Vehicle cleaning control method, device, system, equipment, storage medium and program product

Abstract

The application provides a vehicle cleaning control method, device, system, equipment, storage medium and program product, and relates to the field of vehicle washing control. The method comprises the following steps: acquiring first point cloud data of the outer surface of a vehicle; determining a first confidence field according to the first point cloud data; generating a first cleaning track according to the first confidence field, and controlling a first mechanical arm to clean the vehicle based on the first cleaning track; performing at least one round of scanning on a to-be-updated area with a confidence less than a first confidence threshold in the first confidence field through a point cloud acquisition device arranged on a second mechanical arm, and updating the first cleaning track based on the scanned second point cloud data to obtain a second cleaning track; and controlling the first mechanical arm to clean the vehicle based on the second cleaning track. In this way, the reliability of cleaning can be improved by reducing missed washing and repeated operations.

Description

Technical Field

[0001] This application relates to the field of vehicle cleaning control, and more particularly to a vehicle cleaning control method, device, system, equipment, storage medium, and program product. Background Technology

[0002] With the rapid development of unmanned car cleaning technology, automated car wash equipment with multiple robotic arms working together has been widely used.

[0003] Currently, dual-arm car wash equipment typically configures both robotic arms as cleaning execution units, completing cleaning through a zoned collaborative approach. In cases where the initial point cloud data contains gaps, point cloud completion technology using algorithm fitting is commonly used to fill in the missing areas, and then a global cleaning path is planned based on the completed point cloud data.

[0004] However, in the above scheme, the fitted point cloud completion result deviates from the actual vehicle body outline, which can easily lead to inaccurate cleaning path planning and cause problems such as missed washing and empty washing. Summary of the Invention

[0005] This application provides a vehicle cleaning control method, device, system, equipment, storage medium, and program product, which can reduce problems such as missed washing and empty washing, and improve the reliability of car wash equipment in cleaning vehicles.

[0006] In a first aspect, this application provides a vehicle cleaning control method applied to a car wash equipment, the car wash equipment including a first robotic arm and a second robotic arm. The method includes: acquiring first point cloud data of the vehicle's outer surface; determining a first confidence field based on the first point cloud data; generating a first cleaning trajectory based on the first confidence field, and controlling the first robotic arm to clean the vehicle based on the first cleaning trajectory; scanning at least once the areas to be updated in the first confidence field with confidence levels less than a first confidence threshold using a point cloud acquisition device mounted on the second robotic arm, and updating the first cleaning trajectory based on the scanned second point cloud data to obtain a second cleaning trajectory; and controlling the first robotic arm to clean the vehicle based on the second cleaning trajectory.

[0007] In one possible embodiment, a point cloud acquisition device mounted on a second robotic arm performs at least one round of scanning on the regions to be updated in the first confidence field where the confidence level is less than the first confidence threshold. Based on the scanned second point cloud data, the first cleaning trajectory is updated to obtain a second cleaning trajectory. This includes: for each round of scanning, updating the confidence field obtained in the previous round of scanning based on the second point cloud data acquired during the scan to obtain a second confidence field; determining the cleaning path corresponding to a first region in the second confidence field within the regions to be updated where the confidence level is greater than or equal to the first confidence threshold; fusing the cleaning path corresponding to the first region with the first cleaning trajectory to generate a second cleaning trajectory; and identifying regions in the second confidence field within the regions to be updated where the confidence level is less than the first confidence threshold as new regions to be updated.

[0008] In one possible embodiment, the method further includes: after each round of scanning, determining whether a preset condition is met, and stopping the scanning if the preset condition is met; the preset condition includes the number of scanning rounds reaching a preset number or the absence of a region with a confidence level less than a first confidence level threshold.

[0009] In one possible embodiment, the confidence field obtained from the previous scan is updated based on the second point cloud data acquired by scanning to obtain a second confidence field, including: determining at least one point cloud observation result based on the second point cloud data; the point cloud observation result is used to indicate whether the sensor has observed point cloud data; inputting at least one point cloud observation result into the sensor model to obtain a first observation probability and a second observation probability corresponding to each point cloud observation result; the first observation probability is the probability that the sensor observes point cloud data when the area to be updated is occupied by the vehicle's outer surface; the second observation probability is the probability that the sensor observes point cloud data when the area to be updated is in an idle state; performing recursive Bayesian filtering calculation based on the confidence level in the current confidence field, the first observation probability, and the second observation probability to obtain the confidence level corresponding to the second point cloud data; and obtaining the second confidence field based on the confidence level corresponding to the second point cloud data.

[0010] In one possible embodiment, before each round of scanning begins, the method further includes: performing connected component processing on the region to be updated to obtain at least one region to be detected; for each of the at least one region to be detected, determining a set of feasible viewpoints that the second robotic arm can scan without collision; and determining a detection viewpoint based on the set of feasible viewpoints to control the second robotic arm to scan the region to be detected based on the detection viewpoint.

[0011] In one possible embodiment, determining the detection viewpoint based on the set of feasible viewpoints includes: calculating the total information gain corresponding to each feasible viewpoint in the set of feasible viewpoints for each region to be detected, thereby obtaining multiple total information gains; and using the feasible viewpoint corresponding to the largest total information gain among the multiple total information gains as the detection viewpoint for each region to be detected.

[0012] In one possible embodiment, during the scanning of the area to be updated using a point cloud acquisition device mounted on the second robotic arm, the method further includes: controlling the point cloud acquisition device mounted on the second robotic arm to scan the area to be updated at a corresponding scanning speed; the scanning speed is proportional to the confidence level corresponding to the area to be updated.

