Robot mechanical arm path planning method and system for curtain wall cleaning

Abstract

The invention relates to the technical field of path planning, and discloses a robot mechanical arm path planning method and system for curtain wall cleaning. According to the method, collected data are fused through a depth camera, a laser radar and an inertial measurement unit, a dynamic environment topological map is constructed, and accurate environment perception is achieved. A self-adaptive grid method and an improved spiral traversal algorithm are adopted, a global traversal path covering a cleaning area is generated, the joint motion trail of the mechanical arm is optimized in a layered mode, and motion accuracy and stability are ensured. And meanwhile, the dynamic control module monitors obstacles in real time, the joint track is adjusted online through local track re-planning, it is guaranteed that cleaning operation is safe and continuous, the cleaning pressure is adjusted in real time, and it is guaranteed that the cleaning quality is stable and consistent.

Description

Technical Field

[0001] The present invention relates to the technical field of path planning, and in particular to a method and system for path planning of a robot manipulator for curtain wall cleaning. Background Art

[0002] With the widespread use of curtain walls in high-rise buildings, curtain wall cleaning has become a critical and high-risk task. Traditional manual cleaning methods are not only inefficient but also pose safety risks. In recent years, the development of robotics has provided new solutions for curtain wall cleaning. However, existing robotic arm path planning methods for curtain wall cleaning suffer from inaccurate path planning, inability to effectively handle dynamic obstacles, and unstable pressure control during the cleaning process, resulting in poor cleaning results and safety risks. Therefore, a more efficient and accurate robotic arm path planning method and system are needed to improve the automation level and cleaning quality of curtain wall cleaning. Summary of the Invention

[0003] In response to the above-mentioned technical deficiencies, the purpose of the present invention is to provide a robot arm path planning method and system for curtain wall cleaning, so as to solve the problems of inaccurate path planning, poor dynamic obstacle coping ability and unstable cleaning pressure of the curtain wall cleaning robot arm in the prior art.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for robot arm path planning for curtain wall cleaning, the method comprising: The robot uses a depth camera and lidar to synchronously collect 3D point cloud data of the curtain wall surface. The robot uses an inertial measurement unit to detect the posture offset of the end effector of the robot arm in real time, thereby building a dynamic environment topology map. Based on the topological map, the obstacle boundaries and clean area contours on the curtain wall surface are identified, the clean area is meshed using an adaptive grid method, and a global traversal path covering all clean grids is generated; For the global traversal path, the robot arm joint motion trajectory is planned through a hierarchical optimization strategy: First, a collision-free initial trajectory is generated in the joint space based on the kinematic model. Then, the operating constraints of the end effector are introduced to perform smooth optimization on the trajectory. Dynamic obstacles during the robot arm's execution are monitored in real time, and the joint trajectory is adjusted online through the local trajectory replanning module to ensure that the end effector always moves in close contact with the curtain wall surface at the preset cleaning pressure.

[0005] Preferably, in a possible implementation of the first aspect, the process of constructing the dynamic environment topology map includes: The 3D point cloud data of the curtain wall surface is collected by fusing the depth camera and the lidar, and the point cloud is subjected to noise reduction and registration processing to generate an initial environment model; Combined with the real-time detection of the manipulator end effector posture offset by the inertial measurement unit, the position error of the point cloud data is dynamically compensated; Based on the topological modeling algorithm, the compensated point cloud is divided into continuous surface units, and the boundaries of the obstacle area and the clean area are marked; By updating point cloud data and posture offsets in real time, a dynamic environment topology map including surface curvature, obstacle locations and topological relationships of clean areas is constructed.

[0006] Preferably, in a possible implementation manner of the first aspect, the grid division includes: According to the outline of the curtain wall clean area, calculate its minimum circumscribed rectangle and set the initial grid size; Dynamically adjust the grid size based on the complexity of the obstacle boundary: Adjacent homogeneous grids are integrated through grid merging algorithm to form a non-uniform grid structure; Each grid is marked with a clean status attribute to generate a traversable grid map.

