A robot inspection scheduling method based on gait energy efficiency matrix

Abstract

The application discloses a kind of robot inspection scheduling methods based on gait energy efficiency matrix, obtain the state data of multiple robots and inspection task information, determine task urgency based on task type and establish task-gait energy efficiency mapping relationship in combination with terrain resistance coefficient;Calculate the spatial coverage overlap degree of multiple robots and generate path overlap penalty term;With the goal of minimizing cluster total energy consumption, build and solve global energy consumption optimization model, automatically eliminate and re-solve infeasible solution caused by constraint conflict, obtain task allocation and gait strategy;Before issuing, simulation-reality migration is carried out and noise disturbance is added, and the instruction is issued through the native image compiled in advance;Real-time acquisition of telemetry data is carried out during execution and the terrain resistance coefficient is updated.The application cooperates native image fast rescheduling by simultaneously including task urgency, terrain resistance and spatial coverage overlap degree into global energy consumption optimization, to maximize the endurance time of multiple robot cluster under the condition of ensuring the full coverage of inspection task.

Description

Technical Field

[0001] This invention relates to the field of embodied intelligence and multi-robot collaborative control technology, specifically to a robot inspection scheduling method based on gait energy efficiency matrix. Background Technology

[0002] With the rapid development of embodied intelligence technology, quadruped robots, with their advantages of strong terrain adaptability and flexible movement, are widely used in inspection operations in scenarios such as industrial parks, smart communities, and substations. To improve inspection coverage and task execution efficiency, multi-robot collaborative inspection has become the mainstream technology in the industry.

[0003] Existing inspection scheduling methods are mostly based on the "fixed route - fixed task" model, which means that the robot is assigned to the target location and executed sequentially according to the preset inspection route and task list. Gait parameters (such as walking speed and step frequency) are kept constant or adjusted only based on simple positioning feedback. Some solutions optimize the robot's endurance by introducing a simple mobile energy consumption model.

[0004] However, in campus scenarios with varying levels of urgency, complex terrain, and multi-machine concurrent collaboration, this model reveals the following shortcomings: (1) The energy consumption of a single robot is disconnected from the task. The robot moves with a fixed gait and uniform speed, and does not dynamically match the movement mode according to the urgency of the inspection task (such as emergency fire confirmation and daily instrument readings). This results in slow response to high-urgency tasks or energy waste in low-urgency tasks, and the overall energy utilization efficiency is low.

[0005] (2) Multi-robot task allocation ignores terrain and energy consumption factors. Existing allocation is mostly based on the principle of the shortest physical distance, without taking into account the differences in movement energy consumption caused by terrain resistance (such as grass, slopes, gravel roads) into the allocation decision, and also fails to take into account the current remaining power of each robot, causing individual robots to run out of power too early, affecting the overall endurance of the cluster.

[0006] (3) Redundancy in multi-machine vision coverage has not been effectively eliminated. During the inspection process of multiple robots, the video monitoring range often overlaps. Existing methods do not perform dynamic deduplication based on spatial coverage relationships, resulting in multiple robots accumulating unnecessary energy consumption in the same area or path, causing collective energy waste.

[0007] (4) Emergency dispatch response delay. When a sudden alarm occurs or the environment changes and rapid replanning is required, the startup and calculation of traditional central dispatch services take a long time, making it difficult to complete the redistribution of tasks and gaits across the entire network within milliseconds, which affects the timeliness and security in emergency scenarios.

[0008] Based on the above, the technical problem that urgently needs to be solved by those skilled in the art is: how to provide a scheduling method that can incorporate task urgency, terrain access resistance, multi-machine spatial coverage overlap and single-machine remaining power as collaborative variables into a unified framework for global energy consumption optimization, and achieve deterministic and rapid rescheduling when dealing with sudden and high-concurrency tasks, so as to maximize the overall endurance of the multi-robot cluster while ensuring the task completion rate. Summary of the Invention

[0009] This invention aims to solve the technical problems in existing quadruped robot inspection scheduling methods, such as the disconnect between single-machine energy consumption and task urgency, the lack of coordinated consideration of terrain resistance and remaining power in multi-machine task allocation, energy waste caused by overlapping vision coverage of multiple machines, and uncontrollable scheduling response delays in emergency scenarios. It provides an inspection scheduling method that can incorporate task urgency, terrain resistance, multi-machine spatial coverage overlap, and single-machine remaining power into a unified global energy consumption optimization framework, and achieve deterministic and rapid rescheduling under high-concurrency emergency tasks. This maximizes the overall endurance of a multi-robot cluster while ensuring full coverage of inspection tasks.

