Positioning and compliant operation methods for construction wall inspection robots

Abstract

This invention relates to a positioning and compliant operation method for a construction wall inspection robot, applicable to a robot system including a mobile chassis, a robotic arm, a LiDAR, an inertial measurement unit, and a depth camera for identifying target areas and acquiring local point clouds. The method includes the following steps: S1, degradation perception positioning based on LiDAR-inertial fusion; S2, vision-triggered local operation mode switching and task coordinate system construction; S3, chassis-robotic arm collaborative alignment in the local task coordinate system; S4, end-effector attitude alignment, feedforward compensation, and compliant contact based on the wall normal; S5, performing wall contact detection according to a preset detection sequence. This application does not use a single fixed absolute threshold to hard-cut the degradation direction, but rather uses relative constraint strength and continuous weights to progressively suppress the degradation direction, avoiding pose jumps caused by traditional fixed threshold hard cutting, and improving the positioning continuity and stability in construction weak geometry scenarios.

Description

Technical Field

[0001] This invention relates to the field of construction robot technology, and specifically to a positioning and compliant operation method for a construction wall inspection robot. Background Technology

[0002] Tasks such as wall inspection, concrete quality inspection, and defect verification at construction sites typically require the testing device to maintain stable contact with the wall surface and ensure consistency in the testing point location, orientation, and contact state.

[0003] Traditional manual inspection relies on experience, is labor-intensive and inefficient, and cannot meet the requirements for high consistency inspection. On the other hand, when using a mobile chassis to carry a robotic arm and end-effector for automated inspection, problems such as unstable lighting at the construction site, weak wall texture, long corridors and repetitive structures are encountered, which makes visual positioning prone to failure. Traditional laser point cloud registration is prone to longitudinal drift, pose jump and trajectory tracking errors in scenarios with degraded geometric features. At the same time, wall contact inspection is also affected by residual errors of chassis docking, robotic arm extension deflection and end-effector contact reaction force. If global positioning, visual recognition, robotic arm alignment and compliant contact control are disconnected, it is difficult to ensure the continuity and stability of the entire process of inspection, positioning, alignment and inspection.

[0004] Therefore, there is an urgent need for a positioning and compliant operation method for construction wall inspection robots to improve the stability of global inspection positioning, the accuracy of local visual alignment, and the reliability of end-point contact detection. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a positioning and compliant operation method for a construction wall inspection robot, so as to improve the stability of global inspection positioning, local visual alignment accuracy and end-point contact detection reliability.

[0006] To achieve the above objectives, this invention provides a positioning and compliant operation method for a construction wall inspection robot, applicable to a robot system including a mobile chassis, a robotic arm, a lidar, an inertial measurement unit, and a depth camera for identifying target areas and acquiring local point clouds; the method includes the following steps: S1. In the global inspection phase, the laser point cloud of the current frame is matched with the local map and a normal equation is constructed. The relative constraint strength of each feature direction is calculated by performing eigenvalue decomposition on the coefficient matrix of the normal equation. Continuous degradation weights are constructed and the pose increment is corrected by using the projection matrix to suppress pose drift in the geometric degradation environment. S2. When the depth camera identifies the target area and the distance between the robot and the target area is less than the preset threshold, the robot switches from the global inspection mode to the local operation mode and constructs a local task coordinate system with the target area as the origin and the wall normal, horizontal direction and height direction as references. S3. Decompose the alignment error in the local task coordinate system into components in different directions, and have the moving chassis and robotic arm work together to compensate for the components in different directions. S4. Estimate the local normal vector of the wall based on the local point cloud obtained by the depth camera and perform end pose alignment. Perform configuration-related feedforward compensation on the ideal target point before solving the inverse kinematics, and perform compliant contact control when the end contacts the wall. S5. Perform wall contact detection according to the preset detection sequence. After completing all detection points in a target area, exit the local operation mode and resume the global inspection mode.