[0007] Preferably, in a possible implementation of the first aspect, the process of generating the global traversal path includes: Based on the clean state properties of the grid map, an improved spiral traversal algorithm is used to generate the initial path; According to the grid size and the robot arm motion radius constraints, the path is interpolated so that all grids are continuously covered; Introducing obstacle avoidance strategies to adjust the path direction through local detours; Finally, a non-repeated traversal sequence with the grid center point as the path node is generated.

[0008] Preferably, in a possible implementation of the first aspect, the improved spiral traversal algorithm optimizes the number of path turns through dynamic weights, specifically including: Define the total turning cost function of the path:

[0009] in, is the total turning cost, For the The steering angle of the node, is the obstacle density weight of the area where the node is located, is the total number of path nodes; By minimizing Reduce energy consumption caused by frequent turning of the robotic arm.

[0010] Preferably, in a possible implementation of the first aspect, the process of planning the robot arm joint motion trajectory using the hierarchical optimization strategy includes: In the joint space layer, the global path is converted into a joint angle sequence based on the robot kinematic model, and a random sampling algorithm is used to generate the initial obstacle avoidance trajectory; In the operation constraint layer, the cleaning pressure constraint of the end effector and the normal constraint of the curtain wall surface are introduced to smoothly optimize the initial trajectory. The fifth-order polynomial interpolation is used to ensure the continuity of the velocity and acceleration of the joint motion and avoid vibration of the robotic arm.

[0011] Preferably, in a possible implementation manner of the first aspect, the optimization process of the curtain wall surface normal constraint includes: Construct the end effector pose deviation function:

[0012] in, is the normal vector of the current cleaning point of the curtain wall, is the axial vector of the end effector, is the preset fitting angle threshold; By optimizing the trajectory Approaching 0 ensures that the end effector is always perpendicular to the curtain wall surface.

[0013] Preferably, in a possible implementation of the first aspect, the optimization of the deviation function is achieved by an adaptive step-size gradient descent method: Update joint angles:

[0014] in, For the The joint angle vector of the iteration, For the The joint angle vector of the iteration, is the adaptive step size factor, , is the gradient of the deviation function with respect to the joint angle; By dynamically adjusting Balance convergence speed and stability.

[0015] Preferably, in a possible implementation manner of the first aspect, the process of adjusting the joint trajectory online by the local trajectory replanning module includes: Real-time monitoring of lidar data to identify dynamic obstacles; On the basis of preserving the global path, the local obstacle avoidance trajectory is generated by adopting the rolling horizon control strategy with the current joint state of the robot arm as the starting point; By calculating the deviation between the cleaning pressure feedback value and the preset threshold in real time, the displacement compensation of the end effector is adjusted to ensure constant fitting pressure on the curtain wall surface.

[0016] In a second aspect, the present invention provides a robot arm path planning system for curtain wall cleaning, the system comprising: The environmental modeling module uses the robot's depth camera and lidar to synchronously collect 3D point cloud data of the curtain wall surface, and uses the inertial measurement unit to detect the posture offset of the robot's end effector in real time to build a dynamic environmental topology map. A path planning module, which identifies the obstacle boundaries and clean area contours on the curtain wall surface based on the topological map, meshes the clean area using an adaptive grid method, and generates a global traversal path covering all clean grids; The trajectory planning module plans the motion trajectory of the robot arm joints based on the global traversal path through a hierarchical optimization strategy: first, a collision-free initial trajectory is generated in the joint space based on the kinematic model, and then the operating constraints of the end effector are introduced to smoothly optimize the trajectory; The dynamic control module monitors dynamic obstacles during the execution of the robotic arm in real time, and adjusts the joint trajectory online through the local trajectory replanning module to ensure that the end effector always moves in accordance with the preset cleaning pressure and moves in line with the curtain wall surface.

[0017] The beneficial effects of the present invention are as follows: the present invention constructs a dynamic environmental topology map by fusing data collected by a depth camera, lidar, and inertial measurement unit, achieving accurate environmental perception and path planning. The adaptive grid method and improved spiral traversal algorithm improve the flexibility and efficiency of path planning. A hierarchical optimization strategy and trajectory smoothing ensure the accuracy and stability of the robot arm's joint motion. The dynamic control module and local trajectory replanning mechanism effectively address dynamic obstacles, ensuring the safety and continuity of cleaning operations. Real-time monitoring and adjustment of cleaning pressure ensure the stability and consistency of cleaning quality.