[0013] In one possible embodiment, generating a first cleaning trajectory based on a first confidence field includes: using a path planning algorithm to perform path planning on regions in the first confidence field with a confidence level greater than or equal to a first confidence threshold, thereby generating the first cleaning trajectory.

[0014] In one possible embodiment, based on the second cleaning trajectory, controlling the first robotic arm to clean the vehicle includes: after the second robotic arm completes at least one round of scanning of the area to be updated, performing path allocation processing on the second cleaning trajectory to determine the cleaning path corresponding to the first robotic arm and the cleaning path corresponding to the second robotic arm respectively; and controlling the first robotic arm and the second robotic arm to clean the vehicle according to their respective cleaning paths.

[0015] Secondly, this application provides a vehicle cleaning control device applied to a car wash equipment. The car wash equipment includes a first robotic arm and a second robotic arm. The device includes: an acquisition module for acquiring first point cloud data of the vehicle's outer surface; a determination module for determining a first confidence field based on the first point cloud data; a generation module for generating a first cleaning trajectory based on the first confidence field; a control module for controlling the first robotic arm to clean the vehicle based on the first cleaning trajectory; a scanning module for scanning at least once the areas to be updated in the first confidence field with confidence levels less than a first confidence threshold using a point cloud acquisition device mounted on the second robotic arm; the generation module is further used to update the first cleaning trajectory based on the scanned second point cloud data to obtain a second cleaning trajectory; and the control module is further used to control the first robotic arm to clean the vehicle based on the second cleaning trajectory.

[0016] Thirdly, this application provides a vehicle cleaning control system, including a control device and a car wash device, wherein the control device is communicatively connected to the car wash device; the control device is used to send control commands to the car wash device; the car wash device or the control device is used to execute the method provided above, determine a second cleaning trajectory of the vehicle, and generate control commands based on the second cleaning trajectory to control the cleaning of the vehicle.

[0017] Fourthly, this application provides a car wash device, including a main body, a first robotic arm, a second robotic arm, a memory, and a processor; the main body is connected to the first robotic arm; the main body is connected to the second robotic arm; the first robotic arm is used to clean the surface of a vehicle; the second robotic arm is used to carry a point cloud acquisition device to scan areas in a first confidence field with a confidence level less than a first confidence threshold; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory, causing the processor to perform the method provided above.

[0018] Fifthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method provided above.

[0019] Sixthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the method provided above.

[0020] The vehicle cleaning control method, apparatus, system, equipment, storage medium, and program products provided in this application acquire first point cloud data of the vehicle's outer surface and determine a first confidence field based on the first point cloud data. This enables the generation of corresponding first cleaning trajectories based on the reliability of point clouds in different areas of the vehicle's outer surface, thereby allowing the first robotic arm to prioritize cleaning the vehicle according to targeted planning results. By using a point cloud acquisition device on a second robotic arm to scan the areas to be updated in the first confidence field with confidence levels lower than the first confidence threshold for at least one round, and updating the first cleaning trajectory based on the second point cloud data obtained from the scan, the application can supplement and correct point cloud holes, occlusions, or low-confidence areas, thereby obtaining a more accurate second cleaning trajectory and controlling the first robotic arm to continue cleaning, thus improving the accuracy of cleaning planning, operational efficiency, and reliability of cleaning coverage. Attached Figure Description

[0021] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0022] Figure 1 A flowchart illustrating the vehicle cleaning control method provided in this application;

[0023] Figure 2 A schematic diagram of the dual-arm collaborative vehicle cleaning method provided in this application;

[0024] Figure 3 A schematic diagram of the vehicle cleaning control device provided in this application;

[0025] Figure 4This is a structural diagram of the car wash equipment provided in this application.

[0026] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0027] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application.

[0028] Intelligent vehicle cleaning control falls under the field of industrial robotics and vehicle service automation technology, and is typically applied in tunnel-type or station-type car wash equipment. This type of equipment usually includes dual robotic arms for performing cleaning operations, a point cloud acquisition device mounted on the robotic arms, and a processing unit for trajectory planning and motion control.

[0029] In this scenario, the equipment first needs to acquire point cloud data of the vehicle's outer surface, and then establish a cognitive result of the vehicle's surface based on the point cloud to generate the cleaning trajectory of the robotic arm, thereby completing the automatic cleaning operation of the outer surface of different vehicle models.

[0030] Specifically, the existing dual-arm car wash robot's cleaning process is as follows: First, sensors such as LiDAR are used to perform a single scan of the vehicle to collect raw point cloud data of the vehicle's outer surface; for areas with data gaps or sparse point cloud defects in the raw point cloud, a fitting algorithm is used to fill in the gaps and generate complete vehicle body point cloud data; based on the complete vehicle body point cloud data, the vehicle cleaning trajectory is planned; finally, the cleaning trajectory is divided, and the two robotic arms are controlled to synchronously complete cleaning operations such as spraying and brushing.

[0031] However, the process of filling in gaps may introduce data errors, affecting the accuracy of the subsequently generated cleaning trajectory and potentially causing problems such as missed areas or ineffective empty washing. Therefore, existing car wash solutions suffer from poor reliability in vehicle cleaning.

