A collaborative operation scheduling and recovery method and system for concrete rebound detection using a quadruped robot at a substation construction site.
By establishing a mapping relationship between candidate stations of quadruped robots and inspection surfaces at the substation construction site, and combining the reachability and safety constraints of the robotic arm, the problem that inspection surfaces cannot be completed within a limited time in the concrete rebound inspection of quadruped robots was solved, and efficient inspection surface selection and operation recovery were achieved.
Patent Information
- Application Number
- CN202611457921.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-18
- Publication Date
- 2026-11-17
AI Technical Summary
At substation construction sites, the concrete rebound detection of quadruped robots is limited by the reachability of the robotic arm, the working time window, and safety constraints, which makes it impossible to complete the detection of the surface within a limited time. The existing dynamic scheduling method increases unnecessary task rescheduling and robot relocation, making it difficult to make effective use of the working time.
By establishing a mapping relationship between candidate positions of the quadruped robot and the detection surface, and combining the reachability state of the robotic arm, the springback alignment state, the operation permission time window and safety constraints, the validity of the mapping relationship is determined. When conditions change, only the failed part is reconstructed to generate alternative operation units, thereby realizing the spatiotemporal executability of the detection surface.
It improves the efficiency and stability of testing within a limited working time, reduces redundant planning, adapts to the time and safety constraints of substation construction sites, and ensures that the testing panels can be completed under actual conditions.
Smart Images

Figure CN123113769A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of robot collaborative operation, and in particular to a collaborative operation scheduling and recovery method and system for concrete rebound detection by a quadruped robot at a substation construction site. Background Technology
[0002] With the increasing level of intelligence at substation renovation and expansion construction sites, quadruped robots are gradually being used for tasks such as construction area inspection, environmental perception, and quality inspection. Compared to wheeled robots, quadruped robots can adapt to unstructured environments such as equipment area passages, steps, ramps, and complex construction surfaces. They can also be equipped with perception devices such as vision sensors and LiDAR to detect the construction environment and the objects to be inspected. In concrete structure inspection scenarios, the robot can be further equipped with a robotic arm and a rebound detection device, enabling it to autonomously approach the components to be inspected and perform contact rebound detection.
[0003] Existing robotic springback detection solutions typically require first identifying the component to be inspected and obtaining information such as the location of the inspection point, the component's surface pose, or the surface normal. Then, the robotic arm is controlled to complete the inspection based on its range of motion and end effector posture. For mobile robots, it is also necessary to determine the robot's working position relative to the component to be inspected.
[0004] However, in the substation renovation and expansion construction environment, robotic arm contact operations are not only limited by the spatial accessibility of the robotic arm, but also constrained by work permits, planned power outage windows, safe distances to energized areas, and the stability of the quadruped robot's position. Especially within the limited working window, the robot's arrival at the candidate position, posture adjustment and stabilization, robotic arm performance of detection, and safe exit all require corresponding time. Therefore, the detection areas that are spatially accessible may not necessarily be completed within the currently allowed working time.
[0005] Furthermore, when the power outage window shortens, the safety status changes, or the robot's operational status becomes abnormal, replanning tasks solely on a complete task basis can easily lead to the recalculation of previously determined and still executable combinations of stations and detection points, increasing unnecessary task rescheduling and robot relocation. Although existing dynamic scheduling methods can achieve job recovery through task suspension, breakpoint saving, and rescheduling of remaining tasks, it is still necessary to further address how to correlate the spatial executability of the detection surface with dynamic job permission conditions, and how to reconstruct only the failed parts when the permission conditions change. This would improve the effective detection capability within a limited operation time and reduce the amount of repetitive planning during job recovery. Summary of the Invention
[0006] To address the aforementioned technical problems in the existing technology, this invention proposes a collaborative operation scheduling and recovery method and system for concrete rebound detection using a quadruped robot at a substation construction site, in order to solve the above-mentioned technical problems.
[0007] According to a first aspect of the present invention, a collaborative operation scheduling and recovery method for concrete rebound detection using a quadruped robot at a substation construction site is proposed, comprising: S1: Obtain candidate detection patches of the component to be inspected, candidate positions of the quadruped robot, and work permission time window and safety constraint information for the rebound detection task. Among them, the candidate detection patches have three-dimensional position and surface normal. S2: Establish a job mapping relationship between each candidate station and each candidate detection surface, and determine the validity of the job mapping relationship based on the reachability of the robotic arm, the rebound alignment status, the candidate station status, and the job permission time window. S3: Determine the target station and the corresponding target detection patch subset based on the effective operation mapping relationship, generate an operation unit including the target station, the target detection patch subset and the effective execution range, and control the quadruped robot and the robotic arm to execute the operation unit; S4: When changes occur in the job permission time window, safety constraint information, or the job status of the quadruped robot that cause changes in the execution conditions of the current job unit, determine the affected job mapping relationship based on the time conditions, candidate positions, or candidate detection patches affected by the changes, re-determine the validity of the affected job mapping relationship, mark the job mapping relationship that no longer meets the valid conditions as the invalid mapping relationship, and retain the valid mapping relationship that is not affected. S5: Generate alternative work units based on the incomplete detection patches and candidate stations corresponding to the failure mapping relationship, and form a subsequent recovery work sequence by combining the unaffected valid mapping relationship.
[0008] In some specific embodiments, candidate detection patches are generated based on the three-dimensional detection results of the component to be detected. Each candidate detection patch includes at least the three-dimensional coordinates of the center of the detection patch, the surface normal, and the available detection area. Candidate detection patches whose available detection area does not meet the preset requirements, whose distance from the edge of the component to be detected does not meet the preset net distance requirements, or which contain holes or contaminated areas are rejected.
[0009] In some specific embodiments, the reachability state of the robotic arm in the job mapping relationship is determined based on the reachability of the robotic arm end effector to the candidate detection surface under the corresponding candidate station; the springback alignment state is determined based on the deviation between the detection direction of the robotic arm end effector and the surface normal of the candidate detection surface. When the deviation meets the preset alignment conditions, the corresponding candidate detection surface is marked as a spatially executable surface under the candidate station.
[0010] In some specific embodiments, for any spatially executable surface, based on the estimated time for the quadruped robot to reach the corresponding candidate station, the stabilization time in place, the estimated detection time of the corresponding candidate detection surface, and the safety allowance time required for the robotic arm to exit the operation state, it is determined whether the spatially executable surface can be completed within the permitted operation time window; if it can be completed, the corresponding operation mapping relationship is marked as a valid mapping relationship.
[0011] In some specific embodiments, for the same candidate station, the expected detection time of the candidate detection patches corresponding to the effective mapping relationship is occupied sequentially within the job permission time window, and the executable detection patch subset corresponding to the candidate station is determined by the remaining available time of the job permission time window; the target station is determined based on the coverage of the executable detection patch subset and the movement cost of the quadruped robot to reach the candidate station.
