Robot walking task intelligent scheduling method and system based on historical data analysis

Abstract

The invention provides a robot walking task intelligent scheduling method and system based on historical data analysis, and the method comprises the steps: obtaining space information and historical walking data, and the historical walking data comprises a walking path, walking time and an obstacle condition; extracting spatial features of the walking path based on the spatial information, segmenting the walking path, and generating a region set corresponding to each path segment; for the same area, historical walking time and historical obstacle conditions corresponding to the area are extracted from historical walking data, and a current obstacle prediction sequence corresponding to the area is generated; and according to the spatial information and the current obstacle prediction sequence of each region in the region set, generating a sorting result of the regions in the region set through the trained neural network model, and generating a target patrol path in combination with the spatial information. According to the technical scheme provided by the invention, an efficient patrol path can be analyzed and formulated according to historical data.

Description

Technical Field

[0001] The present application relates to the technical field of robot scheduling management, and in particular to a method and system for intelligent scheduling of robot walking tasks based on historical data analysis. Background Art

[0002] In today's era of rapid technological development, robots have been widely used in many fields. Robots are everywhere in various service scenarios such as patrolling, security, and transportation. As a key link in robot applications, the rationality and efficiency of robot walking task scheduling directly affects the operating efficiency, cost, and service quality of the entire system.

[0003] Traditional robot walking task scheduling methods are mostly based on simple rules. For example, in some simple service scenarios, tasks may be scheduled in the order of arrival, or specific robots may be fixedly assigned to different types of tasks. Although this method is easy to understand and implement, it lacks sufficient consideration of the complexity and dynamics of tasks. In practical applications, task requirements, environmental conditions, and the robot's own state are constantly changing. Traditional scheduling methods are difficult to adjust scheduling strategies in a timely manner according to these dynamic changes, resulting in insufficient utilization of robot resources and low task execution efficiency.

[0004] In the past scheduling process, a large amount of historical scheduling data has not been fully mined and utilized. These historical data contain rich information, such as the operating efficiency of robots under different conditions. In particular, in the application scenario of unmanned patrol in garages, robots need to deal with different areas, different time periods, and different path structures, while avoiding moving and stationary obstacles to ensure safety and patrol efficiency. How to conduct in-depth analysis of historical data, extract valuable information, and apply it to task scheduling, and formulate more scientific and reasonable scheduling strategies are problems that need to be solved urgently by existing technologies. Summary of the invention

[0005] In view of this, the present application provides a method and system for intelligent scheduling of robot walking tasks based on historical data analysis to solve the technical problem of relatively low robot patrol efficiency in the prior art.

[0006] In a first aspect, the present application provides a method for intelligently scheduling robot walking tasks based on historical data analysis, the method comprising:

[0007] Obtain spatial information and historical walking data, where the historical walking data includes walking path, walking time and obstacle conditions;

[0008] Extracting spatial features of the walking path based on the spatial information, segmenting the walking path, and generating a set of regions corresponding to each path segment;

[0009] For the same area, extract the historical walking time and historical obstacle conditions corresponding to the area from the historical walking data, and generate a current obstacle prediction sequence corresponding to the area;

[0010] According to the spatial information and the current obstacle prediction sequence of each area in the area set, the sorting results of the areas in the area set are generated through the trained neural network model, and the target patrol path is generated in combination with the spatial information.

[0011] Preferably, extracting the spatial features of the walking path based on the spatial information and segmenting the walking path includes:

[0012] Using a convolutional neural network to process the data in the spatial information, and identify landmarks and environmental features of the walking path;

[0013] Based on the distribution of landmarks, environmental features, and obstacles in the walking path, an environmental complexity assessment mechanism is used to assess the environmental complexity of different parts of the walking path;

[0014] A dynamic segmentation strategy is adopted, and different segments in the walking path are segmented using different segmentation standards based on environment complexity evaluation data.

[0015] Preferably, the method for intelligent scheduling of robot walking tasks based on historical data analysis further includes: utilizing spatial topological relationships to merge adjacent path segments having similar spatial characteristics and traffic properties into one area.

[0016] Preferably, when extracting the historical walking time corresponding to the area from the historical walking data, the precise timestamp of each time the robot enters and leaves the area is extracted, and the historical walking time of each time in the area is obtained by calculating the difference between the two;

[0017] When extracting historical obstacle conditions, the obstacle types are classified and recorded; wherein the obstacle types include fixed obstacles, mobile obstacles and temporary obstacles, and different types of obstacles are marked separately in the historical data.

[0018] Preferably, when generating the current obstacle prediction sequence corresponding to the area, a state space-based method is used to model the historical travel time and historical obstacle conditions, and a Kalman filter algorithm is used for state estimation and prediction;

[0019] The robot's state prediction formula is:

[0020] x_hat(k|k-1)=F*x_hat(k-1|k-1)+B*u(k);

[0021] Among them, at time k, the robot's state x_hat(k|k-1) at time k is predicted based on the state estimate x_hat(k-1|k-1) at time k-1, the robot's motion model F, and the control input u(k). The motion model F and the control input u(k) are associated with the robot's path planning and walking state. B is the control input matrix, which describes the impact of the control input u(k) on the state; x_hat(k|k-1) represents the predicted state of the robot at time k, including the position, speed, and the prediction of the probability of obstacles in the area;

[0022] Using the cross-validation method, historical data is divided into training set and test set. The training set data is used to generate a prediction sequence, which is compared with the test set data. The accuracy of the prediction is evaluated by calculating the mean square error and mean absolute error indicators. If the prediction result does not meet the preset accuracy requirements, the parameters of the prediction model are adjusted or the prediction model is replaced.

[0023] Preferably, the trained neural network model is a recurrent neural network model;

[0024] Based on the recurrent neural network, the current obstacle prediction sequence of each area is processed to capture the spatial dependency of each area and the temporal dependency in the current obstacle prediction sequence;

[0025] Based on the spatial dependency of each region and the temporal dependency in the current obstacle prediction sequence, a sorting result of the regions in the region set is generated in combination with a sorting basis;

[0026] The sorting criteria include the obstacle risk level of the area, the estimated patrol time, and the distance from the starting point or important target point; the neural network model outputs the priority ranking of each area through comprehensive learning and analysis of the sorting criteria.

[0027] Preferably, generating a target patrol path in combination with spatial information comprises: based on the sorting result of the areas in the area set, in combination with the spatial information, using Dijkstra algorithm to perform path planning on the sorted areas to find the target patrol path;

[0028] After the target patrol path is generated, the actual movement process of the robot is simulated to check whether the impassable area and the repeated pass area on the target patrol path are within the preset threshold range. If not, the path is adjusted or replanned.

[0029] In a second aspect, the present application provides a robot walking task intelligent scheduling system based on historical data analysis, comprising:

[0030] An acquisition module is used to acquire spatial information and historical walking data, wherein the historical walking data includes walking path, walking time and obstacle conditions;

[0031] A partitioning module, used to extract the spatial features of the walking path based on the spatial information, segment the walking path, and generate a set of regions corresponding to each path segment;

[0032] A sorting module is used to extract the historical walking time and historical obstacle conditions corresponding to the same area from the historical walking data, and generate a current obstacle prediction sequence corresponding to the area;

[0033] The generation module is used to generate the sorting results of the areas in the area set according to the spatial information and the current obstacle prediction sequence of each area in the area set through the trained neural network model, and generate the target patrol path in combination with the spatial information.