[0007] Furthermore, step S1 specifically includes: During the robot's inspection along the construction wall, the current frame laser point cloud data and inertial measurement data are collected, the current frame point cloud is matched with the local map, and the local map matching residual function is constructed based on geometric constraints. Linearizing the residual function yields the normal equation: HΔx = b ; in H This is the coefficient matrix of the normal equation; Δx The original pose increment to be solved; b The right-hand term; For the coefficient matrix of the normal equation H Perform eigenvalue decomposition: H =VΛV T ; Obtain the eigenvalues ​​corresponding to each feature direction. l i and eigenvectors v i Where Λ = diag( l 1, l 2,…, l 6), is a diagonal matrix of eigenvalues ​​arranged in descending order; V=[ v 1, v 2, …, v 6] represents the corresponding eigenvector matrix; Based on the largest eigenvalue l max Calculate the relative constraint strength in each characteristic direction: r i = l i / l max ; in l max=max( l 1, l 2,…, l 6); Introducing an absolute degradation threshold t abs Weak degradation threshold or low and reliable constraint threshold or high Construct a continuously degenerate weight function w i And generate a weight diagonal matrix: W=diag( w 1, w 2,…, w 6); And further construct the feature subspace projection matrix: P=VWV T ; Increment of the original pose using the projection matrix Δx After correction, the pose increment after degradation suppression is obtained: Δx′ =P Δx ; It is also used to update the robot's current pose.

[0008] Furthermore, the continuously degenerate weight function w i Construct it as follows: When relative constraint strength r i Less than or equal to the weak degradation threshold or low At that time, weight w i A value of 0 indicates complete suppression of pose updates in that direction. When relative constraint strength r i Greater than or equal to the reliability constraint threshold or high At that time, weight w i Set to 1 to preserve the original pose update in that direction; When relative constraint strength r i At the weak degradation threshold or low With reliable constraint threshold or high Between, weight w i Between 0 and 1, the pose update in that direction is proportionally decayed.

[0009] Furthermore, step S2 specifically includes: During the global inspection, the depth camera continuously acquires images of the wall. When a target area on the wall is identified, the position of the target area in the camera coordinate system is calculated. By combining the robot's current pose, the robotic arm's pose, and the camera's extrinsic parameters, the target area is registered in the global map coordinate system; When the target area meets the visual recognition confidence requirement and the distance between the robot and the target area is less than the preset distance threshold. d switch At that time, the system switches from global inspection mode to local operation mode. After the robot completes docking, the local task coordinate system is constructed using the centroid of the target area, the center of the detection frame, or the intersection of the normal directions.

[0010] Furthermore, step S3 specifically includes: Let the center of the target area measured by the vision system be... p obj The center point of the end tool is p tcp The alignment error in the local task coordinate system is: ; in This is the transformation matrix from the global map coordinate system to the local task coordinate system; Alignment error e ltf Decompose into horizontal error e x Vertical error e z and normal distance error e n ; Among them, horizontal error e x The mobile chassis performs micro-motion compensation along the inspection direction, reducing vertical error. e z The normal distance error is compensated by the lifting and lowering of the robotic arm end effector. e n Used to control the end detection device to approach the target area along the normal direction of the wall.

[0011] Furthermore, the end-effector attitude alignment in step S4 includes: In local operation mode, the depth camera acquires local point clouds near the target area, and estimates the local normal vector of the wall based on the neighborhood point cloud, specifically including: Let the local neighborhood point set be Its mean is Then the covariance matrix is: ; Take the covariance matrix C The eigenvector corresponding to the smallest eigenvalue is used as the local normal vector of the wall. n ; Let the approach direction of the end-effector's center point be... v tcp Calculate the end-effector attitude error: ; when i Greater than the preset attitude error threshold i max At that time, the robotic arm prioritizes end-effector posture adjustment to align the approach direction of the end-effector detection device with the normal direction of the wall; when i Less than or equal to i max At that time, the robotic arm performs an approach motion along the normal direction.