[0018] Overall, the present invention improves the automation level of curtain wall cleaning, reduces labor costs and safety risks, and has significant economic and social benefits for the intelligent manufacturing equipment industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 A flow chart of a robot arm path planning method for curtain wall cleaning is provided for this application.

[0021] Figure 2 A structural diagram of a robotic arm path planning system for curtain wall cleaning is provided for this application.

[0022] Explanation of the accompanying figures: 1-environment modeling module, 2-path planning module, 3-trajectory planning module, 4-dynamic control module. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] Example 1: Figure 1 As shown, the present invention provides a robot arm path planning method for curtain wall cleaning, comprising: The depth camera and lidar mounted on the robot body synchronously collect three-dimensional point cloud data of the curtain wall surface, and the inertial measurement unit is used to detect the posture offset of the end effector of the robotic arm in real time to build a dynamic environment topology map.

[0025] In this embodiment, the depth camera carried by the robot body and the 16-line laser radar are used to synchronously collect the three-dimensional point cloud data of the curtain wall surface. frequency to obtain high-resolution color depth information, LiDAR The frequency scanning surface geometric profile is synchronized in milliseconds through hardware triggering. The collected original point cloud is first pre-processed: the radius The statistical filtering is used to remove outlier noise points, and then the dual-sensor point clouds are registered through the iterative closest point algorithm based on feature point matching to generate an initial environment model that integrates texture and geometric features.

[0026] In order to eliminate the positioning drift caused by the vibration of the robot arm, the end effector attitude offset detected by the six-axis IMU (Inertial Measurement Unit) is read in real time. The quaternion output by the IMU is converted into a rotation matrix and translation vectors , establish the transformation matrix of the current moment t relative to the initial pose The point cloud data is mapped from the robot coordinate system to the global coordinate system through coordinate transformation: , to achieve dynamic position compensation. In response to the slight deformation of the curtain wall surface, the Kalman filter is additionally introduced to fuse IMU and point cloud motion estimation to control the positioning error to within the range.

[0027] Based on point cloud compensation, a topological modeling algorithm is used to construct an environmental map: first, the point cloud is segmented into continuous surface units using a region growth clustering algorithm, and a curvature change threshold of 0.05 rad / m is used to distinguish between plane and curved areas; second, based on the normal vector mutation amount (threshold ) mark the obstacle bounding box and extract the clean area contour Polygon. Calculate curvature eigenvalues ​​for each surface element in real time 、 ,when When Surfaces are marked as cleanable.

[0028] The map update mechanism uses incremental topology maintenance: each time a new frame of point cloud data is added, Accelerate the nearest neighbor search to match existing map nodes. Add new topological nodes when the system is running. At the same time, the obstacle position coordinates are dynamically adjusted according to the latest IMU attitude offset, and a dynamic map containing three types of topological relationships is established: 1) the adjacency matrix between surface units; 2) the spatial octree index of the obstacle bounding box; 3) the B-spline parameterized description of the clean area outline. The final output map is The frequency update includes key attributes such as the center coordinates, normal vector, curvature radius and obstacle semantic labels of each surface unit.

[0029] The obstacle boundaries and clean area contours on the curtain wall surface are identified based on the topological map. The clean area is meshed using the adaptive grid method to generate a global traversal path covering all clean grids.

[0030] In this embodiment, based on the curtain wall cleaning area contour extracted from the dynamic environment topology map, its minimum circumscribed rectangle is first calculated. Based on the diameter of the end effector of the robot arm (200mm), the initial grid size is set to 300mm×300mm. Edge detection algorithm analyzes the complexity of obstacle boundaries: calculates the standard deviation of the curvature of boundary pixels ,when In areas with dense obstacles (such as the intersection of curtain wall keels), a dynamic refinement strategy is used to reduce the grid size to To improve path accuracy; in flat areas (such as the center of the glass panel), enlarge the grid to To reduce the computational load.