[0034] In a third aspect, the present application provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to implement the intelligent scheduling method for robot walking tasks based on historical data analysis as described in the first aspect.

[0035] In a fourth aspect, the present application provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed by a processor, the intelligent scheduling method for robot walking tasks based on historical data analysis described in the first aspect is implemented.

[0036] The method and system for intelligent scheduling of robot walking tasks based on historical data analysis provided in this application have at least the following beneficial effects:

[0037] The technical solution provided by the present application can obtain spatial information and collect historical walking data at the same time. These data record the walking path of the robot in the past patrolling the garage, the time spent on each walk, and the various obstacles encountered. Based on the acquired spatial information, the historical walking path is deeply analyzed, its spatial characteristics are extracted, and the walking path is reasonably segmented. Each segment corresponds to a specific area, and many areas together constitute a regional set. For each area, the corresponding historical walking time and historical obstacles are accurately extracted from the historical walking data. Using data analysis and prediction algorithms, the current obstacle prediction sequence is generated to predict the probability and type of obstacles that may appear in the area at different times. The spatial information and the current obstacle prediction sequence of each area in the regional set are input into the trained neural network model. By learning a large amount of historical data, various factors can be comprehensively considered, the areas in the regional set can be sorted, and the optimal patrol order can be determined. Then, combined with the spatial information, a target patrol path covering all areas and efficiently avoiding obstacles is generated.

[0038] In this embodiment, the patrol path can be dynamically adjusted based on real-time obstacle prediction and spatial information. By analyzing and predicting historical data, the path to avoid obstacles is planned in advance, and the patrol order of the area is optimized, which greatly improves the patrol efficiency. Through scientific area division and path planning, it is ensured that the robot can patrol every corner of the garage comprehensively and carefully, and accurately discover various potential problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of an intelligent scheduling method for robot walking tasks based on historical data analysis in one embodiment of the present application is shown;

[0040] Figure 2 A schematic diagram of an obstacle prediction process based on Kalman filtering in one embodiment of the present application is shown;

[0041] Figure 3 A schematic diagram of an intelligent scheduling system for robot walking tasks based on historical data analysis in one embodiment of the present application is shown;

[0042] Figure 4 A schematic diagram of a computer device in one embodiment of the present application is shown. DETAILED DESCRIPTION

[0043] Example 1

[0044] The intelligent scheduling method for robot walking tasks based on historical data analysis provided by the present application can be applied to the application scenario of unmanned patrol in garages. In this scenario, vehicles frequently enter and exit, and the spatial layout is complex. In order to ensure the safety and order of the garage and ensure the safety of vehicles and facilities, the garage needs to be patrolled regularly. The intelligent scheduling method for robot walking tasks based on historical data analysis has significant application value in such scenarios. Of course, the intelligent scheduling method for robot walking tasks based on historical data analysis provided by the present application can also be applied to scenarios such as industrial plant patrols, large warehouse inventory scenarios, campus security patrol scenarios, hospital tours and inspections, etc.

[0045] See also Figure 1 , an intelligent scheduling method for robot walking tasks based on historical data analysis provided in one embodiment of the present application may include the following steps.

[0046] S101, obtaining spatial information and historical walking data, wherein the historical walking data includes walking path, walking time and obstacle conditions;

[0047] S102, extracting spatial features of the walking path based on the spatial information, segmenting the walking path, and generating a region set corresponding to each path segment;

[0048] S103, for the same area, extracting the historical walking time and historical obstacle conditions corresponding to the area from the historical walking data, and generating a current obstacle prediction sequence corresponding to the area;

[0049] S104, generating a sorting result of the areas in the area set based on the spatial information and the current obstacle prediction sequence of each area in the area set through the trained neural network model, and generating a target patrol path in combination with the spatial information.

[0050] In this embodiment, obtaining the spatial information of the garage is the basis for the robot to conduct effective patrols. This information includes the layout structure of the garage, such as the distribution of floors, the shape and size of each floor, the direction and width of the lanes, the arrangement and numbering rules of parking spaces, etc. By mastering this spatial information, the robot can build an overall understanding of the garage and clarify its position and range of action in the garage. For example, the robot can understand which areas are passages, which are parking spaces, and the connection methods between different floors, etc., providing important spatial references for subsequent path planning.

[0051] The walking paths in the historical walking data record the trajectories that the robot has walked during past patrols. These paths reflect the robot's action choices in different situations, including regular patrol routes under normal circumstances, and detour routes when encountering obstacles or special circumstances. By analyzing these walking paths, the robot can learn which routes are efficient and commonly used, and which areas are prone to traffic difficulties and require advance planning for detours. For example, the robot may find that a passage connecting different parking areas has a large volume of traffic and slow traffic during a specific time period, so it may choose other more suitable routes in subsequent patrols, or avoid patrolling this area at a specific time.

[0052] The walking time information records the time it takes the robot to complete a specific patrol task or walk a specific path in different time periods. This is crucial for optimizing patrol plans. By analyzing historical walking time, the robot can understand the impact of traffic conditions in the garage at different time periods on its patrol speed. For example, during the morning and evening rush hours on weekdays, due to the frequent entry and exit of vehicles, the passages in the garage may be congested, resulting in longer robot walking time. After mastering these rules, the robot can adjust the patrol strategy during peak hours, give priority to relatively unobstructed routes, or plan extra time in advance to complete patrol tasks to ensure that patrols are completed on time.

[0053] The obstacles in the historical walking data record the various factors that hinder the robot's normal passage encountered during past patrols. These obstacles include illegal parking of vehicles, ground construction, equipment failures occupying passages, etc. By analyzing the historical obstacles, the robot can summarize the types and frequencies of obstacles that are prone to occur in different areas and time periods. For example, the robot may find that every Monday morning, there are often vehicles parked illegally near a certain parking space, blocking the passage. After understanding this information, the robot can prepare for the response in advance before patrolling, such as adjusting the patrol route to avoid the area, or being vigilant before arriving at the area, and promptly discovering and dealing with possible obstacles to ensure the smooth progress of the patrol mission.

[0054] By combining spatial information, historical walking paths and walking time, the robot can optimize patrol route planning and choose the fastest and most unobstructed routes for patrols at different time periods, reducing unnecessary detours and waiting time, thereby improving overall patrol efficiency. For example, during off-peak hours, the robot can choose a longer route that can cover more areas to ensure comprehensive inspections; during peak hours, it can choose a shorter route that avoids congested areas to ensure that the patrol task is completed on time.

[0055] Based on its understanding of historical obstacles, the robot can predict possible obstacles in advance and develop corresponding response strategies. This allows the robot to respond more calmly when encountering obstacles, avoiding mission interruptions or delays caused by unexpected obstacles. For example, when the robot predicts that there may be ground construction obstacles in a certain area, it can plan a route to bypass the area in advance, or wait until the construction is completed before patrolling, ensuring the continuity of patrol work.

[0056] Based on the acquired garage space information, the spatial features of the historical walking paths are extracted. This means in-depth analysis of the specific attributes of the path in the garage space, such as the shape and size of the area passed by the path, the relative position relationship with surrounding facilities (such as pillars, walls, entrances and exits, etc.), and the direction and curvature of the path. By extracting these features, the robot can more accurately understand the connection between its own movement trajectory in the garage and the layout of the garage space. For example, it can identify that some path segments are close to the corners of the garage and are easily affected by the parking of vehicles; or some path segments are along longer straight passages, where traffic is relatively smooth but attention needs to be paid to high-speed vehicles.