[0012] Furthermore, the feedforward compensation in step S4 specifically includes: Let the ideal target point for the task be calculated by the vision system. P ideal The configuration-related end offset compensation vector is D ff (q,l,m If the target point is input into the inverse kinematics solver, then: P target = P ideal - D ff (q,l,m ); in q The current joint angle of the robotic arm. l The length of the end extension. m For the end load parameters; using the compensated target point P target Solve for the target joint configuration of the robotic arm.

[0013] Furthermore, the compliant contact control in step S4 specifically includes: When the end-of-line detection device approaches the target point along the normal direction of the wall, and the distance between the end-of-line device and the wall is... d Less than the contact trigger threshold d contact At the same time, the equivalent stiffness in the normal contact direction is reduced, and an upper limit is set on the joint current at the far end of the robotic arm, so that the end-effector can form an approximate admittance compliant response without an independent six-dimensional force sensor. By reducing the normal direction stiffness K n and limit distal joint current Ij This ensures that the joint output torque satisfies: t j ≤ k t I max , d < d contact ; in k t The torque constant of the motor. I max The upper limit of current set for the contact phase.

[0014] Furthermore, in step S5: After each detection point is completed, the robot records the position, end-effector posture, contact status, and detection result of the detection point in the global map coordinate system and the local task coordinate system, and updates the detection status of the target area. When all detection points in the target area have been detected, the robot exits the local operation mode and resumes the global inspection mode.

[0015] Furthermore, the robot system also includes an end-effector contact detection device and a control computing unit; The mobile chassis is used to move along the construction wall or a preset inspection path, the robotic arm is used to adjust the detection points of the end-effector on the wall, the depth camera is used to identify the target area on the wall and acquire local point clouds, and the lidar and inertial measurement unit are used to realize robot pose estimation.

[0016] Compared with the prior art, the present invention has the following advantages: 1. Improve positioning stability in geometrically degraded environments: By using a progressive degradation processing method based on dual threshold weights of relative eigenvalues, instead of using a single fixed absolute threshold to hard-cut the degradation direction, the degradation direction is progressively suppressed by relative constraint strength and continuous weights. This avoids pose jumps caused by traditional fixed threshold hard cutting and improves the continuity and stability of positioning in long corridors, rough walls and repetitive structures on construction sites.

[0017] 2. Reduce the difficulty of chassis-robotic arm collaborative control: By constructing a local task coordinate system and decoupling the direction of alignment error, the horizontal error is compensated by the moving chassis, the vertical error is compensated by the robotic arm, and the normal error is used to control the approach motion, avoiding the need to compensate all residual errors by a single actuator.

[0018] 3. Improve end-effector detection accuracy and contact reliability: Ensure the perpendicularity of the detection device to the wall by estimating the wall normal and aligning the end-effector attitude; eliminate systematic deviations caused by the extension deflection of the robotic arm by configuration-related feedforward compensation; reduce contact impact by approximate admittance compliant contact control; and improve the consistency of detection points. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the operation process of the positioning and compliant operation method for the construction wall inspection robot of the present invention; Figure 2 This is a schematic diagram of the serpentine path detection in the positioning and compliant operation method of the construction wall inspection robot of the present invention. Detailed Implementation

[0020] Terminology Explanation: Current frame laser point cloud: refers to a frame of three-dimensional point cloud data collected by the lidar at the current moment.

[0021] Local map: refers to a 3D point cloud map of the local environment around the robot, which is built in real time based on lidar and inertial measurement unit. It is used to match with the lidar point cloud of the current frame to estimate the robot pose.

[0022] Geometrically degraded environment: refers to scenarios in the construction site that result in insufficient laser point cloud registration constraints, including long corridors, unfinished walls, and repetitive structures.

[0023] Continuous degradation weight: refers to a weight function constructed based on relative constraint strength and preset threshold, with a value range of [0,1]. It is used to progressively adjust the pose update gain in each feature direction, which is different from the traditional fixed threshold hard cut method.

[0024] Local task coordinate system: refers to a local reference coordinate system constructed with the target area of ​​the wall as the origin, with the wall normal direction, the wall horizontal direction, and the wall height direction as the three coordinate axes.