[0010] To achieve the above objectives, this invention provides a robot inspection scheduling method based on a gait energy efficiency matrix, comprising: Acquire status data of multiple robots and inspection task information to be executed. The status data includes at least the remaining battery power and real-time location of each robot. The urgency of the task is determined based on the task type in the inspection task information, and the terrain resistance coefficient is determined in combination with the terrain information of the inspection area to establish a task-gait energy efficiency mapping relationship. Calculate the spatial coverage information of the inspection area by the multiple robots, determine the coverage overlap between each robot, and generate a path overlap penalty term based on the coverage overlap. With the goal of minimizing the total energy consumption of the cluster, the task-gait energy efficiency mapping relationship and the path overlap penalty term are used as variables to construct and solve a global energy consumption optimization model to obtain the task allocation scheme and corresponding gait strategy of each robot. During the solution process, preset constraints are checked. For infeasible solutions caused by constraint conflicts, allocation schemes that violate the constraints are automatically eliminated and the solution is resolved. Before the gait strategy is sent out for execution, the gait strategy is subjected to simulation-reality transfer processing. Preset noise disturbances are added to the simulation environment to generate a robust path. The scheduling logic is loaded based on the pre-compiled native image and the scheduling instructions are sent to each robot for execution. During the inspection process, telemetry data from each robot is collected in real time, and global model parameters, including the terrain resistance coefficient, are dynamically updated based on the telemetry data.

[0011] Furthermore, in the task-gait energy efficiency mapping relationship, transportation energy consumption... Calculate using the following formula: ; in, The rated power under the corresponding gait command, For the distance traveled, For movement speed, This is the terrain resistance coefficient.

[0012] Furthermore, the construction and solution of the global energy consumption optimization model includes: establishing the objective function of the global energy consumption optimization model as minimizing the total power consumption of the robot cluster during the inspection cycle. : ; in, For the total number of robots, This represents the total number of inspection tasks. Let be a binary decision variable, representing the th Is the robot assigned the first One task, Based on travel distance ,gait and terrain resistance coefficient Determined power consumption during transportation To carry out the mission The inherent power consumption at that time, This is a spatial path overlap penalty term. This is the penalty coefficient.

[0013] Furthermore, calculating the spatial coverage information of the inspection area by the multiple robots includes: obtaining the current visual monitoring range of each robot; determining the overlapping area between the visual monitoring ranges of different robots using a spatial relationship function, as the coverage overlap; wherein, the calculation of the coverage overlap includes projecting the video field of view of each robot onto a spatiotemporal grid map, and calculating the overlapping area between each projection area using a topological intersection function; the objective function of the global energy consumption optimization model includes a penalty term positively correlated with the coverage overlap.

[0014] Furthermore, the preset constraints include at least a power balance constraint, a timeliness constraint, and a coverage constraint; wherein, the power balance constraint is used to ensure that the remaining power of each robot after performing the assigned task is not lower than a preset safety threshold, the timeliness constraint is used to ensure that the estimated arrival time of each robot to the corresponding task point with the assigned gait strategy is not later than the deadline of the corresponding task, and the coverage constraint is used to ensure that the union of the visual monitoring range of each robot covers the preset target inspection area.

[0015] Furthermore, in the global energy consumption optimization model, the inherent power consumption during task execution is calculated in stages according to the load devices involved in the inspection task; the inspection task information also includes recognition accuracy requirements. If the recognition accuracy requirements are higher than a preset threshold, the maximum walking speed in the gait strategy is limited, and a high-power sensor load is matched.

[0016] Furthermore, the dynamic updating of global model parameters, including the terrain drag coefficient, based on the telemetry data includes: using foot sensors or inertial measurement units mounted on each robot to sense the physical feedback of the current terrain, and correcting the terrain drag coefficient of the corresponding road segment in real time based on the physical feedback.

[0017] Furthermore, the simulation-reality migration process includes: applying a preset noise disturbance to the gait strategy in the simulation environment to generate a robust path that adapts to the dynamic changes of the real physical environment, and then loading and issuing the control commands corresponding to the robust path through the native image.

[0018] Furthermore, it also includes: responding to a detected sudden triggering event, completing the replanning of the entire network's tasks and gait based on the native image within a preset time threshold and issuing instructions; the sudden triggering event includes at least one of: robot communication interruption, charging position status change, and sudden emergency alarm.

[0019] Furthermore, the acquisition of status data of multiple robots and inspection task information to be executed includes: aggregating and analyzing the spatiotemporal trajectories uploaded by the multiple robots through a spatial database deployed in the cloud, in order to obtain the real-time position and historical trajectory of each robot.

[0020] Compared with existing technologies, the robot inspection scheduling method based on gait energy efficiency matrix provided by this invention has the following advantages: (1) By incorporating the task urgency and terrain resistance coefficient into the task-gait energy efficiency mapping relationship, the gait power, movement distance, movement speed and terrain resistance coefficient are coupled and quantitatively evaluated, so that the task allocation decision not only considers the physical distance, but also comprehensively measures the gait energy consumption and terrain passage cost, avoiding the problem of low energy utilization caused by fixed gait in traditional methods, and effectively extending the overall endurance of multi-robot clusters.

[0021] (2) By constructing a global energy consumption optimization objective that includes transportation power consumption, execution power consumption and spatial path overlap penalty term, the spatial coverage overlap of multiple robots is introduced into the optimization objective in the form of penalty term, so that the solution process automatically tends to plan spatially mutually exclusive non-overlapping paths, effectively eliminating the collective energy waste caused by the redundancy of multi-robot vision coverage.

[0022] (3) By checking the preset constraints during the process of solving the global energy consumption optimization model, and automatically eliminating the allocation scheme that violates the constraints and resolving when the infeasible solution is caused by the conflict of constraints, especially the risk of a single machine being unable to return due to insufficient remaining power is avoided by power balance constraints, the gait strategy is linked with the task deadline based on time constraints, and the inspection area is ensured to be free of blind spots based on coverage constraints, thereby minimizing the energy consumption of the cluster under the premise of ensuring full coverage of tasks and timely emergency response.