[0032] In view of this, this application provides a vehicle cleaning control method. It uses point cloud data to determine a confidence level to distinguish between reliable high-confidence regions and ambiguous low-confidence regions requiring updating. A second robotic arm performs multiple rounds of point cloud scanning and iteratively corrects the confidence level only for the low-confidence regions, while the control is achieved by cleaning the high-confidence regions using the second robotic arm. This allows for parallel cleaning and detection. During the cleaning process, the cleaning trajectory is updated using new point cloud data collected by the second robotic arm, enabling the robotic arm to clean with the new trajectory. This dynamically completes the vehicle body areas missed by the initial trajectory, effectively solving the problems of path planning deviation and localized missed cleaning caused by the uncertainty of single point cloud data, improving the accuracy of cleaning planning, and thus enhancing the reliability of the car wash equipment in cleaning the vehicle.

[0033] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0034] Figure 1 A flowchart illustrating the vehicle cleaning control method provided in this application. The method includes the following steps:

[0035] S101. Obtain the first point cloud data of the vehicle's outer surface.

[0036] The vehicle cleaning control method in this application is applied to a car wash equipment, which serves as the execution carrier of the method and includes at least a first robotic arm, a second robotic arm, and a point cloud acquisition device.

[0037] The first robotic arm is the cleaning execution component in the car wash equipment, and its end can be equipped with a brushing component, a spraying component, a wiping component, or a composite cleaning component.

[0038] The second robotic arm is the active detection and execution component in the car wash equipment. A point cloud acquisition device is mounted on the end of the second robotic arm and can adjust its scanning position and shooting angle to scan the vehicle's outer surface.

[0039] For example, the point cloud acquisition device can be a device with point cloud acquisition function, such as a lidar or a 3D vision sensor, used to acquire 3D point cloud data of the vehicle's outer surface.

[0040] The first robotic arm is used to perform the actual cleaning action on the vehicle's outer surface, and the second robotic arm is used to complete supplementary scanning during the cleaning process. The first point cloud data is used to reflect the position and contour of the vehicle's outer surface in three-dimensional space.

[0041] For example, after the vehicle enters the car wash bay and completes its initial positioning, the point cloud acquisition devices on the first and second robotic arms perform an initial scan of the vehicle's outer surface to form first point cloud data. This first point cloud data is used to characterize the initial observation results of the vehicle's outer surface.

[0042] S102. Determine the first confidence field based on the first point cloud data.

[0043] The first confidence field is used to characterize the confidence distribution of different regions on the outer surface of the vehicle.

[0044] Specifically, after acquiring the first point cloud data, the vehicle's outer surface can be divided into multiple continuous regions or discrete units, and a confidence level can be determined for each region, thereby constructing a first confidence field covering the vehicle's outer surface. The confidence level in the first confidence field can be obtained by comprehensively considering features related to the point cloud quality. For locations prone to voids, such as glass, door handle recesses, the vicinity of rearview mirrors, or the edge of the roof, the confidence value in the first confidence field is usually lower than that of other regions.

[0045] For example, a three-dimensional voxel grid can be used to discretize the space of the vehicle's outer surface. The comprehensive confidence of each voxel is calculated based on three types of features: point cloud density, geometric discreteness, and edge gradient, thereby forming the first confidence field. The specific calculation process is as follows:

[0046] The three-dimensional space corresponding to the vehicle's outer surface is divided into several voxel units of equal size. For any voxel... The density confidence score is calculated based on the number of point clouds contained within it. The calculation formula is as follows:

[0047] ;

[0048] Wherein, min() is the minimum value function, used to constrain the value of the density confidence level, with an upper limit of 1; voxels The number of point clouds contained within; This is a preset reference density threshold, for example, 10. When the number of point clouds within a voxel is greater than or equal to this reference density threshold, the voxel point cloud density is determined to be sufficient, and the density confidence level is set to 1.

[0049] For voxels containing point cloud data Geometric uncertainty is characterized by the degree of spatial dispersion of point clouds within voxels. First, the three-dimensional covariance matrix of the point clouds within voxels is calculated. The calculation formula is as follows:

[0050] ;

[0051] in, For voxels Three-dimensional coordinate vectors of points; The vector of the average coordinates of all points within the voxel; superscript This represents the transpose operation of a vector.

[0052] Voxels are further calculated based on the covariance matrix. Geometric uncertainty The calculation formula is as follows:

[0053] ;

[0054] Here, det() is the determinant operation of a matrix, used to measure the overall dispersion of the covariance matrix; the superscript 1 / 3 indicates the cube root operation, used to restore the volume dimension of the determinant to the length dimension, so as to keep it consistent with the coordinate dimension of the point cloud; the larger the geometric uncertainty value, the more scattered the spatial distribution of the point cloud within the voxel, and the lower the reliability of the geometric shape.

[0055] Based on the density confidence level, calculate the three-dimensional spatial gradient of the density confidence level, and the density gradient. The calculation formula is as follows:

[0056] ;

[0057] in, , , Density confidence levels at , , The partial derivatives in the three spatial directions, the larger the gradient magnitude, the closer the location is to the boundary of the point cloud cavity.

[0058] For voxels containing point cloud data The edge uncertainty is calculated based on the density gradient magnitude at its location. The calculation formula is as follows:

[0059] ;

[0060] Among them, sigmoid() is a sigmoid activation function used to map any real number to the interval (0,1); This is the gradient scaling factor, for example, 5, used to adjust the recognition sensitivity of edge regions; is the L2 norm of the density gradient vector at the voxel; the edge uncertainty ranges from [0,1], and the higher the value, the closer the voxel is to the edge of the hole and the worse the stability of the geometric information.

[0061] The overall confidence level of voxels is calculated based on density confidence, geometric uncertainty, and marginal uncertainty. The calculation formula is as follows:

[0062] ;

[0063] in, The value range is [0,1], and the higher the value, the higher the reliability of the point cloud data corresponding to the voxel; exp() is an exponential function with the natural constant e as the base; This is the weighting coefficient for geometric uncertainty, for example, 2.0; This is the weighting coefficient for edge uncertainty, for example, 1.0.