[0012] In some specific embodiments, the work permit time window is determined based on the valid period of the work order; for rebound detection tasks with power outage requirements, the work permit time window is determined by the overlap of the valid period of the work order and the planned power outage window; the safety constraint information includes at least the status of the work area, the safe distance between the robot and the live conductor, and the stability status of the candidate station.
[0013] In some specific embodiments, when the job permission time window is shortened, only job mapping relationships whose execution completion time exceeds the shortened job permission time window are marked as invalid; job mapping relationships that are still valid under the current target station continue to be executed, and incomplete detection patches that have become invalid due to the shortened job permission time window are moved into the set of patches to be reconstructed.
[0014] In some specific embodiments, when the confidence level of the three-dimensional pose of the component to be detected is lower than a preset threshold, the normal angle error of the candidate detection patch exceeds a preset error threshold, or the difference in surface normal between the visual detection result and the point cloud detection result exceeds a preset threshold, a re-observation is performed on the corresponding component to be detected, the three-dimensional position and surface normal of the candidate detection patch are updated, and the operation mapping relationship related to the candidate detection patch is re-determined.
[0015] In some specific embodiments, when a work ticket failure, a change in the safety status of the work area, a distance between the robot and a live conductor that is less than a safe distance, or an instability in the current target position is detected, the work mapping relationship affected by the change in safety constraints in the current work unit is set to failure, the robotic arm is controlled to stop the rebound detection and exit to a preset safe posture; after the corresponding safety constraints are satisfied again, a replacement work unit is regenerated only for the unfinished detection patches corresponding to the failure mapping relationship.
[0016] According to a second aspect of the present invention, a collaborative operation scheduling and recovery system for concrete rebound detection using a quadruped robot at a substation construction site is proposed, for implementing the method described above, comprising: The perception module is configured to acquire candidate detection patches with three-dimensional position and surface normal, candidate positions of the quadruped robot, and work permission time window and safety constraint information for the rebound detection task. The mapping relationship construction module is configured to establish a task mapping relationship between candidate positions and candidate detection patches for a quadruped robot, and to determine the validity of the task mapping relationship based on the reachability of the robotic arm, the rebound alignment status, the candidate position status, and the task permission time window. The job unit generation module is configured to determine the target station location and target detection patch subset based on the effective mapping relationship, and generate a job unit that includes the target station location, target detection patch subset, and effective execution range; The execution monitoring module is configured to re-evaluate the validity of affected job mapping relationships when changes occur in the job permit time window, safety constraint information, or the quadruped robot's operating status; and The recovery module is configured to generate alternative work units based on the incomplete detection patches and candidate stations corresponding to the failure mapping relationship, and combine them with the unaffected valid mapping relationships to form a subsequent recovery work sequence.
[0017] This invention proposes a collaborative operation scheduling and recovery method and system for concrete rebound detection using a quadruped robot at a substation construction site. Compared with existing technologies, it has the following advantages: This application establishes a task mapping relationship between candidate positions of a quadruped robot and candidate inspection patches of the component to be inspected. It also introduces the reachability state of the robotic arm, the springback alignment state, the candidate position state, and the task permission time window for validity determination. This transforms the inspection patch from simply "spatially reachable" into a "spatiotemporally executable" object that can actually be completed under the current operational conditions. Therefore, when selecting a target position, the robot can comprehensively consider the spatial distribution of the inspection patches covered by that position, the time required for the robot to reach and stabilize, the inspection execution time, and the remaining permission time. This reduces the likelihood of the robot being unable to perform inspection due to insufficient time window or unmet safety conditions after arriving at the position, thus improving the utilization efficiency of the limited task permission time.
[0018] When the job permission time window, safety constraint information, or robot operation status changes, this application does not directly discard the entire original job plan. Instead, it re-evaluates the affected job mapping relationships, only invalidating those mapping relationships that no longer meet the execution conditions, while retaining the unaffected valid mapping relationships. Alternative job units are generated for the incomplete detection patches corresponding to the invalidated mapping relationships. Therefore, it can reduce the need for full replanning and unnecessary robot relocation caused by disturbance events, achieving partial recovery while maintaining completed detection results and still valid job relationships, thus improving the stability and recovery efficiency of the continuous detection process.
[0019] Furthermore, when the target position changes or the reliability of the 3D pose of the component under inspection decreases, the 3D position and surface normal of the component under inspection can be reacquired and the relevant operation mapping relationship can be updated. In cases such as work permit failure, insufficient safety distance, or target position instability, the corresponding operation mapping relationship is invalidated and the robotic arm is controlled to exit to a safe state. This forms a continuous technical closed loop of object perception, robot positioning, robotic arm execution, safety authorization, and anomaly recovery, which is more suitable for quadrupedal robot contact inspection operations with time windows and safety constraints in substation renovation and expansion construction sites. Attached Figure Description
[0020] The accompanying drawings are included to provide a further understanding of the embodiments and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments and, together with the description, serve to explain the principles of the invention. Other embodiments and many anticipated advantages of the embodiments will be readily recognized as they become better understood through reference to the following detailed description. Other features, objects, and advantages of this application will become more apparent from reading the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart of a collaborative operation scheduling and recovery method for concrete rebound detection using a quadruped robot at a substation construction site, according to one embodiment of this application. Figure 2 This is a schematic diagram of the concrete wall panel detection and candidate detection patch generation process according to a specific embodiment of this application; Figure 3 This is an architecture diagram of a collaborative operation scheduling and recovery system for concrete rebound detection using a quadruped robot at a substation construction site, according to one embodiment of this application. Figure 4 This is a schematic diagram of the scheduling process for job unit failure and recovery under disturbance conditions, which is a specific embodiment of this application. Detailed Implementation
[0021] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.
[0022] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0023] Figure 1 A flowchart illustrating a collaborative operation scheduling and recovery method for concrete rebound detection using a quadruped robot at a substation construction site, according to an embodiment of this application, is shown. Figure 1 As shown, the method includes: S1: Obtain candidate detection patches of the component to be inspected, candidate positions of the quadruped robot, and the work permission time window and safety constraint information of the rebound detection task. Among them, the candidate detection patches have three-dimensional position and surface normal.
[0024] In a specific embodiment, candidate detection patches are generated based on the 3D detection results of the component to be detected. Each candidate detection patch includes at least the 3D coordinates of the patch center, the surface normal, and the available detection area. Specifically, the 3D pose of the component to be detected can be obtained based on the visual detection results and point cloud detection results, and multiple candidate detection patches can be generated on the surface of the component. Candidate detection patches whose available detection area does not meet the preset requirements, whose distance from the edge of the component to be detected does not meet the preset clearance requirements, or which contain holes or contaminated areas are eliminated, thereby obtaining a set of candidate detection patches suitable for subsequent rebound detection by the robotic arm. By screening the detection areas on the surface of the component to be detected before scheduling, areas that do not meet the rebound detection requirements can be avoided from continuing to participate in subsequent position matching and operation planning, reducing invalid robotic arm reachability calculations and robot position adjustments, and improving the effectiveness of subsequent operation mapping relationship establishment and the executability of detection tasks.