[0023] (4) By collecting telemetry data of each robot in real time during the inspection task, and using foot sensors or inertial measurement units to sense the physical feedback of the current terrain to dynamically update the terrain resistance coefficient, the global model parameters can be adaptively adjusted with the actual terrain changes, and the accuracy of subsequent scheduling decisions can be continuously optimized.

[0024] (5) By introducing simulation-reality migration processing before the gait strategy is issued and adding preset noise disturbances in the simulation environment to generate a robust path, the difference between the simulation environment and the real physical environment is effectively compensated, and the actual execution stability of the gait strategy under complex terrain conditions is improved.

[0025] (6) By loading the scheduling logic based on the pre-compiled native image, and completing the replanning of the entire network tasks and gait and issuing instructions within a preset time threshold when a sudden triggering event is detected, the garbage collection pause and real-time compilation warm-up delay of the traditional Java Virtual Machine runtime are eliminated, ensuring the determinism and real-time performance of the scheduling response in emergency scenarios.

[0026] (7) By dynamically limiting the maximum speed in the gait strategy based on the recognition accuracy requirements in the inspection task information and matching the high-power sensor load, and by performing graded calculation of the power consumption of the execution based on the load devices involved in the task, a fine balance between task execution accuracy and energy consumption is achieved.

[0027] (8) By aggregating and analyzing the spatiotemporal trajectories uploaded by multiple robots through a spatial database deployed in the cloud, real-time location and historical trajectory are obtained, providing efficient and accurate spatiotemporal data support for the global energy consumption optimization model, and further improving the response speed and spatiotemporal accuracy of scheduling decisions. Attached Figure Description

[0028] Figure 1 This is a flowchart of the robot inspection and scheduling method of the present invention; Figure 2 This is a detailed flowchart of the robot inspection scheduling in Embodiment 2 of the present invention; Figure 3 This is a schematic representation of the scene parameters in Embodiment 2 of the present invention. Detailed Implementation

[0029] Example 1

[0030] like Figure 1 As shown, this embodiment provides a robot inspection scheduling method based on gait energy efficiency matrix. This method is applied to an end-edge-cloud collaborative architecture consisting of a cloud scheduling center, edge computing nodes, and multiple quadruped robots (also known as robot dogs). The specific steps include: Step S1: Obtain status data of multiple robots and information on inspection tasks to be performed.

[0031] The status data includes at least the remaining battery power and real-time location of each robot. Optionally, the status data also includes at least one of the following: the current operating status of each robot, the currently executed gait type, and historical energy consumption data. The operating status includes, for example, inspection, charging, or standby. The inspection task information includes at least the task type (e.g., fire confirmation, equipment meter reading, environmental anomaly monitoring), task location coordinates, task deadline, and recognition accuracy requirements. The status data and task information can be obtained by aggregating and analyzing the spatiotemporal trajectories uploaded by each robot through a spatial database deployed in the cloud, or by summarizing the telemetry data reported in real time by each robot through edge computing nodes. This embodiment does not specifically limit this.

[0032] Step S2: Determine the urgency of the task based on the task type in the inspection task information, and determine the terrain resistance coefficient in combination with the terrain information of the inspection area, and establish a task-gait energy efficiency mapping relationship.

[0033] The urgency of a task can be determined using a pre-set urgency mapping table. For example, tasks such as fire confirmation and toxic gas leak detection are mapped to high urgency, while routine tasks like meter reading and regular patrols are mapped to low urgency. Different urgency levels correspond to different gait types: high urgency corresponds to a fast running gait, while low urgency corresponds to an energy-saving diagonal gait or a slow crawling gait. The terrain resistance coefficient is preset with corresponding values ​​based on the terrain information of the inspection area. For example, the terrain resistance coefficient for hard concrete roads is set to a baseline value of 1.0, for grasslands to 1.2, for slopes to 1.5, and for gravel roads to 1.8. By incorporating both task urgency and terrain resistance coefficient into the task-gait energy efficiency mapping relationship, subsequent scheduling decisions consider not only physical distance but also a comprehensive assessment of gait energy consumption and terrain traversal costs.

[0034] Step S3: Calculate the spatial coverage information of the inspection area by the multiple robots, determine the coverage overlap between each robot, and generate a path overlap penalty term based on the coverage overlap.

[0035] The specific methods for calculating spatial coverage information include: obtaining the current visual monitoring range of each robot, which can be determined based on the field of view of the camera mounted on each robot and its current pose; and using spatial relationship functions (such as spatial intersection functions and spatial inclusion relationship functions) to determine the overlapping area between the visual monitoring ranges of different robots, which is used as the coverage overlap degree. In some embodiments, the video field of view of each robot can be projected onto a spatiotemporal grid map, and the overlapping area between each projected area can be calculated using a topological intersection function, which is used as the quantitative value of the coverage overlap degree. The value of the path overlap penalty term generated based on the coverage overlap degree is positively correlated with the coverage overlap degree, that is, the larger the overlapping area, the higher the cost of the penalty term, thereby forcing the system to tend to plan spatially exclusive non-overlapping paths in subsequent optimization solutions.