[0025] Feedforward compensation: refers to the amount of end-effector position offset compensation calculated in advance based on the current joint configuration, end-effector extension length and load parameters of the robotic arm, used to eliminate systematic positioning errors caused by structural deflection of the robotic arm.

[0026] Approximate admittance compliant response: refers to the compliant characteristics formed by reducing normal contact stiffness and limiting the upper limit of distal joint current in the absence of an independent six-dimensional force sensor, enabling the end-sensing device to absorb impact energy when in contact with a wall. The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0027] This invention relates to a positioning and compliant operation method for a construction wall inspection robot, which is based on a construction wall contact inspection robot system. The system includes: a mobile chassis, a robotic arm, an end-effector contact inspection device, a lidar, an inertial measurement unit, a depth camera, and a control and computing unit. The mobile chassis is used to move along the construction wall or a preset inspection path, the robotic arm is used to adjust the inspection point position of the end-effector on the wall, the depth camera is used to identify the target area on the wall and acquire local point clouds, and the lidar and inertial measurement unit are used to realize robot pose estimation.

[0028] Please refer to the appendix. Figure 1 As shown, the overall workflow of this invention is as follows: S1. Degradation sensing and localization based on laser inertial fusion.

[0029] The robot performs path planning and trajectory tracking according to a preset route or a construction site map. During the robot's inspection along the construction wall, it collects laser point cloud data and inertial measurement data for the current frame, matches the current frame point cloud with the local map, and constructs a local map matching residual function based on point-to-line, point-to-surface, or surface-to-surface geometric constraints. ; Where X represents the robot's current pose state. D X is the pose increment to be solved, r m For the first m Point cloud matching residuals; linearization of the residual function yields the normal equation: HΔx = b ; in H= J T J is the coefficient matrix of the normal equation; J is the Jacobian matrix of the residual function with respect to the pose state variables; Δx The original pose increment to be solved; b =﹣J T r is the right-hand term constructed from the current residual; matrix H can characterize the local constraint capability of the current point cloud observation on the pose state; For the coefficient matrix of the normal equation H Perform eigenvalue decomposition: H =VΛV T ; Obtain the eigenvalues ​​corresponding to each feature direction. l i and eigenvectors v i Where Λ = diag( l 1, l 2,…, l 6), is a diagonal matrix of eigenvalues ​​arranged in descending order; V=[v 1, v 2, …, v 6] represents the corresponding eigenvector matrix; Based on the largest eigenvalue l max Calculate the relative constraint strength in each characteristic direction: r i = l i / l max ; in l max =max( l 1, l 2,…, l 6); Introducing an absolute degradation threshold t abs Weak degradation threshold or low and reliable constraint threshold or high Construct a continuously degenerate weight function w i ; when r i ≤ or low or l i < t abs hour, w i =0; when r i ≥ or high hour, w i =1; when or low < r i < or high hour, w i It decays proportionally between 0 and 1, specifically as follows: ; Construct the weight diagonal matrix: W=diag( w 1, w 2,…, w 6); and the eigenspace projection matrix: P=VWV T ; Increment of the original pose using the projection matrix Δx Make corrections to obtain Δx′ =P Δx and use Δx′ Update the robot's current pose.

[0030] By employing the above methods, effective pose updates are preserved when geometric constraints are sufficient, unreliable directions are smoothly attenuated in environments with weak geometric constraints, and extremely weak constraint directions are suppressed in environments with severe degradation, thereby improving the positioning stability in long corridors, unfinished walls, and repetitive structural environments on construction sites. S2, Visual triggering of local operation mode switching and task coordinate system construction.

[0031] During the robot's global inspection, the depth camera continuously acquires images of the wall. When the vision module detects a target area on the wall, it calculates the position of the target area in the camera coordinate system and, combined with the robot's current pose, the robotic arm pose, and the camera's extrinsic parameters, registers the target area in the global map coordinate system.