[0025] In a specific embodiment, the work permit time window is determined based on the valid period of the work order; for rebound detection tasks requiring power outages, the work permit time window is determined by the overlap between the valid period of the work order and the planned power outage window. Safety constraint information includes at least the status of the work area, the safe distance between the robot and the live conductor, and the stability status of the candidate station. Therefore, for detection tasks without power outage requirements, the permitted work period can be determined based on the work order and the current safety status; for detection tasks requiring power outages, the actual permitted execution time range is further limited by the planned power outage window. By incorporating the valid period of the work order, the planned power outage window, and the on-site safety status into the work permit conditions for rebound detection, the robotic arm can avoid performing detection solely based on spatial accessibility, ensuring that subsequently generated work units simultaneously meet both work time and on-site safety requirements, thus improving the safety of quadruped robots performing contact detection tasks at substation construction sites.
[0026] S2: Establish a job mapping relationship between each candidate station and each candidate detection surface, and determine the validity of the job mapping relationship based on the reachability of the robotic arm, the rebound alignment status, the candidate station status, and the job permission time window.
[0027] In a specific embodiment, for any candidate station, the candidate detection surface that the robotic arm's end effector can reach is determined based on the robotic arm's working range at that candidate station. Furthermore, based on the alignment relationship between the robotic arm's end effector's detection direction and the surface normal of the corresponding candidate detection surface, the reachable candidate detection surfaces are further screened. When the robotic arm's end effector can reach the corresponding candidate detection surface, and the deviation between the robotic arm's end effector's detection direction and the surface normal of the candidate detection surface meets a preset alignment condition, the candidate detection surface is determined as a spatially executable surface at the corresponding candidate station. By simultaneously considering both the spatial reachability of the robotic arm's end effector and the alignment relationship between the rebound detection direction and the surface normal of the detection surface, it avoids situations where the rebound bomb's incident direction does not meet requirements due to using only "being able to reach the detection position" as the working condition. This makes the operational mapping relationship established between the candidate station and the candidate detection surface more consistent with the actual execution conditions of contact rebound detection.
[0028] Furthermore, for any spatially executable surface, based on the estimated time (arrival time) for the quadruped robot to reach the corresponding candidate station, the stabilization time required for the quadruped robot to complete posture adjustment and positioning at the candidate station, the estimated detection time for the corresponding candidate detection surface, and the safety allowance time required for the robotic arm to exit the operation state, it is determined whether the spatially executable surface can be completed within the permitted operation time window. When the spatially executable condition is met and the corresponding detection can be completed within the permitted operation time window, the operation mapping relationship between the candidate station and the candidate detection surface is marked as a valid mapping relationship; otherwise, it is excluded from the current valid mapping relationship. In this way, the correspondence between the candidate station and the candidate detection surface is not only constrained by the spatial position and the reachability of the robotic arm, but also further constrained by the current remaining operation time. This further filters the spatially "executable" detection surfaces into detection surfaces that are "actually completeable" within the current permitted operation time window, avoiding the inability to complete the detection due to insufficient remaining time after the robot arrives at the station and completes posture adjustment, thus improving the effective utilization rate of limited operation time.
[0029] In one specific implementation, the job mapping relationship is used to characterize the corresponding execution relationship between a candidate station and a candidate detection patch under the current working conditions. The job mapping relationship is associated with at least the candidate station identifier, the candidate detection patch identifier, the reachability state of the robotic arm, the rebound alignment state, the time-executable state, and the current valid state. It can also be further associated with the expected arrival time of the quadruped robot to the corresponding candidate station, the stabilization time, the expected detection time of the corresponding candidate detection patch, and the expected completion time. Thus, the spatial execution conditions and time execution conditions can be recorded separately for different candidate stations and different candidate detection patches, without treating the entire detection task as a single valid or invalid object. In this embodiment, the valid execution interval refers to the time interval within the job permission time window that allows a subset of the corresponding target detection patches to perform rebound detection, provided that the quadruped robot reaches the corresponding candidate station and completes stabilization and the reserved time for the robotic arm to safely exit. When the job permission time window changes or the expected arrival time of the robot changes, the valid execution interval can be updated accordingly, and the time-executable state of the corresponding job mapping relationship can be re-determined.
[0030] S3: Determine the target station and the corresponding target detection patch subset based on the effective job mapping relationship, generate a job unit including the target station, the target detection patch subset and the effective execution range, and control the quadruped robot and the robotic arm to execute the job unit.
[0031] In specific embodiments, for the same candidate station, candidate detection patches that can be continuously executed within the current work permit time window are determined based on multiple valid mapping relationships corresponding to the candidate station. Specifically, the work permit time window can be occupied sequentially according to the expected detection time of each candidate detection patch, and the executable detection patch subset corresponding to the candidate station can be determined based on the remaining available time of the work permit time window. When multiple candidate stations correspond to at least one executable detection patch subset, the target station is determined based on the coverage of the executable detection patch subset corresponding to each candidate station (based on the number of candidate detection patches contained in the executable detection patch subset) and the movement cost of the quadruped robot to reach the corresponding candidate station. The movement cost of the candidate station can be determined based on the expected movement time of the quadruped robot from its current position to the corresponding candidate station. In some embodiments, it can also be further determined by combining the path length and expected energy consumption of the corresponding movement path. In some embodiments, candidate stations that can complete a large number of candidate detection patches within the current work permit time window can be preferentially selected. When multiple candidate stations correspond to the same number of candidate detection patches, the target station can be further determined based on the movement time and stabilization time of the quadruped robot to the corresponding candidate station. After determining the target station, the target station, the subset of target detection patches that can be executed at the target station, and the effective execution interval of the subset of target detection patches are associated to form a work unit. The quadruped robot first moves to the target station and completes posture adjustment and stabilization, and then controls the robotic arm to perform rebound detection on the detection patches in the subset of target detection patches in sequence. By binding the target station, the subset of detection patches, and the effective execution interval to form a work unit, the station selection no longer simply pursues the spatial coverage capability of the robotic arm, but further considers the number of detection patches that can actually be completed within the limited work permit time and the cost of robot relocation, thereby reducing the time consumption caused by frequent station transfers and improving the effective detection coverage and overall detection efficiency under a single stable station.
[0032] S4: When changes occur in the job permission time window, safety constraint information, or the job status of the quadruped robot that cause changes in the execution conditions of the current job unit, determine the affected job mapping relationship based on the time conditions, candidate positions, or candidate detection patches affected by the changes, re-determine the validity of the affected job mapping relationship, mark the job mapping relationship that no longer meets the valid conditions as the invalid mapping relationship, and retain the valid mapping relationship that is not affected.