[0036] Step S4: With the goal of minimizing the total energy consumption of the cluster, the task-gait energy efficiency mapping relationship and the path overlap penalty term are used as variables to construct and solve the global energy consumption optimization model, and obtain the task allocation scheme and corresponding gait strategy of each robot.

[0037] The global energy consumption optimization model is constructed by: establishing an objective function to minimize the total power consumption of the robot cluster during the inspection cycle. This total power consumption includes transportation power consumption, execution power consumption, and spatial path overlap penalty terms. The task allocation relationship of each robot is incorporated into the objective function as a binary decision variable. During the solution process, preset constraints are checked. These preset constraints include at least power balance constraints, timeliness constraints, and coverage constraints. If an infeasible solution is obtained due to constraint conflicts (e.g., a robot's expected remaining power after executing the assigned task is lower than a preset safety threshold, or the expected arrival time is later than the task deadline), the allocation scheme that violates the constraints is automatically eliminated. The remaining selectable robots are then re-selected and the solution is re-solved until the optimal or near-optimal solution that satisfies all constraints is obtained. This step effectively avoids the risk of power depletion caused by single-machine overload, minimizing the total energy consumption of the cluster while ensuring full task coverage and timely emergency response.

[0038] Step S5: Before the gait strategy is sent out for execution, the gait strategy is subjected to simulation-reality transfer processing. Preset noise disturbances are added to the simulation environment to generate a robust path. The scheduling logic is loaded based on the pre-compiled native image, and the scheduling instructions are sent to each robot for execution.

[0039] The purpose of simulation-reality transfer processing is to compensate for the differences between the simulation environment and the real physical environment. Since simulation environments typically use idealized terrain models and dynamic parameters, while real environments present uncertainties such as uneven ground, varying friction coefficients, and sensor noise, directly issuing gait strategies obtained from the simulation environment to the robot may lead to deviations from expected performance. Therefore, pre-defined noise perturbations (such as adding random deviations to the terrain friction coefficient and Gaussian noise to joint torques) are applied to the robot's dynamic model in the simulation environment. Through multiple simulation iterations, robust paths adapted to the dynamic changes of the real physical environment are generated. The control commands corresponding to these robust paths are then loaded and issued via a native image. This native image is generated by the scheduling algorithm using pre-compilation technology. Tools such as GraalVM can be used to pre-compile the scheduling microservice code into a machine-code-level binary image, which is then deployed on edge computing nodes. A real-time end-to-edge synchronization link enables rapid issuance of scheduling commands, eliminating garbage collection pauses and just-in-time compilation warm-up delays inherent in traditional Java Virtual Machine runtime.

[0040] Step S6: During the inspection task, telemetry data of each robot is collected in real time, and global model parameters, including the terrain resistance coefficient, are dynamically updated based on the telemetry data.

[0041] The telemetry data includes the real-time position, remaining battery power, actual movement speed, motor current, and sensor feedback data of each robot. Dynamic updates to the terrain drag coefficient are achieved by utilizing foot sensors or inertial measurement units (IMUs) on each robot to perceive the physical feedback of the current terrain. For example, foot force sensors can detect changes in the contact force between the foot and the ground, and IMUs can detect the degree of body sway. Based on this physical feedback, the actual travel resistance of the current road segment is calculated, and the terrain drag coefficient of the corresponding road segment is corrected in real time. Through this feedback loop, the global model parameters can adaptively adjust with changes in the actual terrain, continuously optimizing the accuracy of subsequent scheduling decisions.

[0042] Furthermore, the method described in this embodiment can also respond to detected sudden triggering events during execution, and based on the native image, complete the replanning of all network tasks and gait within a preset time threshold (e.g., milliseconds) and issue instructions; the sudden triggering events include at least one of: robot communication interruption, charging position status change, and sudden emergency alarm. When the above events are detected, the scheduling center does not need to wait for the Java Virtual Machine to warm up or cold start, but directly starts the scheduling logic in the form of the native image, and completes global replanning in a very short time, ensuring the determinism and real-time performance of the scheduling response in emergency scenarios.

[0043] As one implementation method, in the task-gait energy efficiency mapping relationship described in this embodiment, transportation energy consumption Calculate using the following formula: ; in, The rated power under the corresponding gait command, For the distance traveled, For movement speed, This is the terrain resistance coefficient. This formula couples gait power, travel distance, travel speed, and terrain resistance coefficient into a unified energy consumption calculation relationship, enabling quantitative comparison of transportation energy consumption under different task allocation schemes, and providing an accurate energy consumption assessment basis for the construction of a global optimization model.

[0044] As one implementation method, the construction and solution of the global energy consumption optimization model in this embodiment includes: establishing the objective function of the global energy consumption optimization model as minimizing the total power consumption of the robot cluster during the inspection cycle. : ; in, For the total number of robots, This represents the total number of inspection tasks. Let be a binary decision variable, representing the th Is the robot assigned the first One task, Based on travel distance ,gait and terrain resistance coefficient Determined power consumption during transportation To carry out the mission The inherent power consumption at that time, This is a spatial path overlap penalty term. The penalty coefficient is used. This objective function integrates task allocation decisions, gait selection, terrain adaptability, and spatial coverage into a unified model, reflecting a deep decoupling of physics and logic. Through the penalty term mechanism, the system is driven to automatically avoid multi-machine path overlap, effectively reducing collective energy waste.