[0064] For voxels that do not contain point cloud data, their overall confidence level is set to 0.

[0065] The combined confidence scores of all voxels constitute the first confidence field covering the outer surface of the vehicle.

[0066] S103. Generate a first cleaning trajectory based on the first confidence field, and control the first robotic arm to clean the vehicle based on the first cleaning trajectory.

[0067] The first cleaning trajectory is used to instruct the first robotic arm to perform cleaning on the vehicle's exterior surface area, which currently has sufficient confidence.

[0068] In this application, based on the confidence distribution results of each region in the first confidence field, trajectory planning is performed on the regions that meet the first confidence threshold to generate a first cleaning trajectory. After the generated first cleaning trajectory is sent to the first robotic arm, the first robotic arm moves according to the first cleaning trajectory and synchronously drives the cleaning components to brush, spray, or wipe the corresponding surfaces.

[0069] It is understandable that, since the first cleaning trajectory is based on the first confidence field and its coverage is the area with high current cognitive quality, the cleaning action can be started before all low confidence areas are filled in.

[0070] In one possible implementation, generating a first cleaning trajectory based on a first confidence field includes: using a path planning algorithm to perform path planning on regions in the first confidence field where the confidence level is greater than or equal to a first confidence threshold, thereby generating the first cleaning trajectory.

[0071] It is understandable that path planning is performed only on regions within the first confidence field that are greater than or equal to the first confidence threshold, and a first cleaning trajectory matching that region is output. This trajectory is then sent to the first robotic arm to perform the cleaning action. Since the trajectory generation is directly based on regions with high confidence, the first robotic arm can continuously clean regions with high confidence along the confirmed first cleaning trajectory.

[0072] For example, from the first confidence field, surface voxels corresponding to the outer surface of the vehicle are filtered, and only voxels corresponding to the vehicle body surface area are retained to participate in path solving, thereby improving the efficiency of path planning. The kinematic reachability workspace constraints of the first robotic arm are superimposed, and grid cells that cannot be covered by the first robotic arm are eliminated.

[0073] Based on this, each surface grid is marked with a confidence level: high confidence areas are core cleaning units that are required to be fully covered in the initial planning stage, and must be covered without omissions; medium confidence areas are flexible transition units that are not required to be covered, and are included in the trajectory in an exploratory manner in the initial stage, and their coverage path can be dynamically corrected according to the scanning and detection results of the area to be updated.

[0074] The high confidence region is the region in the first confidence field where the confidence level is greater than or equal to the second confidence threshold, and the medium confidence region is the region in the first confidence field where the confidence level is greater than or equal to the first confidence threshold and less than the second confidence threshold, wherein the second confidence threshold is greater than the first confidence threshold.

[0075] When planning paths, a cost function that integrates confidence and path length can be used to guide path search, ensuring both coverage integrity and cleaning efficiency. Continuous trajectories Total cost Defined as an integral form along the arc length of the trajectory:

[0076] ;

[0077] in, For continuous trajectory The arc length parameter, The range of values ​​corresponds to the continuous trajectory The total arc length from the starting point to the ending point; Arc length parameter on the trajectory The corresponding spatial location point; For position The confidence level at that point This is a confidence penalty item; the lower the confidence level, the higher the penalty value. It is used to guide the path to prioritize covering high-confidence areas. This is the path length cost term, used to constrain the total path length and avoid unnecessary detours; This is the weighting coefficient for the confidence penalty term, for example, 0.7; This is the weighting factor for the path length item, for example, 0.3.

[0078] Based on the aforementioned cost model, the path planning algorithm adopted can be the A* path search algorithm. The A* path search algorithm is used to perform unified path planning for high-confidence regions and medium-confidence regions. Specifically, all high-confidence surface grates are set with mandatory coverage constraints, and the search process needs to traverse all high-confidence grates. Medium-confidence surface grates are not set with mandatory coverage constraints, and the cost function guides the path to pass through as needed.

[0079] Specifically, the center point of the grid on the vehicle surface is used as the search node. Each node records the corresponding three-dimensional coordinates in space, the cumulative actual cost from the starting point to the current node, and the set of high-confidence grids that have been covered. Each node expands to the surface neighborhood, with the single-step step size aligned with the grid cell side length. The movement cost of each step is calculated based on the aforementioned single-step incremental cost formula.

[0080] Guided by the cost function and combined with the forced coverage constraint, the A* algorithm will generate a path that fully covers all high-confidence regions. The path segments corresponding to these high-confidence regions are marked with fixed attributes and remain unchanged in the subsequent trajectory fusion process without modification or adjustment.

[0081] For medium-confidence regions, the confidence penalty is moderate, and the path will only pass through them naturally when the detour cost is higher than the transit cost, forming exploratory path segments. The path segments corresponding to this type of medium-confidence region are marked as adjustable attributes, and no action nodes corresponding to deep cleaning are configured. Only light coverage parameters are set. When the area to be updated is scanned, the confidence is upgraded, and the trajectory is fused, the adjacent medium-confidence path segments can be locally corrected without replanning the high-confidence fixed path segments and the global path.