[0032] Set time t The homogeneous coordinates of the detected target point in the camera coordinate system are Then its coordinates in the global map coordinate system can be expressed as: ; in, This refers to the real-time pose of the mobile chassis in the global map coordinate system. This refers to the pose transformation of the robotic arm's end effector relative to the chassis base. The camera extrinsic parameter matrix obtained through hand-eye calibration; When the target area meets the visual recognition confidence requirement and the distance between the robot and the target area is less than the preset distance threshold. d switch At this time, the robot switches from global inspection mode to local operation mode and performs deceleration, docking and local alignment processes; After the robot completes docking, a local task coordinate system is constructed with the centroid of the target area, the center of the detection box, or the intersection of the normals of the target area as the origin. This local task coordinate system uses the wall normal direction, the wall horizontal direction, and the wall height direction as the main reference directions to decouple the subsequent alignment errors. S3, Chassis-robotic arm collaborative alignment in local task coordinate system.

[0033] Let the center of the target area measured by the vision system be... p obj The center point of the end tool is p tcp The alignment error in the local task coordinate system is: ; in, This is the transformation matrix from the global map coordinate system to the local task coordinate system; Alignment error e ltf Decompose into horizontal error e x Vertical error e z and normal distance error e n ; Among them, horizontal error e x The mobile chassis performs small-range micro-motion compensation along the inspection direction; vertical error e z Compensation is provided by the lifting and lowering of the robotic arm end effector; normal distance error. e n Used to control the end detection device to approach the target area along the normal direction of the wall.

[0034] This error decoupling method avoids having all residual errors compensated by a single actuator, reduces the difficulty of multi-degree-of-freedom coupling control between the mobile chassis and the robotic arm, and improves the local alignment accuracy of the end-effector relative to the target area on the wall. S4. End-point attitude alignment, feedforward compensation, and compliant contact based on wall normal.

[0035] In local operation mode, the depth camera acquires a local point cloud near the target area; let the local neighborhood point set be... Its mean is Then the covariance matrix is: ; Take the covariance matrix C The eigenvector corresponding to the smallest eigenvalue is used as the local normal vector of the wall. n Let the approach direction of the end-effector's center point be... v tcp Calculate the end-effector attitude error: ; when i > i max When, the robotic arm prioritizes end-effector posture adjustment; when i ≤ i max At that time, the robotic arm performs an approach motion along the normal direction; Considering the systematic bias of the robotic arm in its extended state, let the ideal target point calculated by the vision system be... P idealThe configuration-related end offset compensation vector is D ff (q,l,m If the target point is input into the inverse kinematics solver, then: P target = P ideal - D ff (q,l,m ); in q The current joint angle of the robotic arm. l The length of the end extension. m For the end load parameters; using the compensated target point P target Solve for the target joint configuration of the robotic arm; When the end-of-line detection device approaches the target point along the normal direction of the wall, and the distance between the end-of-line device and the wall is... d Less than the contact trigger threshold d contact At the same time, the equivalent stiffness in the normal contact direction is reduced, and an upper limit is set on the joint current at the far end of the robotic arm, so that the end-effector can form an approximate admittance compliant response without an independent six-dimensional force sensor. By reducing the normal direction stiffness K n and limit distal joint current I j This ensures that the joint output torque satisfies: t j ≤ k t I max , d < d contact ; in k t The torque constant of the motor. I max The upper limit of current set for the contact phase.

[0036] In this way, the end detection device can absorb some of the impact energy when it comes into contact with the wall, reduce the disturbance of the contact reaction force on the moving chassis, and reduce the offset of the detection point. S5. Perform wall contact detection according to the preset detection sequence.

[0037] After completing local alignment and compliant contact, the system proceeds according to a preset detection sequence (such as...). Figure 2The serpentine path shown performs contact detection at multiple detection points. After each detection point is completed, the system records the position, end pose, contact status, and detection result of the detection point in the global map coordinate system and the local task coordinate system, and updates the detection status of the target area.

[0038] Once all inspection points within the target area have completed their inspections, the robot exits the local operation mode, resumes the global inspection mode, and continues to perform inspection and testing tasks for the next target area. The present invention has been described in detail above with reference to the accompanying drawings and embodiments. Those skilled in the art can make various modifications to the present invention based on the above description. Therefore, certain details in the embodiments should not be construed as limiting the present invention, and the scope of protection of the present invention shall be defined by the appended claims.