[0045] As one implementation method, this embodiment calculates the spatial coverage information of the inspection area by the multiple robots, including: obtaining the current visual monitoring range of each robot; determining the overlapping area between the visual monitoring ranges of different robots using a spatial relationship function, as the coverage overlap; wherein, the calculation of the coverage overlap includes projecting the video field of view of each robot onto a spatiotemporal grid map, and calculating the overlapping area between each projected area using a topological intersection function; the objective function of the global energy consumption optimization model includes a penalty term positively correlated with the coverage overlap. Through the above method, the system can collaboratively optimize the spatial coverage of multiple robots from a global perspective, avoiding unnecessary energy consumption caused by multiple robots repeatedly inspecting the same area, while also ensuring that the preset coverage requirements are met under balanced task allocation.

[0046] As one implementation method, the preset constraints in this embodiment include at least power balance constraints, time constraints, and coverage constraints. The power balance constraint ensures that the remaining power of each robot after executing its assigned task is not lower than a preset safety threshold, thereby reserving power for forced return and preventing robots from being stranded during inspections due to power depletion. The time constraints ensure that the estimated arrival time of each robot at the corresponding task point using its assigned gait strategy is not later than the deadline of the corresponding task, thus guaranteeing the timeliness of response to high-urgency tasks (such as fire confirmation). The coverage constraints ensure that the union of the visual monitoring ranges of each robot covers the preset target inspection area, thereby eliminating blind spots and meeting the requirement of full coverage of the park's inspection tasks.

[0047] As one implementation method, in the global energy consumption optimization model described in this embodiment, the inherent power consumption during task execution is calculated hierarchically based on the load devices involved in the inspection task. The inspection task information also includes recognition accuracy requirements. If the recognition accuracy requirements are higher than a preset threshold, the maximum travel speed in the gait strategy is limited, and high-power sensor loads are matched. For example, for equipment instrument reading tasks requiring high-definition image recognition, the system automatically limits the robot's maximum travel speed to ensure image acquisition quality, while simultaneously activating high-power sensors such as high-resolution cameras or infrared thermal imagers. For simple tasks that only require judging the color of equipment operating status indicator lights, the robot can travel at normal speed, with only the basic camera activated. Through the above methods, a refined balance between task execution accuracy and energy consumption is achieved.

[0048] As one implementation method, the dynamic updating of global model parameters, including the terrain drag coefficient, based on the telemetry data in this embodiment includes: using foot sensors or inertial measurement units (IMUs) on each robot to perceive the physical feedback of the current terrain, and adjusting the terrain drag coefficient of the corresponding road segment in real time based on the physical feedback. For example, when a robot travels through a certain road segment, if the foot sensor detects an abnormal sinking depth or the IMU detects a significant change in body pitch, it indicates that the road segment may have muddy or soft terrain. The system then increases the terrain drag coefficient of that road segment, allowing other robots to predict higher energy consumption costs when traveling through the segment and thus make a more reasonable path selection. This online update mechanism ensures that the global model parameters always remain dynamically consistent with the actual environment.

[0049] As one implementation method, the simulation-reality transfer processing in this embodiment includes: applying a preset noise perturbation to the gait strategy in the simulation environment to generate a robust path that adapts to the dynamic changes of the real physical environment; and then loading and issuing the control commands corresponding to the robust path through the native image. The noise perturbation may include random deviations to the ground friction coefficient, Gaussian noise to joint torques, and small offsets to the positions of environmental obstacles. Through multiple Monte Carlo simulation iterations, robust paths that can maintain stable execution under various perturbation conditions are selected, thereby improving the actual execution success rate of the gait strategy under complex terrain conditions.

[0050] As one implementation method, this embodiment further includes: responding to a detected sudden triggering event, completing the replanning of all network tasks and gait based on the native image within a preset time threshold and issuing instructions; the sudden triggering event includes at least one of: robot communication interruption, charging position status change, and sudden emergency alarm. The preset time threshold can be set according to the security requirements of the application scenario, for example, set to within 500 milliseconds or 1 second. When robot communication is interrupted, the dispatch center can quickly reassign tasks to other robots with normal communication; when the charging position status changes (e.g., a charging pile is occupied or malfunctions), the dispatch center can replan the charging path for low-battery robots; when a sudden emergency alarm (e.g., fire alarm, unauthorized intrusion) is triggered, the dispatch center can immediately adjust task priorities and reassign the nearest robot with sufficient battery power to handle the situation.

[0051] As one implementation method, the acquisition of status data of multiple robots and inspection task information to be executed in this embodiment includes: aggregating and analyzing the spatiotemporal trajectories uploaded by the multiple robots through a spatial database deployed in the cloud to obtain the real-time location and historical trajectory of each robot. The spatial database can be a database system that supports geospatial data types and spatial indexes (such as the PostGIS extension of PostgreSQL). It can achieve efficient aggregation and analysis of robot spatiotemporal trajectories through spatial query statements (such as ST_Intersects, ST_Contains, etc.), providing spatiotemporal data support with second-level response for the global energy consumption optimization model, further improving the response speed and spatiotemporal accuracy of scheduling decisions.