[0031] The grid merging algorithm uses the quadtree region growing method: first, a grid adjacency graph is established, and the surface normal vector mean and curvature variance are calculated for each grid. If the angle between the normal vectors of adjacent grids is less than If the curvature variance difference is less than 0.01, the grids are considered homogeneous and merged. The physical properties of the smallest grid are retained during the merging process, resulting in a non-uniform grid structure. Finally, each grid is labeled with a 3D status attribute: 0 (uncleaned), 1 (cleaning in progress), or 2 (cleaned). A topological map is constructed, containing the grid center coordinates, normal vectors, and adjacency relationships.

[0032] The global traversal path generation uses an improved spiral traversal algorithm: starting from the lower left corner grid of the clean area, a two-dimensional grid coordinate system is established along the curtain wall surface. The algorithm defines the total turning cost function of the path:

[0033] in For the The steering angle of the node, is the obstacle density weight of the area where the node is located ( surrounding areas The number of obstacle grids in the image). Algorithm and cost function optimization, real-time calculation of candidate directions during path expansion Incremental, preferred Minimum direction. When a dead end is detected, the backtracking mechanism is enabled to re-plan and ensure that the traversal sequence is not repeated.

[0034] For the motion constraints of the robot arm, according to the grid size and the minimum turning radius of the end effector ( ) Perform cubic B-spline interpolation: insert three control points between adjacent grid center points to make the path curvature continuous and satisfy The obstacle avoidance strategy uses a local artificial potential field: a repulsive field is established around the obstacle grid. ,in is the repulsion coefficient, is the distance to the obstacle, To affect the radius, when the resultant force on the path point exceeds the threshold, the path is offset along the repulsive force gradient direction, and the maximum offset is controlled within the grid size. within the range.

[0035] The energy consumption balance mechanism is introduced in the path optimization stage: after completing the traversal of 10 grids, the accumulated value of historical turning angles is detected. If the cumulative value exceeds , then start the path smoothing module, by inserting the transition arc (radius ) replaces the sharp angle turn, reducing the actual steering angle to the original value The resulting path node sequence is based on the grid center point and outputs a traversal instruction set containing three-dimensional coordinates, normal vectors, and expected cleaning time.

[0036] To adapt to the characteristics of curtain wall surfaces, curvature adaptation parameters are embedded in mesh properties: the principal curvature is calculated for each mesh. 、 ,when When the path is generated in these areas, additional The interpolation point density and constrains the normal vector change rate of adjacent path points , ensuring that the cleaning tool always adheres to the curved surface. Finally, the obstacle detour verification module performs collision detection: It samples along the path in the octree map and uses the end-effector envelope to detect whether it intrudes into the obstacle buffer zone. Local replanning is initiated for unqualified path segments until they pass verification.

[0037] For the global traversal path, the robot arm joint motion trajectory is planned through a hierarchical optimization strategy: first, a collision-free initial trajectory is generated in the joint space based on the kinematic model, and then the operating constraints of the end effector are introduced to smoothly optimize the trajectory.

[0038] In this embodiment, the hierarchical optimization strategy is implemented in two levels: the joint space layer and the operation constraint layer. The initial trajectory generation of the joint space layer is based on the inverse kinematics model of the six-degree-of-freedom serial manipulator. First, the grid center point sequence of the global path is converted into the Cartesian space pose sequence of the end effector, each pose contains the position coordinates and the pose expressed in Euler angles A collision-free trajectory is generated in the joint space through a random sampling algorithm: with the feasible range of the robot's joint angles as a constraint, joint state nodes are randomly sampled between adjacent path points, the end position is calculated through the kinematic solution, and collision detection is performed in real time in the octree model of the environment topology map. If the end envelope sphere or connecting rod cylinder invades the obstacle buffer, the node is removed and resampled. Finally, an initial trajectory consisting of a sequence of joint angles is generated, and the sampling period is , ensuring that the linear interpolation between trajectory points satisfies the maximum joint angular velocity constraint.

[0039] The smooth optimization of the operation constraint layer focuses on the operational requirements of the end effector, and its core consists of two physical constraints: Curtain wall surface normal constraint: To ensure that the cleaning tool fits the curtain wall surface, the end effector pose deviation function is constructed: ,in is the normal vector of the grid where the current cleaning point is located, is the axial vector of the end effector, is the preset fitting angle threshold (in this embodiment The optimization goal is to adjust the joint angle sequence so that Approaching 0.

