task data according to the scene feature attributes to obtain multiple homogeneous scene subsets with consistent scene features. The optimal solution set acquisition subunit is used to obtain the Pareto optimal solution set for each homogeneous scene subset, using spatial granularity threshold components and semantic granularity threshold components as input variables and the comprehensive task performance index as the optimization objective. It also obtains the spatial granularity threshold components and semantic granularity threshold components based on the Pareto optimal solution set.
[0067] Scene feature attributes include total number of tasks, spatial distribution density, and job scene type. Clustering parameter attributes include the number of clusters used when executing historical jobs, the average number of tasks within a cluster, the number of tool switches within a cluster, and the total cross-regional travel distance. Performance label attributes include total task completion time, total tool switching time, average job time per unit task, and task delay rate.
[0068] In summary, the robot task planning system of this invention obtains a comprehensive similarity by fusing spatial and semantic similarity, clusters based on the comprehensive similarity to obtain multiple regional task clusters, quantifies the importance of regional tasks for each regional task cluster to obtain an importance score, and finally obtains the regional task execution sequence based on the importance score and the correlation score. It realizes a dual-modal clustering method that integrates spatial and semantic similarity, which can achieve spatial proximity aggregation of regional tasks while ensuring the homogeneity of tasks within the region, reducing cross-regional empty driving costs and reducing job switching losses within the region. In addition, iterative sorting using the regional task cluster with the highest importance score as the starting anchor point and the correlation score between regional task clusters as the sorting criterion is conducive to maximizing the execution continuity between regions while ensuring the priority execution of high-importance tasks, thereby improving the overall operational efficiency of robots in multi-region, multi-task scenarios.
[0069] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0070] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.
Claims
1. A robot task planning method, characterized in that, Includes the following steps: Basic semantic parsing and spatial semantic normalization are performed on multiple tasks. The spatial similarity and semantic similarity between each pair of tasks after basic semantic parsing and spatial semantic normalization are calculated. Spatial similarity represents the proximity of physical locations, and semantic similarity represents the semantic matching degree. Spatial similarity and semantic similarity are fused to obtain a comprehensive similarity. Based on the comprehensive similarity, clustering is performed to obtain multiple regional task clusters. For each regional task cluster, the importance of regional tasks is quantified to obtain an importance score. The task cluster with the highest importance score is taken as the first execution unit. The remaining unsorted task clusters are traversed with the first execution unit as the sorting anchor point. The correlation score between each task cluster and the last task cluster in the current sorting sequence is obtained. The task clusters are iteratively sorted according to the correlation score to obtain the task execution sequence. Clustering based on comprehensive similarity to obtain multiple regional task clusters: Based on comprehensive similarity, an improved DBSCAN density clustering algorithm is used to perform clustering to obtain multiple regional task clusters; The distance threshold of the improved DBSCAN density clustering algorithm is decoupled into a spatial granularity threshold component and a semantic granularity threshold component, where the spatial granularity threshold component and the semantic granularity threshold component correspond to the clustering granularity benchmarks of spatial distance and semantic distance, respectively. It also includes the following steps: Obtain real-time network heat data, and based on the real-time network heat data, obtain a spatial heat index to reflect the density of task requests and the intensity of scheduling, and a semantic heat index to reflect the fluctuation and urgency of various task demands. The spatial granularity threshold component is modulated according to the spatial heat index, so that the modulated spatial granularity threshold component decreases as the spatial heat index increases. At the same time, the semantic granularity threshold component is modulated according to the semantic heat index, so that the modulated semantic granularity threshold component decreases as the semantic heat index increases. The modulated spatial granularity threshold components and semantic granularity threshold components are weighted and fused to obtain the comprehensive threshold distance; Among them, clustering is performed based on comprehensive threshold distance to obtain multiple regional task clusters.
2. The robot task planning method as described in claim 1, characterized in that, Clustering based on comprehensive threshold distance to obtain multiple regional task clusters includes the following steps: Based on comprehensive similarity and comprehensive threshold distance, all regional task clusters that meet the comprehensive threshold distance are retrieved to obtain the initial neighborhood set; After performing neighborhood pruning on the initial neighborhood set, weighted neighborhoods are calculated and core points are determined. Then, cluster expansion and clustering are performed until all tasks are traversed, resulting in multiple regional task clusters.