[0057] Based on the extraction of spatial features, the walking path is reasonably segmented. This is not a simple random division, but is based on the similarity of spatial features, the difference in functional areas, and the difference in potential patrol needs. For example, the staircase or elevator area connecting different floors is divided into one section, and the adjacent parking spaces on the same floor are divided into one section. Each section of the path corresponds to a specific area, and many such areas are combined to form an area set. In this way, the complex space of the garage is subdivided into small areas with clear characteristics and boundaries, which makes it easier for the robot to conduct targeted analysis and processing of each area.

[0058] Through the generated area set, the robot can formulate personalized patrol strategies for each area. Different areas may have different characteristics and potential problems. For example, the parking area needs to focus on checking the parking conditions of vehicles and whether there are any abnormal items left behind; the passage area needs to pay attention to whether there are obstacles blocking the passage and whether the ground is damaged. In view of these differences, the robot can adjust the patrol speed, stay time and inspection focus in each area. For example, in the parking area with dense vehicles, the robot slows down and carefully checks the condition of each vehicle; while in the relatively unobstructed passage area, the patrol speed is appropriately accelerated to improve the overall patrol efficiency.

[0059] When the robot finds an abnormal situation during patrol, it can quickly determine the specific area where the abnormality occurred because the walking path has been reasonably segmented and formed into area collections. This helps to quickly respond to and handle problems and improve the efficiency of problem solving. For example, if the robot detects a smoke sensor alarm in a certain area, it can immediately accurately determine the specific area where the alarm is located based on the area collection information, and quickly go to the scene for further inspection and processing, avoiding delays that may lead to the expansion of the problem.

[0060] After the walking path is segmented and a set of regions is generated, it is convenient to independently count and analyze the relevant data of each region. For example, the number of vehicles entering and exiting each region in different time periods, the frequency and type of obstacles, etc. can be counted. By analyzing these data, we can gain an in-depth understanding of the usage and potential problems of each region, providing strong support for subsequent optimization. For example, if it is found that vehicles are often illegally parked in a certain parking space area, we can consider adjusting the signs of the area or strengthening patrols.

[0061] Based on the analysis results of regional data, the robot can optimize and adjust the patrol path. If it is found that a certain area has more problems within a specific time period and needs more frequent patrols, the path can be adjusted to increase the number of patrols in that area; or if it is found that the traffic efficiency of certain path sections is low, you can try to find a better alternative path. This path optimization based on data analysis can continuously improve the execution effect of the robot's patrol tasks and make it more adaptable to the actual situation of the garage.

[0062] Accurately extracting the historical travel time and historical obstacles corresponding to a specific area from historical travel data is the key to gaining a deep understanding of the past conditions of the area. The historical travel time reflects the time required to pass through the area at different times, which is affected by many factors such as traffic volume and vehicle parking methods.

[0063] Based on the extracted historical data, a current obstacle prediction sequence corresponding to the area is generated, enabling the robot to predict the obstacles that may currently appear in the area. By analyzing and modeling historical data, combined with factors such as the current time, date, and special events, the prediction sequence can give the probability of specific obstacles appearing in the area at different time points.

[0064] With the current obstacle prediction sequence, the robot can plan a more reasonable patrol route in advance. If it is predicted that a certain area may have a high probability of obstacles at a specific time, the robot can choose a route to avoid the area in advance, or adjust the patrol order and go to other relatively unobstructed areas first, and then patrol the area after the time period when the obstacles may be eliminated.

[0065] According to the obstacle prediction sequence, the robot can reasonably allocate its own resources, such as power and computing resources. For areas with a high probability of obstacles, the robot can make full preparations in advance, such as increasing the scanning frequency of sensors to more sensitively detect potential obstacles; while for areas with a low probability of obstacles, it can appropriately reduce resource consumption and extend patrol time. This not only ensures effective monitoring of key areas, but also improves overall patrol efficiency and reduces energy consumption.

[0066] The current obstacle prediction sequence is not only helpful for the robot's patrol mission, but also provides an important decision-making basis for garage managers. Managers can formulate more reasonable garage management strategies based on this prediction information. For example, if a certain type of obstacle is found to frequently occur in a certain area, managers can consider renovating the area or strengthening the management of the area to reduce the frequency of obstacles.

[0067] The neural network model takes spatial information and the current obstacle prediction sequence of each area as input, and can fully consider the layout structure of the garage and the obstacles that each area may face at the current moment. For areas with a high probability of obstacles, the model may reasonably arrange the patrol order based on factors such as their importance and their relationship with other areas. For example, areas close to the entrance and exit of the garage with a high probability of obstacles may be patrolled first, because the traffic conditions in these areas have a greater impact on the operating efficiency of the entire garage. Through this comprehensive analysis, the generated regional sorting results are more scientific and reasonable, ensuring that the robot can cover each area in a focused and orderly manner during the patrol process.

[0068] After obtaining the regional ranking results, the target patrol path is generated in combination with spatial information. This means that the model will consider spatial factors such as the actual distance between regions and channel connections to plan the shortest and smoothest path. For example, the model will give priority to connecting various regions with short distances, wide channels, and fewer turns, avoiding unnecessary detours and long-distance detours during the patrol process, thereby greatly improving patrol efficiency and reducing patrol time and energy consumption.

[0069] The garage environment is dynamically changing, and the vehicle flow and obstacles may be different in different time periods. Based on the current obstacle prediction sequence and the trained neural network model, the robot can perceive these changes in real time and adjust the area sorting and patrol path accordingly. For example, when a vehicle accident suddenly occurs in an area with fewer obstacles, causing the passage to be blocked, the neural network model can re-evaluate the situation of each area based on the new obstacle information, adjust the area sorting, and plan a new path for the robot to avoid the blocked area, ensuring that the patrol task is not affected and continues to be carried out stably.

[0070] Through scientific area sorting and path planning, the robot can fully cover all areas of the garage to ensure that no important places are missed. At the same time, for some areas that are prone to problems or have high safety risks, such as distribution rooms and storage areas for flammable and explosive items, the model will give higher priority in area sorting, allowing the robot to patrol these areas more frequently and carefully. This not only ensures the comprehensiveness of the patrol, but also highlights the key points, effectively improves the quality of the patrol, and promptly discovers and handles potential safety issues.

[0071] Reasonable patrol route planning helps the robot to timely discover and warn of various safety hazards and ensure the safety of people and facilities in the garage. For example, during the patrol process, the robot can detect abnormal conditions of vehicles (such as not turning off the engine, doors not closed, etc.), failures of facilities and equipment (such as damaged lighting, poor ventilation, etc.) and possible safety violations (such as illegal parking and occupying fire passages, etc.). By timely feedback of this information and taking corresponding measures, safety accidents can be avoided and the normal operation of the garage and the safety of life and property of people can be ensured.

[0072] As the robot patrols along the generated target patrol path, it will continuously accumulate new data, including actual obstacles encountered, actual patrol time in each area, etc. This data can be fed back to the neural network model for further training and optimization, so that its prediction and path planning capabilities for the garage environment are continuously improved. Over time, the model will adapt more and more accurately to the actual situation of the garage, providing the robot with better patrol path planning.