[0040] Cleaning pressure constraint: Real-time feedback of pressure value through the end six-dimensional force sensor , and the preset threshold (20N in this embodiment) Compare and generate pressure error The error is converted into the displacement compensation of the end effector along the normal direction (Proportional coefficient ).

[0041] In order to satisfy the above constraints at the same time, the adaptive step size gradient descent method is used to iteratively optimize the joint trajectory: Gradient calculation: based on the robot Jacobian matrix , the end pose deviation Mapped to joint space. Deviation function for joint angle vector Gradient Solve by numerical differentiation:

[0042] in is a small perturbation of the joint angle (step length ).

[0043] Adaptive Update: The joint angle update formula for the iteration is:

[0044] in, For the The joint angle vector of the iteration, For the Joint angle vector of iterations, step factor Dynamically decays with the number of iterations, with an initial step size of 0.1 to ensure rapid convergence, and later the step size is reduced to below 0.01 to avoid oscillation. The iteration termination condition is or number of iterations .

[0045] Trajectory smoothness is enhanced by quintic polynomial interpolation: in the optimized joint angle sequence, a quintic polynomial function is inserted between every two adjacent path points:

[0046] coefficient Uniquely determined by boundary conditions, including the joint angles, angular velocities (set to 0), and angular accelerations ( This interpolation ensures the continuity of the velocity and acceleration of the joint movement and eliminates the high-frequency vibration of the robot arm (the amplitude is determined by the non-optimized down to ).

[0047] When the LiDAR detects a dynamic obstacle (such as a moving basket) invading the workspace of the robot arm, it generates a local obstacle avoidance trajectory based on the rolling time domain control with the current joint state as the starting point. The hierarchical optimization strategy is directly called in the replanning, but the upper limit of the number of iterations is compressed to 20 times, and the calculation cycle is controlled within At the same time, the pressure compensation Through feedforward control, it is superimposed to the desired end position in real time to ensure that the pressure fluctuation of the curtain wall surface is less than .

[0048] Dynamic obstacles during the robot arm's execution are monitored in real time, and the joint trajectory is adjusted online through the local trajectory replanning module to ensure that the end effector always moves in close contact with the curtain wall surface at the preset cleaning pressure.

[0049] In this embodiment, dynamic obstacle monitoring uses laser radar ( Scanning frequency) and depth camera ( The LiDAR captures a point cloud of moving targets within the workspace in real time. A Euclidean distance-based clustering algorithm (with a 150mm distance threshold) is used to identify the contours of dynamic obstacles and calculate their velocity vectors relative to the robot's base coordinate system. When an obstacle is detected encroaching upon the robot's safety boundary (a hemisphere with a 1.5-meter radius centered on the end effector), the local trajectory replanning module is triggered. This module first freezes the current joint states and extracts the robot's joint angles, angular velocities, and real-time pose of the end effector as initial conditions for replanning.

[0050] The local obstacle avoidance trajectory generation adopts the rolling horizon control strategy: Prediction model construction: based on the current moment As the starting point, establish a time domain window of 3 seconds in the future. Predict the end trajectory based on the robot arm dynamics model: ,in is the joint torque input vector, and the constraint condition is the joint angular velocity , angular acceleration .

[0051] Obstacle Motion Prediction: A Kalman filter is used to estimate the future motion paths of dynamic obstacles and generate a spatiotemporal occupancy grid map. The grid resolution is set to 50 mm × 50 mm × 100 milliseconds, marking the area occupied by the obstacle in each time slice.

[0052] Trajectory optimization solution: Minimize the cost function within the time domain window: ,in is the global path reference point, is the cleaning pressure deviation term, The joint space trajectory is solved by quadratic programming, and the optimization result is updated every 500 milliseconds.

[0053] Cleaning pressure consistency is ensured through closed-loop feedback control: Pressure-displacement conversion model: Six-dimensional force sensor with Frequency acquisition of actual cleaning pressure , and the preset threshold Comparison of generated errors The error is converted into the displacement compensation of the end effector along the normal direction of the curtain wall through the proportional-integral controller: .

[0054] Normal Snap Compensation: Superimposed in real time to the desired position at the end of the local trajectory. Position coordinates after compensation , and update the joint angle sequence by solving inverse kinematics.