[0082] The discrete grid path sequence obtained in the above steps is smoothed, and B-spline curves are used to fit the discrete path points to eliminate inflection points of the broken lines, ensuring the position, velocity, and acceleration of the trajectory are continuous in three orders, thus avoiding impacts during the movement of the first robotic arm. The Cartesian space path is converted into a joint space trajectory using inverse kinematics, and the joint angles, angular velocities, and angular accelerations are checked point-by-point to ensure they meet the hardware constraints of the first robotic arm. Local adjustments are made for infeasible points. The collision risk of the first robotic arm's end effector, its links, vehicles, and the environment is detected point-by-point, and local detour planning is performed for path segments with collision risks. After successful verification, the output can be directly sent to the first robotic arm for execution of the first cleaning trajectory.

[0083] In this way, the first cleaning trajectory is consistent with the reliable region, the trajectory generation process is less dependent on the uncertain region, the first robotic arm can stably cover the verifiable surface in the initial cleaning stage, and provide clear boundaries for subsequent trajectory updates, thereby improving the executability and coverage consistency of the cleaning trajectory.

[0084] S104. Using the point cloud acquisition device installed on the second robotic arm, at least one round of scanning is performed on the area to be updated in the first confidence field where the confidence level is less than the first confidence threshold, and the first cleaning trajectory is updated based on the scanned second point cloud data to obtain the second cleaning trajectory.

[0085] The second robotic arm is used to supplement the perception of low-confidence areas in the first confidence field during the cleaning process performed by the first robotic arm.

[0086] The region to be updated is the region in the first confidence field with a confidence level lower than the first confidence threshold. This region usually corresponds to the location where there are holes, sparse point clouds, edge distortion, or partial occlusion in the scan.

[0087] The point cloud acquisition device is installed at the end of the second robotic arm, which can perform at least one round of scanning on the area to be updated.

[0088] The second point cloud data, as a supplementary observation result, has stronger regional specificity compared to the first point cloud data, and its collection range is concentrated in the locations where the original confidence level is insufficient.

[0089] In one possible implementation, before each round of scanning begins, the method further includes: performing connected component processing on the region to be updated to obtain at least one region to be detected; for each of the at least one region to be detected, determining a set of feasible viewpoints that the second robotic arm can reach and avoid collisions; and determining a detection viewpoint based on the set of feasible viewpoints to control the second robotic arm to scan the region to be detected based on the detection viewpoint.

[0090] Specifically, the area to be updated is first represented as several grid cells or point sets in the vehicle coordinate system, and connectivity analysis is performed based on grid adjacency relationships to obtain at least one area to be detected. For each area to be detected, a set of candidate viewpoints is generated by combining the kinematic model of the second robotic arm, joint limits, end-effector mounting posture, and the three-dimensional contour of the vehicle's outer surface. Accessibility checks and collision detection are then performed on each candidate viewpoint, retaining those that satisfy the criteria for trajectory planning and safety clearance meeting a set threshold, forming a set of feasible viewpoints. Subsequently, one or more detection viewpoints are determined from the set of feasible viewpoints, and the second robotic arm is controlled to move to that detection viewpoint to perform a scan to obtain the corresponding second point cloud data. If the area to be detected is large, the same area can also be scanned sequentially using multiple detection viewpoints from the set of feasible viewpoints to supplement observation information from different angles.

[0091] In this way, the second robotic arm can complete the directional scanning under the conditions of accessibility and collision-free operation, thereby outputting second point cloud data that better meets the needs of subsequent trajectory updates. At the same time, since the scanning position is determined based on the local area, it can reduce invalid observations and improve the efficiency of point supplementation in low-confidence areas, thereby improving the pertinence and stability of subsequent cleaning trajectory updates.

[0092] In one possible implementation, determining the detection viewpoint based on the set of feasible viewpoints includes: calculating the total information gain corresponding to each feasible viewpoint in the set of feasible viewpoints for each region to be detected, thereby obtaining multiple total information gains; and using the feasible viewpoint corresponding to the largest total information gain among the multiple total information gains as the detection viewpoint for each region to be detected.

[0093] Specifically, for each region to be explored, the set of feasible viewpoints is traversed, and the total information gain of each feasible viewpoint is calculated. During the calculation, the observation coverage of different sub-regions within the region by the viewpoint can be determined first, and the total information gain corresponding to each feasible viewpoint is finally obtained. After comparing all the total information gains, the feasible viewpoint with the largest value is selected as the detection viewpoint for the region to be explored, and the second robotic arm is controlled to move to that position to complete the scan accordingly.

[0094] For example, based on the information gain maximization criterion, the optimal detection viewpoint corresponding to each region to be detected can be solved from the set of feasible viewpoints. By quantifying the information supplementation capability of each candidate viewpoint for the region to be detected, the scanning position with the highest observation benefit is selected. The specific calculation process is as follows:

[0095] For the first obtained after processing the connected components One area to be detected The second robotic arm scans the set of reachable and collision-free feasible viewpoints. Within this area, find the optimal detection viewpoint that maximizes the total information gain. The calculation formula is as follows:

[0096] ;

[0097] in, Indicates the set of feasible viewpoints In the middle, find the function that makes The variable that achieves the maximum value ; The set of feasible viewpoints for the second robotic arm is determined by constraints imposed by the robotic arm's kinematic model, joint constraints, end-effector mounting posture, and collision detection results. For the first A region to be explored, that is, a single connected component obtained after connected component analysis of the region to be updated; Area to be detected Any point in space within; From the candidate viewpoint Observation space point The information gain that can be obtained; For spatial points To candidate viewpoints The Euclidean distance is used, and the denominator uses the square of the distance to reflect the distance attenuation characteristic of the observation gain. That is, the closer the observation point is to the viewpoint, the higher its information gain contributes to the total gain.