[0033] In a specific embodiment, when the job permission time window is shortened, the estimated completion time of each incomplete detection patch in the current job unit is re-determined based on the shortened job permission time window. For detection patches that can still be completed within the shortened job permission time window, their corresponding job mapping relationships remain valid and execution continues; for detection patches whose estimated completion time exceeds the shortened job permission time window, the corresponding job mapping relationships are marked as invalid mapping relationships, and the corresponding incomplete detection patches are moved to the set of patches to be reconstructed. By invalidating only job mapping relationships that no longer meet the execution conditions due to the shortening of the job permission time window, it is possible to avoid canceling all detection tasks under the current target station due to changes in local time constraints, allowing detection patches that can still be completed within the remaining window to continue execution, reducing overall task rearrangement and unnecessary robot transitions, and improving task continuity under conditions of limited job window changes.
[0034] In specific embodiments, the system determines the job mapping relationships that need to be re-evaluated based on the object changed by the disturbance event, rather than uniformly invalidating all job mapping relationships. Specifically, when the job permission time window changes, only the job mapping relationships whose expected completion time is affected by the change in the job permission time window are re-evaluated; when the stable or safe state of a candidate station changes, the job mapping relationships associated with that candidate station are treated as relationships to be re-evaluated, while the valid mapping relationships associated with other candidate stations are retained; when the 3D position or surface normal of a candidate detection patch is updated, only the job mapping relationships associated with that candidate detection patch are re-evaluated; when the safety state of the work area changes locally, only the job mapping relationships associated with the candidate stations or candidate detection patches in the corresponding affected area are re-evaluated. Different disturbance events determine the corresponding mapping relationship failure range through the candidate stations, candidate detection patches, or time execution conditions they affect, so that the reconstruction range of the work unit corresponds to the actual changed physical object; valid mapping relationships not associated with the changed object remain in their original valid state.
[0035] In a specific embodiment, when any of the following situations are detected: work order failure, change in the safety status of the work area, distance between the robot and a live conductor less than the safe distance, or instability of the current target position, the work mapping relationship corresponding to the affected detection patch in the current work unit is set to a failed state, and the robotic arm is prohibited from continuing to perform the corresponding rebound detection. For robotic arms that have already entered the work state, the robotic arm is controlled to stop the current detection and exit to a preset safe posture. The safe posture can be that the end effector of the robotic arm is far away from the component to be detected and the live risk area, and the robotic arm is kept in a retracted state suitable for stable support or subsequent movement of the robot. After the corresponding safety constraints are satisfied again, it is re-determined whether the failed mapping relationship can be restored to a valid mapping relationship, or to enter the generation process of the subsequent alternative work unit. By directly associating the valid state of the work mapping relationship with the real-time safety conditions on site, the corresponding robotic arm contact operation can be stopped in time when the safety conditions change, without having to discard all unaffected work relationships at the same time. This achieves corresponding control between safety interlocks and local task failures, improves the timeliness of robotic arm operation interruption in abnormal situations, and reduces the amount of repetitive planning after the safety state is restored.
[0036] S5: Generate alternative work units based on the incomplete detection patches and candidate stations corresponding to the failure mapping relationship, and form a subsequent recovery work sequence by combining the unaffected valid mapping relationship.
[0037] In a specific embodiment, the set of patches to be reconstructed is first determined based on the failure mapping relationship. This set includes incomplete detection patches that failed to continue execution due to changes in the job permission time window, safety constraints, or robot operation status. Then, candidate stations capable of establishing job mapping relationships with the detection patches in the set of patches to be reconstructed are re-acquired. Based on the updated job permission time window and safety constraint information, a validity assessment is performed again to determine the alternative target station, the target detection patch subset, and the effective execution interval, thereby generating alternative job units. Valid mapping relationships unaffected by the aforementioned changes are retained, and the job mapping relationships corresponding to the alternative job units, together with the unaffected valid mapping relationships, form the subsequent recovery job sequence. Therefore, it is not necessary to completely replan for completed detection patches and detection patches that still have valid job mapping relationships. By implementing local job unit reconstruction with the failure mapping relationship as the boundary, the job recovery scope can be limited to the detection patches and candidate stations that are truly affected by the disturbance. This avoids the need for traditional task-level rescheduling to cause all detection points and all stations to re-participate in the calculation, which helps to reduce the amount of computation and the range of transitions during the recovery process, while maintaining the continuity of the original effective job relationships.
[0038] In a specific embodiment, when the target position of the quadruped robot changes, the confidence level of the three-dimensional pose of the component to be detected is lower than a preset threshold, or the difference in surface normal between the visual detection result and the point cloud detection result exceeds a preset threshold, the component to be detected is re-observed before generating an alternative work unit, and the three-dimensional position and surface normal of the candidate detection patch are updated according to the re-observation result, thereby re-determining the work mapping relationship between the relevant candidate position and the candidate detection patch.
[0039] In one specific implementation, the quadruped robot can acquire image information of the component to be inspected using a surround-view camera and combine it with a LiDAR to acquire point cloud information of the component. When the reliability of the original detection results is insufficient or the visual detection results are inconsistent with the point cloud detection results, a dual-light gimbal is triggered to re-observe the corresponding component and update the 3D pose and candidate detection surface information of the component using the newly acquired detection results. By reconfirming the 3D position and surface normal of the detection surface when the position changes or the reliability of the component's pose is insufficient, the robot can avoid directly using the spatial information obtained before the interruption, which could lead to position or alignment deviations when the robot arm resumes execution. This ensures that the recovery sequence is based on the updated actual spatial state, improving the positioning and alignment reliability during the robot arm's rebound detection recovery process.
[0040] In one specific implementation, candidate detection patches have at least one detection state among ready, need_reobserve, and reject. When the wall panel detection confidence is below 0.60, the corresponding candidate detection patch is not allowed to enter the bounce execution process, and the dual-light gimbal is requested to re-observe. When the normal angle error of the candidate detection patch is greater than 5°, it is marked as need_reobserve. When the normal difference between the visual detection result and the LiDAR plane fitting result is greater than 8°, the corresponding 3D pose confidence is reduced, and re-observation is triggered first. When the available detection area of the candidate detection patch is less than 0.04 m², the candidate detection patch is marked as reject and the point placement is stopped. When the net distance between the candidate detection patch and the corner of the component to be detected is less than 50 mm, the candidate detection patch is regenerated. After re-observation or regeneration of the candidate detection patch, only the job mapping relationship associated with the candidate detection patch whose state has changed is updated, while the valid mapping relationship corresponding to other candidate detection patches is retained.
[0041] Figure 2 This illustration shows a schematic diagram of the concrete wall panel detection and candidate detection patch generation process according to a specific embodiment of this application, as follows: Figure 2As shown, the concrete wall panel detection and candidate detection patch generation process mainly includes data preprocessing, two-dimensional wall panel detection and key point regression, planar constraint screening, three-dimensional pose estimation by fusion of vision and lidar, candidate detection patch output, and detection result release.