[0073] In this embodiment, by acquiring spatial information, details such as the layout of the garage, floor distribution, channel direction, and parking space location are covered. At the same time, historical walking data is collected, which records the walking paths of the robot in the past patrolling the garage, the time spent on each walk, and the various obstacles encountered. Based on the acquired spatial information, the historical walking paths are deeply analyzed to extract their spatial features, such as the length, direction, and distance from key positions of the paths. According to these features, the walking paths are reasonably segmented, each segment corresponds to a specific area, and many areas together constitute a regional set. For each area, the corresponding historical walking time and historical obstacles are accurately extracted from the historical walking data. Using data analysis and prediction algorithms, the current obstacle prediction sequence is generated to predict the probability and type of obstacles that may appear in the area at different times. The spatial information and the current obstacle prediction sequence of each area in the regional set are input into the trained neural network model. By learning a large amount of historical data, various factors can be comprehensively considered, the areas in the regional set can be sorted, and the optimal patrol order can be determined. Then, combined with the spatial information, a target patrol path covering all areas and efficiently avoiding obstacles is generated.

[0074] In this embodiment, the traditional robot patrol path planning method usually adopts a fixed route, which is difficult to adapt to the dynamic changes of the garage environment. The intelligent scheduling method based on historical data analysis can dynamically adjust the patrol path according to real-time obstacle prediction and spatial information. For example, when a sudden vehicle accident causes a blocked passage, the intelligent scheduling method can quickly plan a new path for the robot to ensure that the patrol task is not affected, while the traditional method may be in trouble. The traditional method may cause the robot to encounter frequent obstacles during the patrol process and waste a lot of time. This intelligent scheduling method analyzes and predicts historical data, plans the path to avoid obstacles in advance, optimizes the patrol order of the area, and greatly improves the patrol efficiency. For example, by reasonably arranging the patrol order, the robot can patrol areas prone to obstacles during the time period with less traffic in the garage, reducing waiting and detour time. The traditional method may not be able to fully cover all areas of the garage, or there may be omissions during the inspection process. The intelligent scheduling method ensures that the robot can patrol every corner of the garage comprehensively and carefully through scientific area division and path planning, and accurately discover various potential problems. For example, for some corner areas that are easily overlooked, the intelligent scheduling method will reasonably arrange the patrol time and order to ensure that these areas can also be fully inspected.

[0075] Preferably, extracting the spatial features of the walking path based on the spatial information and segmenting the walking path includes:

[0076] Using a convolutional neural network to process the data in the spatial information, and identify landmarks and environmental features of the walking path;

[0077] Based on the distribution of landmarks, environmental features, and obstacles in the walking path, an environmental complexity assessment mechanism is used to assess the environmental complexity of different parts of the walking path;

[0078] A dynamic segmentation strategy is adopted, and different segments in the walking path are segmented using different segmentation standards based on environment complexity evaluation data.

[0079] In areas with high environmental complexity, a smaller segment distance threshold is used to increase the number of path segments to more finely plan the robot's motion. In areas with low environmental complexity, a larger segment distance threshold is used to reduce the number of path segments.

[0080] In the unmanned patrol scenario of the garage, convolutional neural network (CNN) plays a core role in processing spatial information data and identifying landmarks and environmental features of the walking path.

[0081] In this embodiment, the spatial information of the garage contains rich data, such as the floor plan of the garage, the layout of each area, the location of facilities, etc. The convolutional neural network has a powerful image and data feature extraction capability, and can perform in-depth analysis of this spatial information. It automatically learns the characteristic patterns in the data through a series of operations such as convolutional layers and pooling layers. For the recognition of the environment of the walking path, the convolutional neural network can identify the key locations in the garage, and these landmarks play an important reference role in the path planning and positioning of the robot. For example, the robot can determine its position and direction based on these landmarks to ensure accurate patrol along the preset path. In terms of environmental feature recognition, the convolutional neural network can recognize the environmental features of the garage, such as the width of the passage, the arrangement of parking spaces, the location of the walls, etc. These features are crucial to understanding the spatial characteristics of the walking path. For example, areas with narrow passages may require the robot to drive more carefully to avoid collisions; areas with irregular parking spaces may increase the difficulty of patrolling. Through the recognition of these environmental features by the convolutional neural network, the robot can better adapt to the complex environment of the garage.

[0082] After identifying the landmarks and environmental features of the walking path, combined with the distribution of obstacles, the environmental complexity assessment mechanism is used to quantify the complexity of different parts of the walking path. Landmarks are of great significance in environmental complexity assessment. For example, the area near the entrance and exit has a relatively high environmental complexity due to frequent vehicle in and out and large personnel flow. In the corner area far away from the entrance and exit, there are fewer vehicles and personnel activities, and the environmental complexity is relatively low. Environmental characteristics also affect environmental complexity. In areas with narrow passages and many bends, the robot needs to be more careful when driving, the operation is more difficult, and the environmental complexity is higher; while in areas with wide passages and wide fields of view, the robot drives more smoothly and the environmental complexity is lower. The distribution of obstacles is a key factor in assessing environmental complexity. If an area often has obstacles such as illegal parking of vehicles and obstruction of construction equipment, the environmental complexity of the area will increase significantly. By comprehensively considering the distribution of landmarks, environmental features and obstacles, and adopting a scientific assessment mechanism, accurate environmental complexity assessment can be performed on different parts of the walking path. For example, a corresponding weight can be set for each factor, and a weighted calculation can be performed according to its importance in environmental complexity, and finally a quantitative environmental complexity index is obtained.

[0083] Based on the environmental complexity evaluation data, the dynamic segmentation strategy is used to segment the walking path, which can adapt to the complex environment of the garage more flexibly and reasonably. For parts with low environmental complexity, a larger segmentation standard is used. For example, in areas with wide passages and fewer obstacles, a longer path can be divided into one segment. This can reduce the number of segments and improve patrol efficiency. The robot can travel quickly in these areas and complete patrols over a larger range. For parts with high environmental complexity, a smaller segmentation standard is used. For example, in areas near entrances and exits, where vehicles are densely parked, or where obstacles often appear, the path is divided into smaller segments. This allows the robot to patrol these areas more carefully and discover potential problems in a timely manner. Each small segment can be used as an independent patrol unit, and the robot can perform more flexible operations in each small segment according to actual conditions. The dynamic segmentation strategy can automatically adjust the segmentation standard according to changes in environmental complexity, making the robot's patrol path more reasonable and efficient. In this way, the robot can better cope with the complex situations in different areas of the garage, improve patrol quality and efficiency, and ensure the safety and normal operation of the garage. In this embodiment, the above method provides strong support for the path planning of the robot in the unmanned patrol scenario of the garage, so that the robot can complete the patrol task more intelligently and efficiently.

[0084] Preferably, the method for intelligent scheduling of robot walking tasks based on historical data analysis further includes: utilizing spatial topological relationships to merge adjacent path segments having similar spatial characteristics and traffic properties into one area.

[0085] Specifically, when generating a set of regions corresponding to each path segment, a unique identifier is assigned to each region; the identifier is associated with information such as the region's spatial position, size, shape, and connection relationship with adjacent regions, which facilitates subsequent management and analysis of the region. When generating a set of regions corresponding to each path segment, the regions are classified and labeled; the labeled categories include but are not limited to open areas, narrow passages, dangerous areas, must-pass areas, etc., and the labeled information is determined based on spatial characteristics and historical obstacles. After generating a set of regions corresponding to each path segment, the region sets are organized hierarchically; the regions are divided into different levels based on factors such as the importance of the region and the distance from the target location, providing a clearer structure for subsequent path planning and task scheduling.