[0098] Furthermore, the information gain of a single spatial point is jointly determined by the current confidence level of that point and the observation incident angle; the information gain of a single point... The calculation formula is as follows:

[0099] ;

[0100] in, For spatial points The current confidence level at the location, As a fundamental term for information gain, the lower the initial confidence of a point, the greater the potential for information enhancement after active detection, and the higher the corresponding information gain. For spatial points Surface normal vector at the location With the observation direction vector The angle between them; The standard deviation of the observation angle is denoted as Gaussian, for example, 30°. It is used to characterize the degree of influence of the incident angle on the observation quality. The smaller the angle between the observation direction and the surface normal, the higher the observation quality and the greater the information gain.

[0101] In this way, the second robotic arm can always locate the viewpoint with the highest value for supplementing information in the area to be detected, so that the subsequent second point cloud data collected can more effectively cover the missing and uncertain areas, reducing the repeated collection caused by invalid scanning.

[0102] In one possible implementation, a point cloud acquisition device mounted on a second robotic arm performs at least one scan on the regions to be updated in the first confidence field where the confidence level is less than the first confidence threshold. Based on the scanned second point cloud data, the first cleaning trajectory is updated to obtain a second cleaning trajectory. This includes: for each scan, updating the confidence field obtained in the previous scan based on the second point cloud data acquired during the scan to obtain a second confidence field; determining a cleaning path corresponding to a first region in the second confidence field within the regions to be updated where the confidence level is greater than or equal to the first confidence threshold; fusing the cleaning path corresponding to the first region with the first cleaning trajectory to generate a second cleaning trajectory; and identifying regions in the second confidence field within the regions to be updated where the confidence level is less than the first confidence threshold as new regions to be updated.

[0103] Specifically, when the confidence level of certain regions reaches or exceeds the first confidence threshold, a corresponding cleaning path is generated based on the region's boundary contour, normal distribution, and robotic arm accessibility. This cleaning path is then spliced, inserted, or partially replaced with an existing first cleaning trajectory to obtain a second cleaning trajectory suitable for the current surface perception state. For regions still below the threshold, they remain as new regions to be updated, and point clouds are collected again in subsequent scanning cycles.

[0104] For example, when the confidence of the area to be updated increases from less than the first confidence threshold to greater than or equal to the second confidence threshold, a local replanning of the first cleaning trajectory will be triggered: the area is moved from the low-confidence area to the high-confidence area, and the incremental graph search algorithm is called to perform local replanning on the original first cleaning trajectory, updating only the path segments affected by the area, and inserting a cleaning path covering the area into the original trajectory; the incremental graph search algorithm can use the DLite algorithm, and the local replanning can be completed within 50ms, avoiding excessively long interruptions in the cleaning operation of the first robotic arm.

[0105] In this way, the second cleaning trajectory can be corrected round by round as the area to be updated shrinks, so that the robotic arm's coverage path on the vehicle surface is consistent with the current real surface state, and the continuity, fit and updating efficiency of the cleaning trajectory are improved.

[0106] In one possible implementation, the method further includes: after each round of scanning, determining whether a preset condition is met; if the preset condition is met, stopping the scanning; the preset condition includes the number of scanning rounds reaching a preset number or the absence of a region with a confidence level less than a first confidence level threshold.

[0107] In actual operation, after each round of scanning, the point cloud acquisition device synchronously updates the current confidence field and then determines whether there are still regions below the first confidence threshold based on the update results. If they still exist, the next round of scanning continues; if they no longer exist, the scanning ends directly, and the current update results are used for subsequent cleaning trajectory fusion. By setting the upper limit of the number of scanning rounds and the region confidence status together as the stopping condition, the scanning process can be terminated in a timely manner after the target update effect is achieved.

[0108] For example, when the scanning rounds reach a preset number and the scanning is terminated, if there are still areas within the region to be updated whose confidence level has not reached the second confidence threshold, then differential processing is performed based on the actual confidence interval of that area:

[0109] For areas with a confidence level greater than or equal to the first confidence threshold but less than the second confidence threshold, they are determined to be highly likely to be real vehicle body surfaces. These areas typically correspond to locations that are difficult to identify using laser point clouds, such as deep grooves, curved surface angles, and matte light-absorbing materials. Even after multiple rounds of scanning, the confidence level cannot be raised to a high level. A corresponding trial cleaning path is generated for these areas and integrated into the second cleaning trajectory. During cleaning operations, the movement speed of the first robotic arm's end effector is reduced, and the safe working distance between the end effector and the vehicle's outer surface is increased. Only light-duty cleaning (e.g., spray rinsing, only rinsing away surface dust with water flow, without using brush components to contact the vehicle body) is performed on these areas. Deep cleaning actions (e.g., reciprocating brushing, high-pressure water rinsing, etc.) are not performed. These areas are marked as verification areas for final verification.

[0110] For areas where the confidence level is still below the first confidence threshold, they are determined to be idle areas that do not belong to the actual vehicle surface. The suspected signals corresponding to the initial point cloud are noise, clutter, or misjudgments. These areas are officially marked as idle areas and will not be subject to further detection and cleaning planning, thus avoiding the execution of invalid detection and cleaning operations and improving the overall operational accuracy and efficiency.

[0111] In this way, the scanning process can be dynamically terminated based on the actual confidence level update, which can reduce invalid repeated scanning and match the re-sampling behavior of the second robotic arm with the actual disappearance state of the area to be updated.