[0052] Example 2

[0053] This embodiment is a further supplementary explanation of Embodiment 1. It elaborates on the complete implementation process of the method of the present invention in detail, combining specific application scenarios and numerical calculation examples.

[0054] like Figure 2 As shown, the method of this invention is implemented through an edge-cloud collaborative architecture. A high-concurrency scheduling microservice is deployed in the cloud, and a spatial database is used to perform second-level analysis of the spatiotemporal trajectories uploaded by the robot dog. In this embodiment, the spatial database adopts a database system that supports geospatial data types and spatial indexes (such as the PostGIS extension of PostgreSQL), and efficient aggregation analysis of the robot's spatiotemporal trajectories is achieved through spatial query statements (such as ST_Intersects, ST_Contains, etc.).

[0055] The complete process of this embodiment includes: Step 1 (task perception and analysis), Step 2 (multi-machine collaborative scheduling decision and global energy consumption optimization calculation), Step 3 (Sim-to-Real migration and path generation), and Step 4 (execution and feedback closed loop). The following is a combination of... Figure 2 The entire process will be explained in detail.

[0056] Step 1: Task perception and analysis; like Figure 2 As shown in Step 1, Task Perception and Analysis, the entire process begins with receiving a new inspection task or an emergency alarm.

[0057] The system acquires status data from all robot dogs (SoC remaining battery power, location, and operating status) and information on inspection tasks to be executed, analyzing task attributes, including task type, recognition accuracy requirements, and operation type. Based on this, the system establishes a multi-dimensional task-gait energy efficiency matrix. Specifically, it defines a task urgency weight α and a terrain resistance coefficient β, adaptively calculating the optimal gait command according to the inspection task type. For example, tasks involving safety risks, such as fire detection, are set to high urgency, corresponding to a fast running gait; routine tasks, such as daily equipment inspections, are set to normal urgency, corresponding to an energy-saving diagonal trotting gait. The terrain resistance coefficient β is preset with corresponding values ​​based on the terrain information of the inspection area, for example, 1.0 for hard concrete surfaces, 1.2 for grass, 1.5 for slopes, and 1.8 for gravel surfaces.

[0058] By incorporating both task urgency and terrain resistance coefficient into the task-gait energy efficiency mapping relationship, subsequent scheduling decisions can consider not only physical distance but also a comprehensive assessment of gait energy consumption and terrain access costs. Among these, transportation energy consumption... Calculate using the following formula: ; in, The rated power under the corresponding gait command, For the distance traveled, For movement speed, This is the terrain resistance coefficient. This formula couples gait power, travel distance, travel speed, and terrain resistance coefficient into a unified energy consumption calculation relationship, enabling quantitative comparison of transportation energy consumption under different task allocation schemes, and providing an accurate energy consumption assessment basis for the construction of a global optimization model.

[0059] like Figure 2 As shown in step two, “Multi-machine collaborative scheduling decision-making”, after completing task perception and analysis, the system enters the core optimization calculation stage.

[0060] First, the system performs preprocessing based on spatiotemporal grid dynamic deduplication and task allocation. Using spatial relationship functions (such as intersection and inclusion functions), it calculates in real-time the video coverage blind spots and overlapping areas of multiple robots during the inspection process. Specifically, this includes: obtaining the current visual monitoring range of each robot and projecting it onto the spatiotemporal grid map; calculating the overlap area between projected regions using a topological intersection function. If robot A has already covered a certain area, the system automatically reduces the dwell time of robot B in that area to plan the optimal non-overlapping path.

[0061] Subsequently, the system constructs a global multi-objective energy consumption optimization mathematical model. In the patent, the mathematical model is the core element embodying "non-obviousness," and this invention defines the scheduling problem as a constrained multi-objective optimization problem. Assume there are [missing information - likely related to energy consumption optimization within the park]. Taiwanese robot dog, performing Each inspection task aims to minimize the total power consumption of the cluster within a time period T. .

[0062] The objective function is to minimize global power consumption.

[0063] The variables are explained as follows: Binary decision variables, such as those for robot dogs. Assign tasks If it is 1, then it is 1; otherwise, it is 0.

[0064] Transportation power consumption, affected by distance ,gait and terrain resistance Influence.

[0065] : Inherent power consumption during task execution (such as turning on high-power infrared sensors, robotic arm operation, etc.).

[0066] : Penalty term, used to reduce the overlap of video coverage areas of multiple machines, λ is the penalty coefficient.

[0067] The core constraints of the model include: Time constraints: Ensure the robot dog moves in a gait The journey will be completed on time.

[0068] Power balance constraints: Reserve power for forced return to base to prevent the robot from being stranded during inspections due to depleted power.

[0069] Coverage constraints: To ensure that there are no blind spots in the park inspection.

[0070] like Figure 2 As shown in the "Solve" section of step two, the system calculates transportation energy consumption, execution energy consumption, and path overlap penalty in sequence, and finally sums up the total power consumption.

[0071] For ease of explanation, assume there are 2 robot dogs (N=2) and 2 unexpected tasks (M=2) deployed in the park. Specific scenario parameters are as follows: Figure 3 As shown.

[0072] The following compares the J-values ​​(total power consumption) of the two schemes.