[0042] Specifically, in Figure 2 In step 1, multi-source sensing data collected by the quadruped robot at the substation construction site is first acquired and preprocessed. This multi-source sensing data can include RGB images, LiDAR point clouds, and time-synchronized pose information corresponding to the robot's current state. The quadruped robot's onboard sensing system can simultaneously process data from multiple RGB cameras, front and rear multi-line LiDARs, dual-light gimbal images, and the robot's pose. A unified spatiotemporal reference is established through time synchronization, coordinate unification, image enhancement, and point cloud preprocessing. Therefore, in this embodiment, denoising, contrast enhancement, and dark / highlight adjustment are performed on the RGB images; downsampling, outlier removal, and ground point processing are performed on the LiDAR point clouds; and the images, point clouds, and robot pose are unified to the corresponding time and coordinate references to reduce the impact of robot motion, sensor noise, and multi-sensor time asynchrony on subsequent wall panel recognition and 3D pose estimation. The quadruped robot serves as a mobile operational platform, equipped with visual and LiDAR sensing devices. A dual-light gimbal camera can be used for long-distance target confirmation or supplementary observation in low-confidence scenarios. A LiDAR sensor provides three-dimensional geometric information about the wall panel and its surrounding space. The dual-light gimbal performs long-distance detection, high-resolution confirmation, and low-confidence re-observation functions. When the surrounding detection distance is large or the visual and LiDAR results are inconsistent, it can re-lock onto the wall panel and acquire high-resolution images.
[0043] In step 2, the preprocessed RGB image is input into the WallPanel-YOLO concrete wall panel detection model to perform two-dimensional target detection and keypoint regression on the component to be detected. Specifically, the two-dimensional detection box of the wall panel in the image and its category confidence can be obtained first, and then keypoints that can characterize the edge or planar region of the wall panel can be predicted. WallPanel-YOLO sets up a keypoint regression branch and planar confidence correlation processing on the basis of the YOLO detection framework to output the two-dimensional position of the wall panel instance and the geometric information required for further three-dimensional pose calculation. In order to verify the detection performance of the wall panel detection model in images of substation construction sites, Faster R-CNN, YOLOv8m, preprocessed YOLOv8m and WallPanel-YOLO were compared and tested on the SubStation-WallPanel test set, and mAP@0.5, mAP@0.5:0.95, Precision, Recall, F1 and single-frame inference time were used as evaluation indicators. The test results are shown in Table 1.
[0044] Table 1. Detection performance of different methods on the SubStation-WallPanel test set. As shown in Table 1, the preprocessed YOLOv8m improved its mAP@0.5 from 84.6% to 89.1% compared to the original YOLOv8m. Further application of WallPanel-YOLO resulted in mAP@0.5 reaching 91.2%, Precision reaching 91.6%, and Recall reaching 88.7%, while the single-frame inference time was 16.8 ms. This demonstrates that the detection model, while maintaining onboard real-time processing capabilities, can provide relatively stable front-end detection results for subsequent wall panel keypoint extraction, 3D pose estimation, and candidate detection patch generation.
[0045] In step 3, planar constraint filtering is performed based on the wall panel detection results obtained in step 2. Planar constraint filtering integrates the size of the candidate wall panel region, the planar normal, and the detection confidence level to filter out candidate regions that do not conform to the geometric characteristics of the concrete wall panel or are unsuitable for rebound detection. Since concrete wall panels are usually approximately planar structures, relying solely on two-dimensional color or texture features is easily affected by interference from the ground, fences, and other gray structures. Therefore, planar geometric information can be further used to verify the two-dimensional detection results. One of the important differences between the wall panel and the background lies in the planar normal and geometric continuity, requiring the combination of LiDAR point cloud planar fitting to improve the reliability of 3D recognition.
[0046] In step 4, LiDAR plane fitting and visual PnP pose estimation are performed on the candidate wall panel targets selected through planar constraints, and the two types of pose estimation results are fused. Specifically, wall panel point clouds can be extracted from the LiDAR point clouds corresponding to the two-dimensional region of the wall panel, and plane fitting is performed on these point clouds to obtain the wall panel's planar center position, planar equation, and planar normal. Simultaneously, PnP pose estimation is performed based on the wall panel's key points and visual imaging relationships to obtain the visual three-dimensional pose information. Subsequently, based on visual detection confidence, point cloud plane fitting residuals, and point cloud density, the visual PnP pose and LiDAR plane fitting results are fused to obtain the three-dimensional position and surface normal of the component to be detected relative to the quadruped robot's coordinate system. The three-dimensional pose of the wall panel is output using a fusion of visual PnP and LiDAR plane fitting, and the fusion result is determined using detection confidence, planar residuals, and point cloud density. Considering that the distance between the quadruped robot and the wall panel to be inspected at the actual construction site will change with the inspection position and traffic conditions, the wall panel detection and three-dimensional pose estimation performance under different shooting distance conditions were further tested, and the impact of dual-light gimbal supplementary sampling on long-distance detection results was statistically analyzed. The test results are shown in Table 2.
[0047] Table 2 Performance of wall panel detection and pose estimation at different shooting distances Table 2 shows that as the distance between the quadruped robot and the wall panel to be inspected increases, the recall and pose estimation accuracy of the wall panel detection decrease when using only conventional surround-view detection. For a distance range of 5–8 m, after introducing a dual-light gimbal for supplementary image acquisition, the detection recall increased from 82.3% to 88.9%, mAP@0.5 increased from 86.4% to 90.2%, the normal angle error decreased from 5.1° to 3.9°, and the center point error decreased from 0.14 m to 0.09 m. This indicates that at long distances or when the reliability of conventional surround-view results decreases, re-acquiring high-resolution images using a dual-light gimbal and re-performing wall panel detection and pose fusion can improve the 3D position and surface normal information upon which candidate detection patch generation depends. Furthermore, to verify the role of fusion of visual information and LiDAR point cloud in determining the spatial attributes of candidate detection patches, three-dimensional pose estimation was compared using only visual PnP, only LiDAR plane fitting, and fusion of visual PnP and LiDAR plane fitting. The test sample included 1120 sets of aligned samples, and the results are shown in Table 3.
[0048] Table 3. Error in 3D Pose Estimation of Wall Panel As shown in Table 3, compared with using only visual PnP or only LiDAR plane fitting, the normal angle error of the visual-LiDAR fusion method is reduced to 3.4°, the center point error is reduced to 0.07 m, the plane fitting residual is 0.012 m, and the success rate of generating usable test areas is increased to 93.5%. Therefore, by fusing visual two-dimensional detection with LiDAR three-dimensional geometric information, the estimation stability of the center position and surface normal of the component to be detected can be improved, and the success rate of generating usable test areas that meet the springback detection conditions can be increased. This provides a more reliable spatial input for judging the reachability of the robotic arm between candidate detection patches and candidate stations, as well as for determining the springback alignment status.