[0112] In one possible implementation, the confidence field obtained from the previous scan is updated based on the second point cloud data acquired by scanning to obtain a second confidence field. This includes: determining at least one point cloud observation result based on the second point cloud data; the point cloud observation result is used to indicate whether the sensor has observed point cloud data; inputting at least one point cloud observation result into the sensor model to obtain a first observation probability and a second observation probability corresponding to each point cloud observation result; the first observation probability is the probability that the sensor observes point cloud data when the area to be updated is occupied by the vehicle's outer surface; the second observation probability is the probability that the sensor observes point cloud data when the area to be updated is in an idle state; performing recursive Bayesian filtering calculation based on the confidence level in the current confidence field, the first observation probability, and the second observation probability to obtain the confidence level corresponding to the second point cloud data; and obtaining the second confidence field based on the confidence level corresponding to the second point cloud data.

[0113] For example, for each spatial location within the region to be updated, a recursive Bayesian filter is performed based on the current confidence level, the first observation probability, and the second observation probability to obtain the updated confidence level. The calculation formula is as follows:

[0114] ;

[0115] in, Spatial location before update The confidence level at a given position is the prior confidence level at that position in the current confidence field. For the updated spatial location The confidence level at the second point cloud data point is the posterior confidence level. These are the point cloud observation results acquired in the current scan. The first observation probability is the probability that the sensor will observe the point cloud when the area to be updated is occupied by the vehicle's outer surface. The second observation probability is the probability that the sensor will observe the point cloud when the area to be updated is in an idle state.

[0116] After updating the confidence level of each spatial location within the update area, the second confidence field can be obtained.

[0117] In this way, the updated confidence field can simultaneously reflect historical scanning information and current scanning results, thereby ensuring that the state judgment of the area to be updated is consistent with the subsequent cleaning trajectory generation. At the same time, due to the use of recursive Bayesian filtering to iteratively correct the confidence, the second confidence field can continuously update local holes, occlusions and sparse echoes, thereby improving the reliability of the trajectory and the consistency of surface coverage after regional fusion.

[0118] In one possible implementation, during the scanning of the area to be updated using a point cloud acquisition device mounted on the second robotic arm, the method further includes: controlling the point cloud acquisition device mounted on the second robotic arm to scan the area to be updated at a corresponding scanning speed; the scanning speed is proportional to the confidence level corresponding to the area to be updated.

[0119] Specifically, when the confidence level of the area to be updated is high, the point cloud acquisition device on the second robotic arm is controlled to scan at a high speed, allowing it to quickly cover the area with more complete information. When the confidence level of the area to be updated is low, the point cloud acquisition device on the second robotic arm is controlled to scan at a lower speed, allowing it to stay in that area for a longer time and collect more observation points, thus obtaining more comprehensive second point cloud data. The scanning speed can be mapped based on the confidence level value. The mapping relationship can be a linear proportional relationship, a piecewise proportional relationship, or a continuous function relationship. The specific mapping parameters can be set according to the point cloud density, the robotic arm's response performance, and the sampling period.

[0120] In this way, the point cloud acquisition device takes less time to scan in high-confidence areas and can increase the number of effective samples in low-confidence areas, making the second point cloud data of the area to be updated more closely match the real vehicle surface, thereby improving the accuracy and coverage continuity of subsequent cleaning trajectory updates, and making the allocation of scanning resources of the second robotic arm more in line with the differentiated needs of the region.

[0121] To avoid motion interference between the first and second robotic arms during collaborative operations, a coordination mechanism based on task priority is used for dual-arm motion control: the first robotic arm, responsible for the main cleaning task, is defined as the high-priority arm, while the second robotic arm, responsible for scanning the area to be updated, is defined as the low-priority arm. When there is a spatiotemporal conflict between the planned scanning path of the second robotic arm and the predicted cleaning trajectory of the first robotic arm, the second robotic arm automatically performs a pause and wait or partial detour planning to prioritize the continuous execution of the first robotic arm's cleaning operation.

[0122] The first and second robotic arms establish a spatial mapping relationship through a unified world coordinate system. They share their current pose and planned motion trajectory data in real time to perform real-time collision detection and motion coordination.

[0123] S105. Based on the second cleaning trajectory, control the first robotic arm to clean the vehicle.

[0124] The second cleaning trajectory is a cleaning trajectory that is locally corrected, supplemented, or reconstructed based on the first cleaning trajectory and combined with the second point cloud data. It is used to instruct the first robotic arm to continue to complete the subsequent cleaning of the vehicle's outer surface.

[0125] In one possible implementation, based on the second cleaning trajectory, controlling the first robotic arm to clean the vehicle includes: after the second robotic arm completes at least one round of scanning of the area to be updated, performing path allocation processing on the second cleaning trajectory to determine the cleaning path corresponding to the first robotic arm and the cleaning path corresponding to the second robotic arm respectively; and controlling the first robotic arm and the second robotic arm to clean the vehicle according to their respective cleaning paths.

[0126] Specifically, after obtaining the second cleaning trajectory, the trajectory points are sequentially analyzed, and the trajectory is redistributed based on the current poses of the two robotic arms, the accessible workspace, and the safe distance between them. For continuous trajectory segments, if they are located within the vehicle body area that the first robotic arm can stably cover, they are assigned to the cleaning path corresponding to the first robotic arm; if they are located in an area that the second robotic arm can access more easily, they are assigned to the cleaning path corresponding to the second robotic arm. For areas with overlapping coverage requirements, the first segment of the trajectory can be assigned to one robotic arm and the second segment to another robotic arm according to a preset priority rule to form a non-conflicting collaborative cleaning path. Subsequently, two sets of motion commands are output synchronously, causing the two robotic arms to perform cleaning actions along their respective paths, maintaining posture continuity and speed matching at the trajectory intersections.