3. The robot task planning method as described in claim 2, characterized in that, Decoupling the distance threshold of the improved DBSCAN density clustering algorithm into spatial granularity threshold components and semantic granularity threshold components specifically includes the following steps: Obtain the robot's historical task data, where each historical task data includes scene feature attributes, clustering parameter attributes, and performance label attributes; Multiple efficiency tags of efficiency tag attributes are weighted and fused to obtain a normalized comprehensive operation efficiency index. Historical operation task data are divided according to scene feature attributes to obtain multiple homogeneous scene subsets with consistent scene features. For each homogeneous scenario subset, using spatial granularity threshold components and semantic granularity threshold components as input variables and the comprehensive operational efficiency index as the optimization objective, we obtain the Pareto optimal solution set and then obtain the spatial granularity threshold components and semantic granularity threshold components based on the Pareto optimal solution set.
4. A robot task planning system for implementing the robot task planning method as described in any one of claims 1-3, characterized in that, include: The similarity acquisition module is used to perform basic semantic parsing and spatial semantic normalization on multiple tasks, and to calculate the spatial similarity and semantic similarity between each pair of multiple tasks after basic semantic parsing and spatial semantic normalization. Spatial similarity represents the degree of proximity of physical locations, and semantic similarity represents the degree of semantic matching. The task cluster acquisition module is used to fuse spatial similarity and semantic similarity to obtain a comprehensive similarity, perform clustering based on the comprehensive similarity to obtain multiple regional task clusters, and quantify the importance of regional tasks for each regional task cluster to obtain an importance score. The execution sequence acquisition module is used to take the regional task cluster with the highest importance score as the first execution unit, and use the first execution unit as the sorting starting anchor point to traverse all the remaining unsorted regional task clusters. It obtains the correlation score between each regional task cluster and the last task cluster in the current sorted sequence, and completes the iterative sorting of regional task clusters according to the correlation score to obtain the regional task execution sequence.
5. The robot task planning system as described in claim 4, characterized in that, The task cluster acquisition module includes: The granularity threshold decoupling unit is used to decouple the distance threshold of the improved DBSCAN density clustering algorithm into spatial granularity threshold components and semantic granularity threshold components, wherein the spatial granularity threshold components and semantic granularity threshold components correspond to the clustering granularity benchmarks of spatial distance and semantic distance, respectively. The heat index acquisition unit is used to acquire real-time network heat data, and to acquire spatial heat index reflecting task request density and scheduling intensity, and semantic heat index reflecting the fluctuation and urgency of various task demands based on the real-time network heat data. The granularity threshold modulation unit is used to modulate the spatial granularity threshold component according to the spatial heat index, so that the modulated spatial granularity threshold component decreases as the spatial heat index increases. At the same time, it modulates the semantic granularity threshold component according to the semantic heat index, so that the modulated semantic granularity threshold component decreases as the semantic heat index increases. The distance threshold acquisition unit is used to perform weighted fusion of the modulated spatial granularity threshold component and the semantic granularity threshold component to obtain the comprehensive threshold distance; The task cluster acquisition unit is used to perform DBSCAN density clustering based on the comprehensive threshold distance to acquire multiple regional task clusters.
6. The robot task planning system as described in claim 5, characterized in that, The granularity threshold decoupling unit includes: The historical data acquisition subunit is used to acquire the robot's historical task data. Each historical task data includes scene feature attributes, clustering parameter attributes, and performance label attributes. The scene subset acquisition subunit is used to perform weighted fusion of multiple efficiency tags of efficiency tag attributes to obtain a normalized comprehensive operation efficiency index, and to divide historical operation task data according to scene feature attributes to obtain multiple homogeneous scene subsets with consistent scene features. The optimal solution set acquisition sub-unit is used to obtain the Pareto optimal solution set for each homogeneous scenario subset, with spatial granularity threshold components and semantic granularity threshold components as input variables and the comprehensive operation efficiency index as the optimization objective. The spatial granularity threshold components and semantic granularity threshold components are then obtained based on the Pareto optimal solution set.
7. The robot task planning system as described in claim 6, characterized in that, Scene feature attributes include total number of tasks, spatial distribution density, and job scene type. Clustering parameter attributes include the number of clusters used when executing historical jobs, the average number of tasks within a cluster, the number of tool switches within a cluster, and the total cross-regional travel distance. Performance label attributes include total task completion time, total tool switching time, average job time per unit task, and task delay rate.
Citation Information
Patent Citations
Multi-task intelligent coordination execution method and system based on health care accompanying robot
CN120494765A
Unmanned aerial vehicle multi-task collaborative planning method based on large language model
CN121523407A