Claims

1. A positioning and compliant operation method for a construction wall inspection robot, applied to a robot system including a mobile chassis, a robotic arm, a lidar, an inertial measurement unit, and a depth camera for identifying target areas and acquiring local point clouds; characterized in that, The method includes the following steps: S1. In the global inspection phase, the laser point cloud of the current frame is matched with the local map and a normal equation is constructed. The relative constraint strength of each feature direction is calculated by performing eigenvalue decomposition on the coefficient matrix of the normal equation. Continuous degradation weights are constructed and the pose increment is corrected by using the projection matrix to suppress pose drift in the geometric degradation environment. S2. When the depth camera identifies the target area and the distance between the robot and the target area is less than the preset threshold, the robot switches from the global inspection mode to the local operation mode and constructs a local task coordinate system with the target area as the origin and the wall normal, horizontal direction and height direction as references. S3. Decompose the alignment error in the local task coordinate system into components in different directions, and have the moving chassis and robotic arm work together to compensate for the components in different directions. S4. Estimate the local normal vector of the wall based on the local point cloud obtained by the depth camera and perform end pose alignment. Perform configuration-related feedforward compensation on the ideal target point before solving the inverse kinematics, and perform compliant contact control when the end contacts the wall. S5. Perform wall contact detection according to the preset detection sequence. After completing all detection points in a target area, exit the local operation mode and resume the global inspection mode.

2. The positioning and compliant operation method for a construction wall inspection robot according to claim 1, characterized in that, Step S1 specifically includes: During the robot's inspection along the construction wall, the current frame laser point cloud data and inertial measurement data are collected, the current frame point cloud is matched with the local map, and the local map matching residual function is constructed based on geometric constraints. Linearizing the residual function yields the normal equation: HΔx = b ； in H This is the coefficient matrix of the normal equation; Δx The original pose increment to be solved; b The right-hand term; For the coefficient matrix of the normal equation H Perform eigenvalue decomposition: H =VΛV T ; Obtain the eigenvalues ​​corresponding to each feature direction. λ i and eigenvectors v i Where Λ = diag( λ 1, λ 2,…, λ 6), is a diagonal matrix of eigenvalues ​​arranged in descending order; V=[ v 1, v 2, …, v 6] represents the corresponding eigenvector matrix; Based on the largest eigenvalue λ max Calculate the relative constraint strength in each characteristic direction: r i = λ i / λ max ； in λ max =max( λ 1, λ 2,…, λ 6); Introducing an absolute degradation threshold τ abs Weak degradation threshold η low and reliable constraint threshold η high Construct a continuously degenerate weight function w i And generate a weight diagonal matrix: W=diag( w 1, w 2,…, w 6); And further construct the feature subspace projection matrix: P=VWV T ; Increment of the original pose using the projection matrix Δx After correction, the pose increment after degradation suppression is obtained: Δx′ =P Δx ； It is also used to update the robot's current pose.

3. The positioning and compliant operation method for a construction wall inspection robot according to claim 2, characterized in that, Continuous degradation weight function w i Construct it as follows: When relative constraint strength r i Less than or equal to the weak degradation threshold η low At that time, weight w i A value of 0 indicates complete suppression of pose updates in that direction. When relative constraint strength r i Greater than or equal to the reliability constraint threshold η high At that time, weight w i Set to 1 to preserve the original pose update in that direction; When relative constraint strength r i At the weak degradation threshold η low With reliable constraint threshold η high Between, weight w i Between 0 and 1, the pose update in that direction is proportionally decayed.