[0049] In a series of field verifications, tests were conducted continuously for three weeks at the construction site of a 500 kV substation renovation and expansion project, with a total of 18 concrete wall panels or column surfaces to be inspected. During the autonomous inspection by the quadruped robot, the wall panel detection rate was 100% (18 / 18), with a false detection rate of 2.8%. The generated recommended test areas were adopted by the subsequent robotic arm in 94.4% of cases. The end-to-end process of 'detecting a wall panel—approaching the component to be inspected—triggering a rebound task' was tested three times and completed successfully in all three tests, with an average time reduction of 22.4% compared to manual point selection. These results indicate that the generated 3D pose of the wall panels and candidate detection patches can be used for subsequent collaborative rebound detection by the quadruped robot and robotic arm.
[0050] Through the aforementioned two-dimensional detection and three-dimensional fusion process, the two-dimensional target of the wall panel in the image can be transformed into a three-dimensional working object that can be used by the quadruped robot and robotic arm. The subsequent steps go beyond simply knowing "where there is a wall in the image," but further obtaining the position of the wall panel in the robot's coordinate system and the orientation of the wall surface. This provides a spatial basis for determining the robotic arm's workspace, aligning the rebound spring with the wall's normal, and identifying candidate positions for the quadruped robot.
[0051] In step 5, the detection results required for rebound detection are generated and output based on the fused 3D pose of the wall panel and the effective area of the wall surface. The detection results may include a 2D detection frame, key points of the wall panel, 3D pose, and recommended test areas. The recommended test area is a region generated on the wall panel plane that can be used for rebound detection. Its output information includes at least the test area center, test area normal, and available area. In one specific embodiment, a 200 mm × 200 mm candidate test area mesh can be generated in the wall panel plane according to the rebound detection requirements, and the candidate test areas can be screened. For example, the test area should maintain a corresponding clearance from the component edge, avoid template lines, holes, or obviously contaminated areas, and prioritize areas with better surface and normal conditions. The aforementioned recommended test areas can further serve as candidate detection patches. Each candidate detection patch not only has its position on the surface of the component to be detected but also is associated with corresponding 3D coordinates, surface normal, and available detection area. Therefore, subsequent steps can use candidate detection patches as the basic operation object, rather than only the entire wall panel or a single 2D detection point as the scheduling object.
[0052] Furthermore, in one embodiment, candidate patches with insufficient available detection area, insufficient distance from the edge of the component to be detected, or unsuitable for springback detection can be eliminated before forming the candidate detection patch set. In this way, the detection patches entering the subsequent mapping relationship establishment process already possess actual springback detection value, reducing the computational workload of the robotic arm in subsequent accessibility judgments and position matching for inoperable areas.
[0053] In step 6, the wall panel detection and candidate detection patch generation results are published to the subsequent robotic arm control and scheduling module via the ROS2 interface. Figure 2 The code schematically illustrates three types of output information: `detections`, `pose`, and `test_area`. `detections` can be used to transmit the 2D detection results of the wall panel, `pose` can be used to transmit the 3D pose information of the wall panel, and `test_area` can be used to transmit recommended test areas or candidate detection patches. The corresponding interface can define fields such as the wall panel's planar center position, planar normal, pose confidence, recommended test area center, test area normal, test area size, available area, and edge clearance.
[0054] Figure 3 This application illustrates an embodiment of the collaborative operation scheduling and recovery system for concrete rebound detection using a quadruped robot at a substation construction site, as shown in the diagram. Figure 3As shown, the system includes a perception module 301, a mapping relationship construction module 302, a job unit generation module 303, an execution monitoring module 304, and a recovery module 305. The system comprises the following modules: a perception module 301, configured to acquire candidate detection patches with 3D position and surface normal, candidate positions of the quadruped robot, and the work permission time window and safety constraint information for the rebound detection task; a mapping relationship construction module 302, configured to establish a work mapping relationship between the candidate positions and candidate detection patches of the quadruped robot, and determine the validity of the work mapping relationship based on the reachability state of the robotic arm, the rebound alignment state, the candidate position state, and the work permission time window; a work unit generation module 303, configured to determine the target position and a subset of target detection patches based on the valid mapping relationship, and generate a work unit including the target position, the subset of target detection patches, and the effective execution interval; an execution monitoring module 304, configured to re-determine the validity of the affected work mapping relationship when the work permission time window, safety constraint information, or the quadruped robot's work state changes; and a recovery module 305, configured to generate alternative work units based on the incomplete detection patches and candidate positions corresponding to the failed mapping relationship, and combine them with the unaffected valid mapping relationship to form a subsequent recovery work sequence. The system can be used to execute the specific steps of the aforementioned method.
[0055] Figure 4 This diagram illustrates a specific embodiment of the scheduling process for job unit failure and recovery under disturbance conditions, representing a specific embodiment of this application. Figure 4 As shown, when the quadruped robot and robotic arm are performing concrete rebound testing, the system continuously monitors the work permit conditions, robot status, and external events so that when anomalies occur or scheduling conditions change during the testing process, the affected work units can be identified, frozen, verified, and reconstructed.
[0056] In step 401, disturbance events are monitored. Disturbance events can originate from external scheduling instructions, changes in job permission conditions, changes in robot operating status, or safety interlock signals. For example, they may include P0-level safety alarms (high-priority safety events), temporary task insertion by the platform, changes in the job permission time window, robot path obstruction, battery level below a preset threshold, and triggering of a robotic arm safety interlock. All of these disturbance events may cause changes in the execution conditions of existing job units; therefore, it is necessary to further determine whether they will affect the effective job mapping relationship between the currently established candidate positions and candidate detection patches. In some implementations, changes in the robotic arm job window, early termination of a power outage plan, path obstruction, insufficient battery power, and safety interlocks can all serve as disturbance events triggering rescheduling.
[0057] In step 402, the detected disturbance events are identified and classified. By identifying the event type, the affected objects and scope of the disturbance can be determined. For example, P0-level alarms typically correspond to high-risk events requiring immediate priority handling; platform order insertion mainly changes the priority order of subsequent tasks; a shortened job permit time window may render some inspection patches that could previously be completed within that time window incomplete; path obstruction may change the estimated time for the robot to reach a candidate station; insufficient power may limit the robot's ability to continue relocation or perform subsequent inspection tasks; and triggering of the robotic arm safety interlock may directly cause the current contact inspection operation to no longer meet the safety execution conditions. In one implementation, the system does not immediately discard all existing scheduling results due to any disturbance event. Instead, it determines the affected job units and the job mapping relationships associated with those job units based on the disturbance type. That is, it first determines whether the disturbance affects the job permit time, target station, robot arrival time, execution conditions of candidate inspection patches, or robotic arm safety execution conditions, and then determines the range of mapping relationships that need to be re-evaluated, thus providing a basis for subsequent local recovery.