[0055] The smoothness of the replanned trajectory is handled using the following mechanism: Joint space interpolation: Perform quintic polynomial interpolation on the optimized discrete joint angle point sequence to ensure the continuity of velocity and acceleration. The interpolation function form is: coefficient It is uniquely determined by the boundary conditions (angle, velocity, and acceleration of the start / end points).

[0056] Dynamic collision verification: During trajectory execution, real-time detection of the intrusion between the robot arm link envelope (cylindrical radius 50 mm) and the obstacle octree model. mm, immediately interrupting the trajectory and initiating an emergency retraction strategy to generate a safe retreat path in the opposite direction of the terminal velocity.

[0057] The exception handling mechanism includes three levels of response strategies: When the predicted obstacle collision time is greater than or equal to 3 seconds, only the trajectory speed curve is adjusted; When the predicted collision time with an obstacle is greater than or equal to 1 second and less than 3 seconds, a local path deflection is triggered, with a maximum offset of no more than 200 mm; When the obstacle prediction collision time is less than 1 second, an emergency stop is executed and a manual intervention request is reported. All replanning operations are completed within a 50 millisecond cycle, and the pressure fluctuation is controlled within the preset value. Within, the end positioning accuracy is maintained Finally, the dynamic topology map is updated in real time and synchronized with the robot controller to form a closed-loop motion control architecture.

[0058] Example 2: Figure 2 As shown, the present invention provides a robot arm path planning system for curtain wall cleaning, comprising: Environmental modeling module 1 uses the robot's depth camera and lidar to synchronously collect 3D point cloud data of the curtain wall surface, and uses an inertial measurement unit to detect the posture offset of the robot's end effector in real time to build a dynamic environmental topology map. Path planning module 2 identifies the obstacle boundaries and clean area contours on the curtain wall surface based on the topological map, meshes the clean area using an adaptive grid method, and generates a global traversal path covering all clean grids; Trajectory Planning Module 3 plans the robot arm joint motion trajectory based on the global traversal path through a hierarchical optimization strategy: first, a collision-free initial trajectory is generated in the joint space based on the kinematic model, and then the operating constraints of the end effector are introduced to smooth and optimize the trajectory; The dynamic control module 4 monitors dynamic obstacles during the execution of the robotic arm in real time, and adjusts the joint trajectory online through the local trajectory replanning module to ensure that the end effector always moves in accordance with the preset cleaning pressure and in line with the curtain wall surface.

[0059] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A robot arm path planning method for curtain wall cleaning, characterized in that: The method comprises: The robot uses a depth camera and lidar to synchronously collect 3D point cloud data of the curtain wall surface. The robot uses an inertial measurement unit to detect the posture offset of the end effector of the robot arm in real time, thereby building a dynamic environment topology map. Based on the topological map, the obstacle boundaries and clean area contours on the curtain wall surface are identified, the clean area is meshed using an adaptive grid method, and a global traversal path covering all clean grids is generated; For the global traversal path, the robot arm joint motion trajectory is planned through a hierarchical optimization strategy: First, a collision-free initial trajectory is generated in the joint space based on the kinematic model. Then, the operating constraints of the end effector are introduced to perform smooth optimization on the trajectory. Dynamic obstacles during the robot arm's execution are monitored in real time, and the joint trajectory is adjusted online through the local trajectory replanning module to ensure that the end effector always moves in close contact with the curtain wall surface at the preset cleaning pressure.

2. A robot arm path planning method for curtain wall cleaning according to claim 1, characterized in that: The process of constructing the dynamic environment topology map includes: The 3D point cloud data of the curtain wall surface is collected by fusing the depth camera and the lidar, and the point cloud is subjected to noise reduction and registration processing to generate an initial environment model; Combined with the real-time detection of the manipulator end effector posture offset by the inertial measurement unit, the position error of the point cloud data is dynamically compensated; Based on the topological modeling algorithm, the compensated point cloud is divided into continuous surface units, and the boundaries of the obstacle area and the clean area are marked; By updating point cloud data and posture offsets in real time, a dynamic environment topology map including surface curvature, obstacle locations and topological relationships of clean areas is constructed.