[0086] In this embodiment, the spatial topological relationship describes the position and connection relationship between spatial objects. In the garage environment, it is reflected in the characteristics of adjacency and connectivity between path segments. After determining the adjacent path segments based on the spatial topological relationship, the path segments with similar spatial characteristics and traffic attributes are further screened out and merged into one area. After merging multiple path segments into areas, the number of path units that need to be processed is reduced, and the patrol path planning process of the robot is simplified. When planning the path, the robot no longer needs to perform complex calculations and decisions on each individual path segment, but considers it in units of areas, which greatly improves the efficiency of path planning. For areas with similar characteristics and attributes, a unified patrol strategy can be adopted. By merging path segments into areas, it can better adapt to changes in the garage environment. When the layout or traffic rules of the garage change, only the relevant areas need to be adjusted, without the need to make large-scale modifications to the entire path planning system. Using spatial topological relationships to merge adjacent path segments with similar spatial characteristics and traffic attributes into one area is an effective method for optimizing garage space management and robot patrol path planning, which can improve the efficiency, adaptability and manageability of the system.

[0087] Preferably, when extracting the historical walking time corresponding to the area from the historical walking data, the precise timestamp of each time the robot enters and leaves the area is extracted, and the historical walking time of each time in the area is obtained by calculating the difference between the two;

[0088] When extracting historical obstacle conditions, the obstacle types are classified and recorded; wherein the obstacle types include fixed obstacles, mobile obstacles and temporary obstacles, and different types of obstacles are marked separately in the historical data.

[0089] In this embodiment, during the actual operation of the unmanned patrol in the garage, each action of the robot is accurately recorded. When extracting the historical walking time corresponding to a specific area from the historical walking data, the key is to obtain the accurate timestamp of each time the robot enters and leaves the area. These timestamps record the specific moment when the robot interacts with a specific area in the garage space, and the time record is accurate to seconds or even milliseconds, providing an accurate data basis for subsequent time calculations. For example, in a patrol mission on a certain day, the robot enters parking area A at 9:15:20 in the morning and leaves the area at 9:23:10. These two time points are the entry and exit timestamps of this patrol in the area. By calculating the difference in the timestamps of each entry and exit area, the historical walking time of the robot in the area can be accurately obtained. This calculation method is simple and direct, but it provides key data for analyzing the patrol efficiency of robots in different areas. For example, by statistically analyzing the historical walking time of the robot in each parking area over a period of time, it can be understood which areas are due to complex layouts, dense parking of vehicles, etc., causing the robot to spend more time patrolling; which areas are relatively smooth and the robot can quickly complete the patrol. This helps to evaluate the difficulty and efficiency of patrolling different areas, and provides a strong basis for subsequent route planning and task scheduling.

[0090] When extracting historical obstacles, it is very important to carefully classify and record the obstacle types. In the garage environment, there are various types of obstacles, which are divided into three categories: fixed obstacles, mobile obstacles, and temporary obstacles, and marked separately in the historical data.

[0091] Fixed obstacles: These obstacles are relatively fixed, such as columns, fire equipment boxes, fixed isolation piers, etc. in the garage. They are long-term elements in the spatial layout of the garage and have a lasting impact on the robot's patrol path planning and actions. For example, columns will limit the robot's driving route, and the robot needs to always avoid these fixed obstacles when planning the path.

[0092] Mobile obstacles: Mobile obstacles refer to objects that move in the garage, such as vehicles, pedestrians, and mobile cleaning equipment. The position and movement of these obstacles are dynamically changing, which increases the uncertainty of the robot patrol process. For example, the driving and parking positions of vehicles in the garage are not fixed, and they may suddenly appear on the robot's patrol path, requiring the robot to respond and adjust in time.

[0093] Temporary obstacles: Temporary obstacles refer to obstacles that appear due to special circumstances or temporary events, such as warning signs set up in the ongoing equipment maintenance area, temporarily stacked goods, etc. The appearance of these obstacles is temporary and sudden, and the robot needs to be able to perceive and respond in time.

[0094] Different types of obstacles are marked in historical data, so that robots and managers can clearly understand the various obstacles that have appeared in each area. By analyzing these marked data, we can find the regularity of the appearance of different types of obstacles in different areas and time periods. Understanding these regularities helps robots prepare for response in advance and optimize patrol strategies, such as increasing patrol frequency during periods when mobile obstacles are likely to appear, or planning detours in advance in areas where temporary obstacles may appear.

[0095] By accurately extracting historical walking time and recording historical obstacles in detail, the robot can have a more comprehensive understanding of the characteristics and potential problems of each area. When planning the path, it can fully consider these factors, combine the traffic efficiency and obstacle risks of different areas, and formulate a more reasonable and efficient patrol path. For example, avoid path sections where mobile obstacles often appear, or plan safer driving routes around fixed obstacles. In-depth analysis of historical data enables the robot to predict the types and situations of obstacles that may be encountered in advance, thereby improving its own response capabilities. During the patrol process, when encountering similar obstacles, the robot can respond quickly and take corresponding measures, such as slowing down, pausing and waiting, adjusting the path, etc., to ensure the smooth progress of the patrol mission.

[0096] Preferably, when generating the current obstacle prediction sequence corresponding to the area, the historical travel time and historical obstacle conditions are modeled based on the state space method, and the Kalman filter algorithm is used for state estimation and prediction;

[0097] Using the cross-validation method, historical data is divided into training set and test set. The training set data is used to generate a prediction sequence, which is compared with the test set data. The accuracy of the prediction is evaluated by calculating the mean square error and mean absolute error indicators. If the prediction result does not meet the preset accuracy requirements, the parameters of the prediction model are adjusted or the prediction model is replaced.

[0098] In this embodiment, when generating the current obstacle prediction sequence, the impact of different operation modes of the robot on the obstacle situation is considered; different operation modes include fast patrol mode, fine inspection mode, etc., and the historical walking data under different operation modes are analyzed and predicted respectively to generate an obstacle prediction sequence that is more in line with the actual situation. The area is divided into sub-areas; according to the spatial layout, functional characteristics and other factors of the area, the area is divided into several sub-areas, and the historical walking time and historical obstacle situation of each sub-area are extracted respectively to generate respective current obstacle prediction sequences, so as to more accurately reflect the obstacle changes in different parts of the area. The historical walking data is updated regularly, and the current obstacle prediction sequence is regenerated; as time goes by and the environment changes, new historical data is continuously generated. By regularly updating the data and re-predicting, the obstacle prediction sequence can timely reflect the latest situation in the area, and improve the timeliness and reliability of the prediction.

[0099] In this embodiment, a state space-based method is used to model the historical travel time and historical obstacle conditions. This method describes the state and observation data using state equations and observation equations, and estimates the state and makes predictions through a Kalman filter algorithm.

[0100] See also Figure 2 As shown, Figure 2 The robot state estimation and obstacle prediction process based on Kalman filtering is described as follows:

[0101] A (initial state estimation): At the beginning of the process (k = 0), the robot's state estimation needs to be initialized

[0102] x_hat(0|0). This includes the robot's initial position in the garage (e.g., based on the garage map and the starting point), the initial velocity (usually 0), and estimates of the initial obstacle conditions in each area (e.g., based on historical data or default values). P(0|0) represents the uncertainty of the initial state estimate, which is usually set to a diagonal matrix, with the elements on the diagonal representing the initial variance of each state variable.