[0058] In step 403, the tasks affected by the disturbance are frozen and execution breakpoints are saved. Freezing means temporarily stopping the continued execution of the affected job unit, rather than immediately deleting the original detection results or all job plans. For detection patches that have completed bounce detection, their completion status can be retained; for detection patches that have not yet been completed, their corresponding job mapping relationship, current execution status, and incomplete status are retained, so that they can participate in recovery scheduling when subsequent safety conditions or time conditions are met again. Valid job mapping relationships that are not affected by the disturbance can continue to be retained and are not recalculated along with the affected tasks. In some embodiments, execution breakpoints can at least be used to characterize the division status of completed and incomplete detection patches in the current job unit, as well as the corresponding execution progress when the disturbance occurs. In this way, after the disturbance is resolved, subsequent recovery job sequences can be directly generated based on incomplete detection patches without re-detecting the completed detection patches.
[0059] Step 404 involves safety verification and priority reordering. After freezing the affected tasks, the current job permit conditions and robot status are reacquired, and safety verification is performed on the tasks to be restored. Safety verification may include work ticket status, work area safety status, safe distance between the robot and live parts, candidate station stability status, and updated job permit time window. For any job mapping relationship that no longer meets the safety or time conditions, it is changed from a valid state to a invalid state; for job mapping relationships that still meet the conditions after reverification, they remain valid. Robotic arm operations themselves have strong safety constraints, requiring verification of the area safety status, valid work tickets, safe distances, and whether candidate stations are flat and stable before execution. Simultaneously, task processing priorities can be adjusted based on the urgency of the disturbance event. When the disturbance is a P0-level event requiring immediate handling, step 405 is entered for P0-level emergency handling judgment; when the disturbance is not a P0-level event, the subsequent job sequence can be directly re-optimized.
[0060] In step 405, it is determined whether the current disturbance requires P0-level emergency handling. If the determination result is yes, proceed to step 406, immediately switch to emergency inspection or corresponding safety handling task, and suspend the current rebound detection task or robotic arm operation task. During this process, the robotic arm can stop the current contact detection operation according to safety interlock requirements and exit to a preset safe posture, while the quadruped robot performs the corresponding task according to the emergency handling requirements. For ordinary detection and robotic arm operation, suspending when a high-priority alarm occurs and resuming after the alarm handling is completed is a typical processing method in the existing dynamic scheduling process. After the P0-level emergency task is handled, the process can be returned to the recovery scheduling process, and the previously frozen unfinished detection patches can be re-evaluated based on the updated job permission time window, safety status, and the robot's current position. Thus, even if a high-priority event temporarily interrupts the original rebound detection, the still valid detection results and job mapping relationship can be used after the handling is completed, without having to build the entire detection plan from scratch.
[0061] When step 405 determines that the current disturbance does not require P0-level emergency handling, or when the emergency task in step 406 has been handled, the process proceeds to step 407 to re-optimize subsequent tasks and generate a new task sequence π′. In this embodiment, a partial reconstruction approach can be prioritized, i.e., only the incomplete detection patches corresponding to the failed mapping relationship and their associated candidate stations are recalculated; for task mapping relationships that are not affected by the disturbance and remain valid, they are not re-involved in the complete planning. For example, when the disturbance is a shortening of the task permission time window, the expected completion time of each incomplete detection patch under the current target station can be recalculated, and only mapping relationships whose expected completion time exceeds the shortened time window are marked as failed, while valid mapping relationships that can be completed within the remaining time are retained. At this time, only the incomplete detection patches corresponding to the failed mapping relationship are added to the set to be reconstructed, and new candidate stations or new valid execution intervals are selected for these detection patches. Existing scheduling strategies also adopt the idea of minimizing disturbance by retaining tasks within the window that meet safety conditions when the window shrinks and only performing local rearrangement on affected subsequent tasks. For example, when the disturbance is path obstruction, the estimated time for the quadruped robot to reach the original candidate station may change. In this case, the corresponding job mapping relationship can be re-evaluated based on the updated estimated arrival time, stabilization time, detection execution time, and safe exit time to determine if it still falls within the current job permission time window. If not, the mapping relationship is invalidated, and other candidate stations are selected or the effective execution range of the corresponding detection patch is readjusted. When the disturbance is a robot arm safety interlock or instability of the current target station, the affected job mapping relationship related to the target station can be directly invalidated, and a candidate station that meets the stability requirements and robot arm reachability conditions can be re-selected during the subsequent recovery process. Detection patches that have already been detected remain in a completed state, thus avoiding the re-execution of completed springback detection due to station re-selection.
[0062] During the re-optimization process in step 407, if alternative work units that meet the work permission and safety conditions can be obtained through local reconstruction, there is no need to re-plan all tasks. Only when no local candidate station can establish an effective work mapping relationship with the detection patch to be reconstructed, or when the remaining work permission time cannot accommodate local alternative solutions, can the scope of re-optimization be expanded. Thus, existing effective work results can be maintained first, reducing global task reordering caused by disturbances.
[0063] In step 408, the regenerated job sequence is sent to the quadruped robot and the robotic arm, and the rebound detection is resumed or continued. The new job sequence may include existing unaffected job units and alternative job units generated based on the failure mapping relationship. For alternative job units, they may include at least a new target position, a subset of target detection patches that have an effective mapping relationship with the target position, and the corresponding effective execution range. The quadruped robot moves to the corresponding target position according to the new job sequence, and controls the robotic arm to perform subsequent rebound detection after the position stability condition is met. During the execution process, the job permission conditions and robot status are continuously monitored, thus forming a closed loop of "execution—monitoring—failure determination—partial reconstruction—resumption of execution".
[0064] In step 409, the cause of the disturbance, the scheduling changes, the estimated completion time, and the execution log are written back to the local area network platform. By writing back this information, the scheduling terminal can obtain the task changes before and after the disturbance and the restored execution status, providing a basis for subsequent manual review, task tracking, and operation recording. For different disturbances such as shortened job permit time windows, path obstruction, insufficient power, or safety interlocks, the corresponding reasons for the changes and the affected job units can also be recorded to trace the scheduling process.
[0065] In the joint debugging and testing of related systems, a total of 120 tests were conducted on dynamic allocation scenarios, including changing patrol routes, alarm queue jumping, escalation / suspension and resumption of detection tasks, and remote pause / resumption. Of these, 112 tests executed correctly and maintained consistent feedback, resulting in an overall dynamic allocation accuracy rate of 93.3%. Specifically, the detection task escalation / suspension and resumption scenario was tested 30 times, with 27 successful executions. These tests demonstrate that the execution chain based on task status monitoring, suspension, and resumption can achieve a stable closed loop on the robot platform, providing an execution foundation for further local operation recovery according to the failure mapping relationship in this embodiment.