[0073] Option 1: Assignment based on proximity (traditional logic); Assignment method: B goes to task 1 (100m, grass), A goes to task 2 (100m, cement road).

[0074] Calculate transportation energy consumption : B goes to Task 1 (Grassland β=1.2): ; A goes to Task 2 (Cement Road β=1.0): ; Substitute into the formula (Note: 50Wh = 180,000J): ; Option 2: Global Energy Efficiency Optimization (Logic of this Implementation Example); Considering that B has low battery and task 1 requires high power consumption, the system decides to let A run a long distance and B walk slowly over a short distance.

[0075] Assignment method: A goes to task 1 (200m, cement road), B goes to task 2 (300m, cement road).

[0076] Calculate transportation energy consumption : A goes to Task 1: ; B goes to Task 2: ; Substitute into the formula: ; like Figure 2 As shown in step two, "Check Constraints," although in terms of absolute power consumption... However, the constraint checking mechanism of the algorithm provided in the implementation example will come into play. In Scheme 1, after robot dog B completes the fire alarm task, its remaining battery power will be lower than the safety threshold (E_safe_to_charge), constituting an infeasible solution. Figure 2 As shown, the scheme was identified and eliminated by the system because it "did not meet the requirements (low battery, timeout)". The algorithm then "removed the unqualified robot dog and solved it again", and finally output the second scheme as the "optimal task allocation and gait strategy" that satisfies all constraints.

[0077] In addition, the overlap penalty mechanism is equally crucial: if the paths of A and B overlap in the GIS map, the λ·Ω term will surge, for example, causing an additional penalty of 10,000J on the total power consumption J, thereby forcing the system to plan the optimal path that is spatially mutually exclusive.

[0078] Step 3: Sim-to-Real migration and path generation; like Figure 2As shown in Step 3, "Sim-to-Real Migration and Path Generation," after obtaining the optimal task allocation and gait strategy, the system performs simulation-to-reality migration processing. Preset noise perturbations (such as random deviations to the ground friction coefficient and Gaussian noise to joint torques) are applied to the gait strategy in the simulation environment. Through multiple Monte Carlo simulation iterations, a robust path that adapts to the dynamic changes of the real physical environment is generated.

[0079] After the path is generated, the system begins issuing commands. This process demonstrates the engineering implementation of "how the system runs efficiently in a real-world environment." The scheduling center logic employs Ahead-of-Time Compilation (AOC) technology, using tools such as GraalVM to pre-compile the scheduling microservices into machine-code-level native binary images. For example... Figure 2 As shown, the system uses the GraalVM native image rapid loading control algorithm to quickly distribute the control instructions corresponding to the robust path to each machine dog for execution. This method eliminates the garbage collection pauses and just-in-time compilation warm-up delays (GC pauses and JIT compilation time) of traditional Java Virtual Machine runtime, enabling global energy-efficient scheduling to execute at hardware-level response speed.

[0080] Step 4: Execution and Feedback Loop; like Figure 2 As shown in step four, “Execution and Feedback Closed Loop,” each robot dog performs inspection tasks according to the issued gait strategy.

[0081] During mission execution, the system collects telemetry data in real time and dynamically updates global model parameters, including terrain coefficients and historical energy consumption. Specifically, it uses foot sensors or inertial measurement units on each robot dog to perceive the physical feedback of the current terrain and corrects the terrain resistance coefficient of the corresponding road segment in real time, thereby continuously optimizing the accuracy of subsequent scheduling decisions.

[0082] Meanwhile, the system continuously monitors its status. For example... Figure 2 The system's judgment logic involves detecting any anomalies: if an anomaly is detected, an alarm is reported; if the task is fully completed, the current scheduling cycle ends; otherwise, it continues inspection or waits to receive new tasks. Furthermore, the trajectory and recognition results generated by the robot during the task are ultimately stored in a PostGIS spatial database, providing efficient and accurate spatiotemporal data support for future scheduling decisions.

[0083] The method described in this embodiment has the ability to handle emergencies, demonstrating the advantages of ultra-fast rescheduling based on AOT compilation. When encountering situations such as network interruption, charging pile queuing, or sudden alarms, the dispatch center does not need to wait for the Java Virtual Machine to warm up or cold start. Instead, it directly starts the scheduling logic in the form of a native image, completes the replanning of all network tasks and gait within milliseconds, and issues instructions, ensuring the determinism and real-time performance of the scheduling response in emergency scenarios.

[0084] The above calculation examples and process descriptions demonstrate that this invention is not a simple shortest distance search, but rather incorporates a terrain coefficient β and task power consumption. The model is solved under multi-objective constraints. In multi-machine concurrent scenarios, this model can effectively avoid the risk of single-machine overload (power depletion) and reduce the power consumption of invalid overlapping paths in the cluster by 12.5%.

[0085] The key innovations of this invention are: (1) Deep decoupling of physics and logic: Instead of simply prioritizing the shortest distance, it models gait frequency, terrain resistance, and GIS video coverage as energy consumption variables. (2) Deterministic computing architecture: In response to the stringent latency requirements of embodied intelligence in emergency obstacle avoidance and sudden scheduling, AOT native compilation technology is used to eliminate the nondeterministic latency of the Java Virtual Machine (GC pauses and JIT compilation time), enabling global energy efficiency scheduling to be executed at hardware-level response speed.