4. The positioning and compliant operation method for a construction wall inspection robot according to claim 1, characterized in that, Step S2 specifically includes: During the global inspection, the depth camera continuously acquires images of the wall. When a target area on the wall is identified, the position of the target area in the camera coordinate system is calculated. By combining the robot's current pose, the robotic arm's pose, and the camera's extrinsic parameters, the target area is registered in the global map coordinate system; When the target area meets the visual recognition confidence requirement and the distance between the robot and the target area is less than the preset distance threshold. d switch At that time, the system switches from global inspection mode to local operation mode. After the robot completes docking, the local task coordinate system is constructed using the centroid of the target area, the center of the detection frame, or the intersection of the normal directions.

5. The positioning and compliant operation method for a construction wall inspection robot according to claim 1, characterized in that, Step S3 specifically includes: Let the center of the target area measured by the vision system be... p obj The center point of the end tool is p tcp The alignment error in the local task coordinate system is: ； in This is the transformation matrix from the global map coordinate system to the local task coordinate system; Alignment error e ltf Decompose into horizontal error e x Vertical error e z and normal distance error e n ; Among them, horizontal error e x The mobile chassis performs micro-motion compensation along the inspection direction, reducing vertical error. e z The normal distance error is compensated by the lifting and lowering of the robotic arm end effector. e n Used to control the end detection device to approach the target area along the normal direction of the wall.

6. The positioning and compliant operation method for a construction wall inspection robot according to claim 1, characterized in that, Step S4, end-effector attitude alignment, includes: In local operation mode, the depth camera acquires local point clouds near the target area, and estimates the local normal vector of the wall based on the neighborhood point cloud, specifically including: Let the local neighborhood point set be Its mean is Then the covariance matrix is: ； Take the covariance matrix C The eigenvector corresponding to the smallest eigenvalue is used as the local normal vector of the wall. n ; Let the approach direction of the end-effector's center point be... v tcp Calculate the end-effector attitude error: ； when θ Greater than the preset attitude error threshold θ max At that time, the robotic arm prioritizes end-effector posture adjustment to align the approach direction of the end-effector detection device with the normal direction of the wall; when θ Less than or equal to θ max At that time, the robotic arm performs an approach motion along the normal direction.

7. The positioning and compliant operation method for a construction wall inspection robot according to claim 1, characterized in that, The feedforward compensation in step S4 is specifically as follows: Let the ideal target point for the task be calculated by the vision system. P ideal The configuration-related end offset compensation vector is Δ ff (q,l,m If the target point is input into the inverse kinematics solver, then: P target = P ideal － Δ ff (q,l,m )； in q The current joint angle of the robotic arm. l The length of the end extension. m For the end load parameters; using the compensated target point P target Solve for the target joint configuration of the robotic arm.

8. The positioning and compliant operation method for a construction wall inspection robot according to claim 1, characterized in that, The compliant contact control in step S4 is specifically as follows: When the end-of-line detection device approaches the target point along the normal direction of the wall, and the distance between the end-of-line device and the wall is... d Less than the contact trigger threshold d contact At the same time, the equivalent stiffness in the normal contact direction is reduced, and an upper limit is set on the joint current at the far end of the robotic arm, so that the end-effector can form an approximate admittance compliant response without an independent six-dimensional force sensor. By reducing the normal direction stiffness K n and limit distal joint current I j This ensures that the joint output torque satisfies: τ j ≤ k t I max , d < d contact ； in k t The torque constant of the motor. I max The upper limit of current set for the contact phase.

9. The positioning and compliant operation method for a construction wall inspection robot according to claim 1, characterized in that, In step S5: After each detection point is completed, the robot records the position, end-effector posture, contact status, and detection result of the detection point in the global map coordinate system and the local task coordinate system, and updates the detection status of the target area. When all detection points in the target area have been detected, the robot exits the local operation mode and resumes the global inspection mode.

10. The positioning and compliant operation method for a construction wall inspection robot according to claim 1, characterized in that, The robot system also includes an end-effector contact detection device and a control computing unit; The mobile chassis is used to move along the construction wall or a preset inspection path, the robotic arm is used to adjust the detection points of the end-effector on the wall, the depth camera is used to identify the target area on the wall and acquire local point clouds, and the lidar and inertial measurement unit are used to realize robot pose estimation.

Images