[0066] In this embodiment, the recovery scheduling process follows the principles of safety priority and minimal disturbance. When a disturbance occurs, instead of abandoning and completely replanning all rebound detection tasks, the process first identifies the actual operation mappings affected by the disturbance, determines their failure, and retains the unaffected valid mappings. Subsequently, alternative work units are generated only for the incomplete detection patches corresponding to the failed mappings. This approach ensures the safety of the robotic arm's contact detection while reducing the impact of changes in work permit conditions, path changes, or emergencies on existing detection plans, thus improving the continuous execution capability and recovery efficiency of rebound detection tasks in dynamic construction environments.
[0067] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.
Claims
1. A collaborative operation scheduling and recovery method for concrete rebound detection using a quadruped robot at a substation construction site, characterized in that, include: S1: Obtain candidate detection patches of the component to be inspected, candidate positions of the quadruped robot, and work permission time window and safety constraint information for the rebound detection task, wherein the candidate detection patches have three-dimensional position and surface normal. S2: Establish a job mapping relationship between each candidate station and each candidate detection surface, and determine the validity of the job mapping relationship based on the reachability of the robotic arm, the rebound alignment status, the candidate station status, and the job permission time window; S3: Determine the target station and the corresponding target detection patch subset according to the effective operation mapping relationship, generate an operation unit including the target station, the target detection patch subset and the effective execution range, and control the quadruped robot and the robotic arm to execute the operation unit; S4: When changes occur in the job permission time window, safety constraint information, or the job status of the quadruped robot that cause changes in the execution conditions of the current job unit, determine the affected job mapping relationship based on the time conditions, candidate positions, or candidate detection patches affected by the changes, re-determine the validity of the affected job mapping relationship, mark the job mapping relationship that no longer meets the valid conditions as a failed mapping relationship, and retain the unaffected valid mapping relationship. S5: Generate alternative work units based on the incomplete detection patches and candidate stations corresponding to the failure mapping relationship, and form a subsequent recovery work sequence by combining the unaffected valid mapping relationship.
2. The collaborative operation scheduling and recovery method for concrete rebound detection using a quadruped robot at a substation construction site according to claim 1, characterized in that, The candidate detection patches are generated based on the three-dimensional detection results of the component to be detected. Each candidate detection patch includes at least the three-dimensional coordinates of the center of the detection patch, the surface normal, and the available detection area. Candidate detection patches whose available detection area does not meet the preset requirements, whose distance from the edge of the component to be detected does not meet the preset net distance requirements, or which contain holes or contaminated areas are rejected.
3. The collaborative operation scheduling and recovery method for concrete rebound detection using a quadruped robot at a substation construction site according to claim 1, characterized in that, The reachability state of the robotic arm in the operation mapping relationship is determined based on the reachability of the robotic arm end to the candidate detection surface under the corresponding candidate station; the rebound alignment state is determined based on the deviation between the detection direction of the robotic arm end and the surface normal of the candidate detection surface. When the deviation meets the preset alignment conditions, the corresponding candidate detection surface is marked as a spatially executable surface under the candidate station.
4. The collaborative operation scheduling and recovery method for concrete rebound detection using a quadruped robot at a substation construction site according to claim 3, characterized in that, For any spatially executable surface, based on the estimated time for the quadruped robot to reach the corresponding candidate station, the stabilization time in position, the estimated detection time for the corresponding candidate detection surface, and the safety allowance time required for the robotic arm to exit the operation state, it is determined whether the spatially executable surface can be completed within the permitted operation time window; if it can be completed, the corresponding operation mapping relationship is marked as a valid mapping relationship.
5. The collaborative operation scheduling and recovery method for concrete rebound detection using a quadruped robot at a substation construction site according to claim 4, characterized in that, For the same candidate station, the expected detection time of the candidate detection patches corresponding to the effective mapping relationship is used to occupy the job permission time window in sequence, and the executable detection patch subset corresponding to the candidate station is determined by the remaining available time of the job permission time window; the target station is determined according to the coverage of the executable detection patch subset and the movement cost of the quadruped robot to reach the candidate station.
6. The collaborative operation scheduling and recovery method for concrete rebound detection using a quadruped robot at a substation construction site according to claim 1, characterized in that, The work permit time window is determined based on the valid period of the work order; for rebound detection tasks with power outage requirements, the work permit time window is determined by the overlap of the valid period of the work order and the planned power outage window; the safety constraint information includes at least the status of the work area, the safe distance between the robot and the live conductor, and the stability status of the candidate station.
7. The collaborative operation scheduling and recovery method for concrete rebound detection using a quadruped robot at a substation construction site according to claim 1, characterized in that, When the job permission time window is shortened, only job mapping relationships whose execution completion time exceeds the shortened job permission time window are marked as invalid; job mapping relationships that are still valid under the current target station continue to be executed, and incomplete detection patches that have become invalid due to the shortening of the job permission time window are moved into the set of patches to be reconstructed.
8. The collaborative operation scheduling and recovery method for concrete rebound detection using a quadruped robot at a substation construction site according to claim 1, characterized in that, When the confidence level of the three-dimensional pose of the component to be detected is lower than a preset threshold, the normal angle error of the candidate detection patch exceeds a preset error threshold, or the difference in surface normal between the visual detection result and the point cloud detection result exceeds a preset threshold, a re-observation is performed on the corresponding component to be detected, the three-dimensional position and surface normal of the candidate detection patch are updated, and the operation mapping relationship related to the candidate detection patch is re-determined.
9. The collaborative operation scheduling and recovery method for concrete rebound detection using a quadruped robot at a substation construction site according to claim 1, characterized in that, When the work ticket is detected to be invalid, the safety status of the work area is changed, the distance between the robot and the live part is less than the safe distance, or the current target position is unstable, the work mapping relationship in the current work unit affected by the change in the safety constraint is set to invalid, and the robotic arm is controlled to stop the rebound detection and exit to the preset safe posture. Once the corresponding safety constraints are satisfied again, alternative work units are regenerated only for the incomplete detection patches corresponding to the failure mapping relationship.
10. A collaborative operation scheduling and recovery system for concrete rebound detection using a quadruped robot at a substation construction site, used to implement the method as described in any one of claims 1-9, characterized in that, include: The perception module is configured to acquire candidate detection patches with three-dimensional position and surface normal, candidate positions of the quadruped robot, and work permission time window and safety constraint information for the rebound detection task. The mapping relationship construction module is configured to establish a task mapping relationship between candidate positions and candidate detection patches for a quadruped robot, and to determine the validity of the task mapping relationship based on the reachability state of the robotic arm, the rebound alignment state, the candidate position state, and the task permission time window. The job unit generation module is configured to determine the target station location and target detection patch subset based on the effective mapping relationship, and generate a job unit that includes the target station location, target detection patch subset, and effective execution range; The execution monitoring module is configured to re-determine the validity of the affected job mapping relationship when the job permission time window, safety constraint information, or quadruped robot operation status changes. as well as The recovery module is configured to generate alternative work units based on the incomplete detection patches and candidate stations corresponding to the failure mapping relationship, and combine them with the unaffected valid mapping relationships to form a subsequent recovery work sequence.