[0086] Compared with existing technologies, the advantages of this invention are: (1) Improved battery life: By decoupling task-gait and deduplicating multi-machine paths, the overall battery life of the robot cluster in the park is expected to increase by 15%-25%. (2) Extreme response: The cold start speed at the native image level eliminates the warm-up delay of traditional Java microservices in the edge cloud native environment, ensuring the real-time nature of emergency response. (3) Spatiotemporal accuracy: By introducing geospatial topology analysis, task allocation not only considers "how far away" but also "how much power it takes to get there".

[0087] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A robot inspection scheduling method based on gait energy efficiency matrix, characterized in that, include: Acquire status data of multiple robots and inspection task information to be executed. The status data includes at least the remaining battery power and real-time location of each robot. The urgency of the task is determined based on the task type in the inspection task information, and the terrain resistance coefficient is determined in combination with the terrain information of the inspection area to establish a task-gait energy efficiency mapping relationship. Calculate the spatial coverage information of the inspection area by the multiple robots, determine the coverage overlap between each robot, and generate a path overlap penalty term based on the coverage overlap. With the goal of minimizing the total energy consumption of the cluster, the task-gait energy efficiency mapping relationship and the path overlap penalty term are used as variables to construct and solve a global energy consumption optimization model to obtain the task allocation scheme and corresponding gait strategy of each robot. During the solution process, preset constraints are checked. For infeasible solutions caused by constraint conflicts, allocation schemes that violate the constraints are automatically eliminated and the solution is resolved. Before the gait strategy is sent out for execution, the gait strategy is subjected to simulation-reality transfer processing. Preset noise disturbances are added to the simulation environment to generate a robust path. The scheduling logic is loaded based on the pre-compiled native image and the scheduling instructions are sent to each robot for execution. During the inspection process, telemetry data from each robot is collected in real time, and global model parameters, including the terrain resistance coefficient, are dynamically updated based on the telemetry data. In the task-gait energy efficiency mapping relationship, transportation energy consumption Calculate using the following formula: ； in, The rated power under the corresponding gait command, For the distance traveled, For movement speed, This is the terrain resistance coefficient; The construction and solution of the global energy consumption optimization model includes: establishing the objective function of the global energy consumption optimization model as minimizing the total power consumption of the robot cluster during the inspection cycle. ; ； in, For the total number of robots, This represents the total number of inspection tasks. Let be a binary decision variable, representing the th Is the robot assigned the first One task, Based on travel distance ,gait and terrain resistance coefficient Determined power consumption during transportation To carry out the mission The inherent power consumption at that time, This is a spatial path overlap penalty term. This is the penalty coefficient.

2. The method according to claim 1, characterized in that, Calculating the spatial coverage information of the inspection area by the multiple robots includes: obtaining the current visual monitoring range of each robot; determining the overlapping area between the visual monitoring ranges of different robots using a spatial relationship function, which is used as the coverage overlap; wherein, the calculation of the coverage overlap includes projecting the visual monitoring range of each robot onto a spatiotemporal grid map, and calculating the overlapping area between each projected area using a topological intersection function; the objective function of the global energy consumption optimization model includes a penalty term that is positively correlated with the coverage overlap.

3. The method according to claim 1, characterized in that, The preset constraints include at least a power balance constraint, a timeliness constraint, and a coverage constraint; wherein, the power balance constraint is used to ensure that the remaining power of each robot after performing the assigned task is not lower than a preset safety threshold, the timeliness constraint is used to ensure that the estimated arrival time of each robot to the corresponding task point with the assigned gait strategy is not later than the deadline of the corresponding task, and the coverage constraint is used to ensure that the union of the visual monitoring range of each robot covers the preset target inspection area.

4. The method according to claim 1, characterized in that, In the global energy consumption optimization model, the inherent power consumption during task execution is calculated in stages based on the load devices involved in the inspection task; the inspection task information also includes recognition accuracy requirements. If the recognition accuracy requirements are higher than a preset threshold, the maximum walking speed in the gait strategy is limited, and a high-power sensor load is matched.

5. The method according to claim 1, characterized in that, The dynamic updating of global model parameters, including the terrain drag coefficient, based on the telemetry data includes: using foot sensors or inertial measurement units on each robot to sense the physical feedback of the current terrain, and correcting the terrain drag coefficient of the corresponding road segment in real time based on the physical feedback.

6. The method according to claim 1, characterized in that, The simulation-to-reality migration process includes: applying a preset noise disturbance to the gait strategy in the simulation environment to generate a robust path that adapts to the dynamic changes of the real physical environment, and then loading and issuing the control commands corresponding to the robust path through the native image.

7. The method according to claim 1, characterized in that, Also includes: In response to detected sudden triggering events, the system completes the replanning of all network tasks and gait within a preset time threshold based on the native image and issues instructions. The sudden triggering events include at least one of the following: robot communication interruption, charging position status change, and sudden emergency alarm.

8. The method according to any one of claims 1 to 7, characterized in that, Acquire status data and inspection task information of multiple robots, including: aggregate and analyze the spatiotemporal trajectories uploaded by the multiple robots through a spatial database deployed in the cloud, in order to obtain the real-time position and historical trajectory of each robot.

Images