3. A robot arm path planning method for curtain wall cleaning according to claim 1, characterized in that: The grid division includes: According to the outline of the curtain wall clean area, calculate its minimum circumscribed rectangle and set the initial grid size; Dynamically adjust the grid size based on the complexity of the obstacle boundary: Adjacent homogeneous grids are integrated through grid merging algorithm to form a non-uniform grid structure; Each grid is marked with a clean status attribute to generate a traversable grid map.

4. A robot arm path planning method for curtain wall cleaning according to claim 3, characterized in that: The process of generating the global traversal path includes: Based on the clean state properties of the grid map, an improved spiral traversal algorithm is used to generate the initial path; According to the grid size and the robot arm motion radius constraints, the path is interpolated so that all grids are continuously covered; Introducing obstacle avoidance strategies to adjust the path direction through local detours; Finally, a non-repeated traversal sequence with the grid center point as the path node is generated.

5. A robot arm path planning method for curtain wall cleaning according to claim 4, characterized in that: The improved spiral traversal algorithm optimizes the number of path turns through dynamic weights, specifically including: Define the total turning cost function of the path: in, is the total turning cost, For the The steering angle of the node, is the obstacle density weight of the area where the node is located, is the total number of path nodes; By minimizing Reduce energy consumption caused by frequent turning of the robotic arm.

6. A robot arm path planning method for curtain wall cleaning according to claim 5, characterized in that: The process of planning the motion trajectory of the robot arm joints with the hierarchical optimization strategy includes: In the joint space layer, the global path is converted into a joint angle sequence based on the robot kinematic model, and a random sampling algorithm is used to generate the initial obstacle avoidance trajectory; In the operation constraint layer, the cleaning pressure constraint of the end effector and the normal constraint of the curtain wall surface are introduced to smoothly optimize the initial trajectory. The fifth-order polynomial interpolation is used to ensure the continuity of the velocity and acceleration of the joint motion and avoid vibration of the robotic arm.

7. A robot arm path planning method for curtain wall cleaning according to claim 6, characterized in that: The optimization process of the curtain wall surface normal constraint includes: Construct the end effector pose deviation function: in, is the normal vector of the current cleaning point of the curtain wall, is the axial vector of the end effector, is the preset fitting angle threshold; By optimizing the trajectory Approaching 0 ensures that the end effector is always perpendicular to the curtain wall surface.

8. A robot arm path planning method for curtain wall cleaning according to claim 7, characterized in that: The optimization of the bias function is achieved by adaptive step-size gradient descent: Update joint angles: in, For the The joint angle vector of the iteration, For the The joint angle vector of the iteration, is the adaptive step size factor, , is the gradient of the deviation function with respect to the joint angle; Through dynamic adjustment Balance convergence speed and stability.

9. A robot arm path planning method for curtain wall cleaning according to claim 1, characterized in that: The process of online adjustment of joint trajectories by the local trajectory replanning module includes: Real-time monitoring of lidar data to identify dynamic obstacles; On the basis of preserving the global path, the local obstacle avoidance trajectory is generated by adopting the rolling horizon control strategy with the current joint state of the robot arm as the starting point; By calculating the deviation between the cleaning pressure feedback value and the preset threshold in real time, the displacement compensation of the end effector is adjusted to ensure constant fitting pressure on the curtain wall surface.

10. A robot arm path planning system for curtain wall cleaning, characterized in that: The system comprises: The environmental modeling module uses the robot's depth camera and lidar to synchronously collect 3D point cloud data of the curtain wall surface, and uses the inertial measurement unit to detect the posture offset of the robot's end effector in real time to build a dynamic environmental topology map. A path planning module, which identifies the obstacle boundaries and clean area contours on the curtain wall surface based on the topological map, meshes the clean area using an adaptive grid method, and generates a global traversal path covering all clean grids; The trajectory planning module plans the motion trajectory of the robot arm joints based on the global traversal path through a hierarchical optimization strategy: first, a collision-free initial trajectory is generated in the joint space based on the kinematic model, and then the operating constraints of the end effector are introduced to smoothly optimize the trajectory; The dynamic control module monitors dynamic obstacles during the execution of the robotic arm in real time, and adjusts the joint trajectory online through the local trajectory replanning module to ensure that the end effector always moves in accordance with the preset cleaning pressure and moves in line with the curtain wall surface.

Images