[0103] B (state prediction): At time k, based on the state estimate x_hat(k-1|k-1) at time k-1 and the robot's motion model F (e.g., a uniform motion model or a more complex motion model) and the control input u(k) (e.g., robot control instructions, such as speed, steering angle, etc.), predict the robot's state x_hat(k|k-1) at time k, including position, speed, and the probability of obstacles in the area. The prediction formula is as follows:

[0104] x_hat(k|k-1)=F*x_hat(k-1|k-1)+B*u(k)

[0105] Among them, the motion model F and control input u(k) are associated with the robot's path planning and walking state, and can be selected and adjusted according to the actual situation. If the probability of an obstacle in area i being predicted at the previous moment is high, then at the current moment, the predicted value of the probability of an obstacle in the area will also be affected. B is the control input matrix, which describes the impact of the control input u(k) on the state. It needs to be selected and adjusted according to the actual control input used. If there is no control input, it can also be set to 0.

[0106] C (forecast error covariance): The forecast uses the formula:

[0107] P(k|k-1)=F*P(k-1|k-1)*F T +Q

[0108] Where P(k|k-1) represents the uncertainty of the predicted state, which takes into account the uncertainty of the state estimate at the previous moment P(k-1|k-1), the state transfer matrix F (related to the motion model), and the process noise Q. Q represents the inaccuracy of the motion model and control input, for example, the deviation of the actual motion trajectory of the robot from the expected trajectory, which may be affected by environmental factors or the robot's own performance.

[0109] D (observation data): At time k, the robot obtains observation data z(k) through sensors (e.g., cameras, lidar), including the robot's actual position, speed, and the actual number or type of obstacles in each area. These observation data correspond to the spatial information and historical walking data obtained in step S101. The observation matrix H maps the robot's state to the observation space. For example, the number of obstacles in a certain area is obtained through camera image analysis. The formula used for mapping is as follows:

[0110] z(k)=H*x(k)+v(k)

[0111] Where v(k) is the sensor observation noise at time k, which obeys a Gaussian distribution with a mean of 0 and a covariance matrix of R.

[0112] E (calculate Kalman gain): Kalman gain K(k) is calculated based on the prediction error covariance P(k|k-1), the observation matrix H and the observation noise covariance R. The formula is as follows:

[0113] K(k)=P(k|k-1)*H T *(H*P(k|k-1)*H T +R) -1

[0114] The Kalman gain determines the weight of the observed value and the predicted value when the state is updated. R represents the uncertainty of the sensor observation data, for example, the error rate of the camera image identifying obstacles.

[0115] F (state update): Update the robot's state estimate x_hat(k|k) at time k based on the predicted state x_hat(k|k-1), the observed value z(k) and the Kalman gain K(k). The formula used is as follows:

[0116] x_hat(k|k)=x_hat(k|k-1)+K(k)*(z(k)-H*x_hat(k|k-1))

[0117] A more accurate estimate of the robot's position, speed, and obstacle probability in each area is obtained through state update calculation for subsequent path planning (as described in S104).

[0118] G (updated error covariance): P(k|k) is the updated state estimate covariance, which indicates the uncertainty of the updated state estimate. The calculation formula is as follows:

[0119] P(k|k)=(IK(k)*H)*P(k|k-1)

[0120] Among them, I is the unit matrix.

[0121] Loop iteration: The updated state estimate x_hat(k|k) and covariance P(k|k) will be used as the prior information for the next moment (k+1), and the prediction and update steps will be executed cyclically to continuously optimize the robot's state estimate and prediction of obstacle conditions.

[0122] Through the state space model and Kalman filter algorithm, the robot state is estimated and the future obstacle situation is predicted. The current obstacle prediction sequence of each area will be used as the input of the neural network model to generate the target patrol path.

[0123] Use cross-validation to divide historical data into training sets and test sets. Common division ratios can be 70%-30% or 80%-20%, etc., that is, most of the data is used as the training set for model training, and a small part of the data is used as the test set to evaluate model performance. For example, divide the historical data of the past year into 80%-20%, with the first 80% of the data as the training set and the last 20% of the data as the test set.

[0124] Generate a prediction sequence using the training set data, and then compare the prediction sequence with the test set data. Evaluate the accuracy of the prediction by calculating indicators such as mean square error (MSE) and mean absolute error (MAE). Calculate the average of the squares of the difference between the predicted value and the actual value. MSE can reflect the degree of deviation between the predicted value and the actual value. The smaller the value, the closer the predicted value is to the actual value, and the higher the prediction accuracy of the model. For example, if the MSE value is 0.5, it means that the average square error between the predicted value and the actual value is 0.5. Calculate the average of the absolute values ​​of the difference between the predicted value and the actual value. MAE more intuitively reflects the average error between the predicted value and the actual value. Similarly, the smaller the value, the better the prediction effect of the model. For example, if the MAE value is 0.3, it means that the average absolute error between the predicted value and the actual value is 0.3. Preset an accuracy requirement, such as MSE does not exceed a certain threshold (such as 0.8) and MAE does not exceed another threshold (such as 0.5). If the prediction result does not meet the preset accuracy requirement, it means that the current model may not accurately capture the rules in the historical data and the model needs to be adjusted.

[0125] By using the state space model, Kalman filter algorithm, and cross-validation to perform prediction evaluation and optimization, we can fully tap into the information in historical data and improve the accuracy of the current obstacle prediction sequence. This allows the robot to understand the obstacles that may be encountered in advance during the patrol process, better plan the path, and improve patrol efficiency and safety. The prediction model and parameters can be flexibly adjusted according to the characteristics and historical data of different areas to adapt to changes in the garage environment. This process of continuous evaluation and optimization allows the prediction model to be continuously improved as new historical data accumulates. Over time, the model will reflect the actual situation of the garage more and more accurately, providing more reliable support for unmanned patrols in the garage.

[0126] Preferably, the trained neural network model is a recurrent neural network model;

[0127] Based on the recurrent neural network, the current obstacle prediction sequence of each area is processed to capture the spatial dependency of each area and the temporal dependency in the current obstacle prediction sequence;

[0128] Based on the spatial dependency of each region and the temporal dependency in the current obstacle prediction sequence, a sorting result of the regions in the region set is generated in combination with a sorting basis;

[0129] The sorting criteria include the obstacle risk level of the area, the estimated patrol time, and the distance from the starting point or important target point; the neural network model outputs the priority ranking of each area through comprehensive learning and analysis of the sorting criteria.

[0130] Specifically, the recurrent neural network (RNN) is a neural network model suitable for processing sequence data, and its unique structure can capture the temporal dependencies in the data. In the unmanned patrol scenario of the garage, the current obstacle prediction sequence of each area is a sequence data with time characteristics, so the RNN model is very suitable for processing this data. During the training process of the neural network model, a variety of loss functions are used for optimization; in addition to the commonly used mean square error (MSE) loss function, the cross entropy loss function or a custom loss function is also introduced to comprehensively consider the difference between the regional sorting results and the actual situation, the importance weights of different areas and other factors, so as to improve the training effect of the model.

[0131] Spatial information includes environmental data, landmark location information, and access attribute information of different areas. Access attribute information is used to indicate whether an area is accessible, adjacent accessible areas, etc. Access attribute information serves as an important constraint when generating area sorting results and target patrol paths.

[0132] Before inputting the current obstacle prediction sequence of each area in the area set into the trained neural network model, the prediction sequence is preprocessed; the preprocessing includes mapping different obstacle prediction values ​​to corresponding intervals and converting the prediction sequence into a format suitable for neural network input.

[0133] In this embodiment, the various areas in the garage do not exist in isolation, but are spatially interrelated. For example, adjacent areas may affect each other due to factors such as the flow of vehicles and the activities of personnel. The recurrent neural network can capture this spatial dependency by processing the current obstacle prediction sequence of each area. The model can learn the positional relationship between different areas and their potential connection in the occurrence of obstacles. For example, the occurrence of obstacles in a certain area may cause changes in the traffic conditions in adjacent areas, thereby affecting the patrol order of the robot in these areas. The current obstacle prediction sequence itself changes with time, and there is a certain correlation between the obstacle conditions at different times. The RNN model can effectively capture this time dependency and understand the evolution of the obstacle situation over time by learning from historical data. For example, the probability of obstacle occurrence in different areas of the garage will be different at different time periods of the day. The RNN model can learn these laws, thereby more accurately predicting the obstacle conditions in each area at future moments, and incorporating this time-dependent information into the decision-making of regional sorting.

[0134] Based on the historical obstacle conditions and the current obstacle prediction sequence, an obstacle risk level is determined for each area. Areas with high risk levels may have more uncertainty and potential dangers and need to be patrolled first. Considering the size of different areas, the complexity of the layout, and the obstacles that may be encountered, the estimated patrol time that the robot needs to spend in each area is calculated. Areas with longer patrol times are expected to need to be arranged more reasonably in the patrol route to ensure that the entire patrol mission can be completed efficiently. The starting point is usually the starting point of the robot, and the important target point may be a key location in the garage. The distance between the area and these points will affect the planning of the patrol path. Areas with closer distances may be considered for patrol first to reduce the robot's movement time on the path. The recurrent neural network model can weigh the relationship between various factors and assign a reasonable priority to each area through comprehensive learning and analysis of these sorting criteria. The model automatically learns how to sort the areas in different situations based on spatial dependencies, temporal dependencies, and sorting criteria to achieve the best patrol effect. For example, in some cases, although the obstacle risk level of an area is high, due to the distance from the starting point and the estimated patrol time, the model may first arrange patrols to areas with a closer distance and relatively lower obstacle risk to ensure that as many important areas as possible are covered within a limited time.

[0135] In this embodiment, after learning and analyzing various information, the recurrent neural network model finally outputs the priority ranking of each area. This ranking result provides an important basis for the patrol path planning of the robot, so that the robot can patrol each area in a reasonable order. By considering the spatial and temporal dependencies and multiple ranking bases, the ranking results generated by the model can enable the robot to avoid obstacles more efficiently during the patrol process, arrange time reasonably, reduce unnecessary movement and waiting time, and thus improve the overall patrol efficiency. Prioritizing patrolling areas with high obstacle risk levels helps the robot to promptly discover and deal with potential safety hazards and ensure the safety of people and vehicles in the garage. The model can learn the dynamic changes of the garage environment, and continuously optimize the regional ranking strategy as time goes by and data accumulates, so as to better adapt to various complex situations in the garage, such as changes in traffic flow in different time periods, and increased obstacles caused by special events. In this embodiment, the regional ranking method based on the recurrent neural network model has significant advantages in the unmanned patrol scene of the garage, and can provide strong support for the efficient and safe patrol of the robot.

[0136] Preferably, generating a target patrol path in combination with spatial information comprises: based on the sorting result of the areas in the area set, in combination with the spatial information, using Dijkstra algorithm to perform path planning on the sorted areas to find the target patrol path;

[0137] After the target patrol path is generated, the actual movement process of the robot is simulated to check whether the impassable area and the repeated pass area on the target patrol path are within the preset threshold range. If not, the path is adjusted or replanned.

[0138] Specifically, the energy consumption factor of the robot is taken into consideration in the process of generating the target patrol path. By modeling and analyzing the energy consumption of the robot in different areas and different path segments, energy consumption is used as an optimization goal of path planning, so that the generated target patrol path can minimize the energy consumption of the robot while meeting the task requirements.

[0139] After the target patrol path is generated, a detailed execution plan of the path is also generated; the execution plan includes specific operation instructions such as the robot's driving speed, turning angle, and dwell time in each path segment to ensure that the robot can perform patrol tasks accurately and efficiently.

[0140] When the robot is performing patrol tasks, if there is a large difference between the environmental information obtained in real time and the predicted situation, such as the emergence of new obstacles or changes in the area traffic properties, the latest spatial information and the current obstacle prediction sequence are re-input, the area sorting results are regenerated through the neural network model, and the target patrol path is adjusted in real time in combination with the spatial information.

[0141] In this embodiment, the sorting results of the areas in the area set determine the order in which the robot patrols each area. This sorting takes into account factors such as the obstacle risk level of the area, the expected patrol time, and the distance from the starting point or important target point, and provides a basic framework for patrol path planning. Spatial information includes the layout structure of the garage, such as the location, length, width of the passage, the distribution of parking spaces, and the connection relationship between the various areas. This information is crucial for planning a feasible and efficient patrol path. The Dijkstra algorithm is a classic graph search algorithm used to find the shortest path from one node to all other nodes in a weighted directed graph. In a garage environment, each area can be regarded as a node in the graph, the passages between the areas can be regarded as edges, and the length of the passage or the difficulty of passage can be used as the weight of the edge. Based on the regional sorting results, the Dijkstra algorithm is used to plan the shortest or optimal path for the robot from the starting point to the sorted areas in this graph constructed based on the garage space information, that is, the target patrol path. In this way, it can be ensured that the robot reduces the walking distance as much as possible during the patrol process and improves the patrol efficiency.

[0142] After the target patrol path is generated, the rationality of the path is checked by simulating the actual movement process of the robot. This simulation process takes into account the robot's motion characteristics, possible obstacles, and the actual environmental limitations of the garage. There may be some temporary impassable areas in the garage due to construction, equipment failure, or other reasons. During the simulation process, it is necessary to check whether there are such impassable areas on the target patrol path, and whether the number and influence range of these areas are within the preset threshold range. If there are too many impassable areas or the influence range is too large, the robot may not be able to complete the patrol task smoothly. Repeatedly passing through certain areas may waste the robot's time and energy and reduce patrol efficiency. Therefore, it is also necessary to check the repeated passing areas on the target patrol path to ensure that they are within the preset threshold range. If the impassable area and the repeated passing area are not within the preset threshold range, the path needs to be adjusted or replanned. When adjusting the path, you can try to bypass the impassable area, or optimize the path to reduce the repeated passing area. If the adjustment cannot meet the requirements, the path needs to be replanned. When replanning the path, you may need to reconsider the order of the areas, or combine more constraints, such as the robot's power limit, task time limit, etc., to generate a target patrol path that better meets actual needs.

[0143] By combining the Dijkstra algorithm with spatial information to generate the target patrol path, it can ensure that the robot can patrol along the optimal path in theory. Simulation inspection and adjustment of the path further improves the feasibility and reliability of the path, avoiding the robot from encountering problems such as impassability or low efficiency during the actual patrol process, thereby ensuring that the patrol mission can be successfully completed.

[0144] Reasonable path planning and adjustment can reduce the robot's walking distance and repeated passing areas, reduce energy consumption, and extend the robot's service life. At the same time, it also improves the robot's work efficiency, enabling it to complete more patrol tasks within a limited time and better meet the garage's safety management needs. The garage environment may change at any time, such as the emergence of new obstacles, temporary closure of passages, etc. Through the path inspection and adjustment mechanism, the impact of these changes on the patrol path can be discovered in time, and corresponding measures can be taken to adjust, so that the robot can quickly adapt to environmental changes and maintain good patrol effects. The process of generating a target patrol path in combination with spatial information and inspecting and adjusting it is an important link to ensure the efficient and reliable operation of the garage's unmanned patrol system, and can provide strong support for garage safety patrols.

[0145] Example 2

[0146] This embodiment provides a robot walking task intelligent scheduling system based on historical data analysis, which is described by applying the robot walking task intelligent scheduling method based on historical data analysis provided in the above embodiment 1. Figure 3 As shown, a robot walking task intelligent scheduling system based on historical data analysis provided by one embodiment of the present application may include the following multiple modules.

[0147] An acquisition module is used to acquire spatial information and historical walking data, wherein the historical walking data includes walking path, walking time and obstacle conditions;

[0148] A partitioning module, used to extract the spatial features of the walking path based on the spatial information, segment the walking path, and generate a set of regions corresponding to each path segment;

[0149] A sorting module is used to extract the historical walking time and historical obstacle conditions corresponding to the same area from the historical walking data, and generate a current obstacle prediction sequence corresponding to the area;

[0150] The generation module is used to generate the sorting results of the areas in the area set according to the spatial information and the current obstacle prediction sequence of each area in the area set through the trained neural network model, and generate the target patrol path in combination with the spatial information.

[0151] The robot walking task intelligent scheduling system based on historical data analysis provided in the embodiment of the present application can be applied to the robot walking task intelligent scheduling method based on historical data analysis provided in the above-mentioned embodiment 1. For relevant details, refer to the above-mentioned method embodiment. Its implementation principle and technical effects are similar and will not be repeated here.

[0152] Example 3

[0153] See also Figure 4 As shown, one embodiment of the present application also provides a computer device, which can be a computer device such as a desktop computer, a laptop computer, a handheld computer, and a cloud server. The computer device can include, but is not limited to, a processor and a memory. The processor and the memory are connected via a bus or other means.

[0154] As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as program instructions / modules corresponding to the methods in the above-mentioned embodiments of the present application. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions and modules stored in the memory, that is, implementing the methods in the above-mentioned method embodiments.

[0155] One embodiment of the present application further provides a computer-readable storage medium, the computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the method in the above method implementation is implemented. Those skilled in the art can understand that the implementation of all or part of the process in the above method implementation of the present application can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the process of the implementation of the above methods.

Claims

1. A method for intelligent scheduling of robot walking tasks based on historical data analysis, characterized in that: include: Obtain spatial information and historical walking data, where the historical walking data includes walking path, walking time and obstacle conditions; Extracting spatial features of the walking path based on the spatial information, segmenting the walking path, and generating a set of regions corresponding to each path segment; For the same area, extract the historical walking time and historical obstacle conditions corresponding to the area from the historical walking data, and generate a current obstacle prediction sequence corresponding to the area; According to the spatial information and the current obstacle prediction sequence of each area in the area set, the sorting results of the areas in the area set are generated through the trained neural network model, and the target patrol path is generated in combination with the spatial information.

2. The method for intelligent scheduling of robot walking tasks based on historical data analysis according to claim 1 is characterized in that: The step of extracting the spatial features of the walking path based on the spatial information and segmenting the walking path includes: Using a convolutional neural network to process the data in the spatial information, and identify landmarks and environmental features of the walking path; Based on the distribution of landmarks, environmental features, and obstacles in the walking path, an environmental complexity assessment mechanism is used to assess the environmental complexity of different parts of the walking path; A dynamic segmentation strategy is adopted, and different segments in the walking path are segmented using different segmentation standards based on environment complexity evaluation data.

3. The method for intelligently scheduling robot walking tasks based on historical data analysis according to claim 2 is characterized in that: Also includes: Adjacent path segments with similar spatial characteristics and traffic attributes are merged into one area using spatial topological relationships.

4. The method for intelligently scheduling robot walking tasks based on historical data analysis according to claim 3 is characterized in that: When extracting the historical walking time corresponding to the area from the historical walking data, the precise timestamp of each time the robot enters and leaves the area is extracted, and the historical walking time of each time in the area is obtained by calculating the difference between the two; When extracting historical obstacle conditions, the obstacle types are classified and recorded; wherein the obstacle types include fixed obstacles, mobile obstacles and temporary obstacles, and different types of obstacles are marked separately in the historical data.

5. The method for intelligent scheduling of robot walking tasks based on historical data analysis according to claim 4 is characterized in that: When generating the current obstacle prediction sequence corresponding to the area, a state space-based method is used to model the historical travel time and historical obstacle conditions, and a Kalman filter algorithm is used for state estimation and prediction; The robot's state prediction formula is: x_hat(k|k-1)=F*x_hat(k-1|k-1)+B*u(k); Among them, at time k, the robot's state x_hat(k|k-1) at time k is predicted based on the state estimate x_hat(k-1|k-1) at time k-1, the robot's motion model F, and the control input u(k). The motion model F and the control input u(k) are associated with the robot's path planning and walking state. B is the control input matrix, which describes the impact of the control input u(k) on the state; x_hat(k|k-1) represents the predicted state of the robot at time k, including the position, speed, and the prediction of the probability of obstacles in the area; Using the cross-validation method, historical data is divided into training set and test set. The training set data is used to estimate model parameters and compared with the test set data. The accuracy of the prediction is evaluated by calculating the mean square error and mean absolute error indicators. If the prediction result does not meet the preset accuracy requirements, the model parameters are adjusted or the prediction model is replaced.

6. The method for intelligent scheduling of robot walking tasks based on historical data analysis according to claim 5 is characterized in that: The trained neural network model is a recurrent neural network model; Based on the recurrent neural network, the current obstacle prediction sequence of each area is processed to capture the spatial dependency of each area and the temporal dependency in the current obstacle prediction sequence; Based on the spatial dependency of each region and the temporal dependency in the current obstacle prediction sequence, a sorting result of the regions in the region set is generated in combination with a sorting basis; The sorting criteria include the obstacle risk level of the area, the estimated patrol time, and the distance from the starting point or important target point; the neural network model outputs the priority ranking of each area through comprehensive learning and analysis of the sorting criteria.

7. The method for intelligently scheduling robot walking tasks based on historical data analysis according to claim 6 is characterized in that: The generating of the target patrol path by combining the spatial information comprises: based on the sorting result of the areas in the area set, combining the spatial information, using the Dijkstra algorithm to perform path planning on the sorted areas, and finding the target patrol path; After the target patrol path is generated, the actual movement process of the robot is simulated to check whether the impassable area and the repeated pass area on the target patrol path are within the preset threshold range. If not, the path is adjusted or replanned.

8. A robot walking task intelligent scheduling system based on historical data analysis, characterized in that: include: An acquisition module is used to acquire spatial information and historical walking data, wherein the historical walking data includes walking path, walking time and obstacle conditions; A partitioning module, used to extract the spatial features of the walking path based on the spatial information, segment the walking path, and generate a set of regions corresponding to each path segment; A sorting module is used to extract the historical walking time and historical obstacle conditions corresponding to the same area from the historical walking data, and generate a current obstacle prediction sequence corresponding to the area; The generation module is used to generate the sorting results of the areas in the area set according to the spatial information and the current obstacle prediction sequence of each area in the area set through the trained neural network model, and generate the target patrol path in combination with the spatial information.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor implements the intelligent scheduling method for robot walking tasks based on historical data analysis as described in any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by the processor, the method for intelligent scheduling of robot walking tasks based on historical data analysis as described in any one of claims 1 to 7 is implemented.

Images