[0127] This allows the updated cleaning trajectory to be redistributed according to the capabilities of the two robotic arms, thereby incorporating the area to be updated into the collaborative operation process, ensuring the continuity and completeness of trajectory execution. At the same time, since the first and second robotic arms each undertake their respective path segments, the cleaning operation can continue in an orderly manner after the update, and achieve collaborative cleaning of the vehicle's outer surface, thereby improving the consistency of path execution and the completeness of vehicle surface coverage.

[0128] After the two robotic arms complete the cleaning operation, the cleaning quality verification stage can begin, where the first and second robotic arms work together to perform multi-dimensional verification.

[0129] One method is visual verification, which uses an RGB-D camera mounted on the second robotic arm to collect images and depth data of the vehicle's exterior surface after cleaning. The image segmentation algorithm is then used to identify whether there are any stains or residues on the vehicle's surface, with a focus on verifying the cleaning effect in low-confidence areas such as the original area to be updated.

[0130] The second method is water film detection, which uses capacitive or optical sensors at the end of a robotic arm to detect the thickness of the water film on the vehicle body surface and determines whether the corresponding area is sufficiently cleaned based on the water film distribution characteristics.

[0131] If the calibration detects areas with residual stains, it will automatically generate a rewash trajectory for the corresponding area and control the robotic arm to perform a second cleaning of that area.

[0132] Furthermore, the verification results can serve as feedback signals to update relevant parameters of the sensor observation model, optimize the accuracy of subsequent confidence calculations, and form a closed-loop optimization mechanism for long-term learning.

[0133] Therefore, the vehicle cleaning control method of this application involves a first robotic arm performing the cleaning of the vehicle's outer surface, and a second robotic arm performing supplementary scanning of low-confidence areas. A first confidence field is established using first point cloud data, and a first confidence threshold is used to divide areas into directly cleanable areas and areas awaiting updating. Then, second point cloud data is used to update the local trajectory, thus forming a dynamic trajectory control process for vehicle cleaning operations. This implementation allows cleaning operations to proceed even when the initial scan contains holes, sparse areas, or uncertainties, and enables local observation compensation and trajectory correction during the operation, ensuring that updates are concentrated on surface areas with insufficient confidence.

[0134] Figure 2 This is a schematic diagram of the dual-arm collaborative vehicle cleaning method provided in this application.

[0135] S201. Input the first point cloud data of the vehicle's outer surface and the operating status parameters of the dual robotic arms.

[0136] S202. Perform three-dimensional voxelization on the first point cloud data, calculate the uncertainty measure, and generate the first confidence field.

[0137] S203. Generate a first cleaning trajectory based on the first confidence field, identify the areas to be updated with a confidence level less than the first confidence threshold, and assign dual-arm operation tasks.

[0138] S204. The second point cloud data of the area to be updated is collected by the second robotic arm, the second confidence field is updated, the trajectory fusion is performed on the area with the confidence level meets the standard, and the second cleaning trajectory is generated.

[0139] S205. After the cleaning operation is completed, the cleaning effect is verified by the end sensor, a rewash trajectory is generated and feedback is given to optimize the sensor observation model.

[0140] Figure 3 A schematic diagram of the vehicle cleaning control device provided in this application is shown below. Figure 3 As shown, the vehicle cleaning control device 30 provided in this embodiment includes:

[0141] The acquisition module 301 is used to acquire the first point cloud data of the vehicle's outer surface;

[0142] The determination module 302 is used to determine the first confidence field based on the first point cloud data;

[0143] The generation module 303 is used to generate a first cleaning trajectory based on a first confidence field;

[0144] Control module 304 is used to control the first robotic arm to clean the vehicle based on the first cleaning trajectory;

[0145] The scanning module 305 is used to perform at least one round of scanning on the area to be updated in the first confidence field where the confidence level is less than the first confidence threshold, using the point cloud acquisition device set on the second robotic arm.

[0146] The generation module 303 is also used to update the first cleaning trajectory based on the scanned second point cloud data to obtain the second cleaning trajectory;

[0147] The control module 304 is also used to control the first robotic arm to clean the vehicle based on the second cleaning trajectory.

[0148] In one possible implementation, the determining module 302 is further configured to update the confidence field obtained in the previous scan based on the second point cloud data obtained in each scan, so as to obtain the second confidence field.

[0149] The generation module 303 is also used to determine the cleaning path corresponding to the first region for the first region in the second confidence field of the region to be updated, whose confidence is greater than or equal to the first confidence threshold;

[0150] The generation module 303 is also used to merge the cleaning path corresponding to the first region with the first cleaning trajectory to generate a second cleaning trajectory;

[0151] The determination module 302 is further configured to determine the regions in the second confidence field of the region to be updated that have a confidence level less than the first confidence threshold as new regions to be updated.

[0152] In one possible implementation, the scanning module 305 is further configured to determine whether a preset condition is met after each round of scanning, and to stop scanning if the preset condition is met.

[0153] The preset conditions include reaching a preset number of scan rounds or the absence of regions with a confidence level less than the first confidence threshold.

[0154] In one possible implementation, the determining module 302 is further configured to determine at least one point cloud observation result based on the second point cloud data; the point cloud observation result is used to indicate whether the sensor has observed point cloud data;

[0155] The determination module 302 is also used to input at least one point cloud observation result into the sensor model to obtain the first observation probability and the second observation probability corresponding to each point cloud observation result; the first observation probability is the probability that the sensor observes point cloud data when the area to be updated is occupied by the vehicle's outer surface; the second observation probability is the probability that the sensor observes point cloud data when the area to be updated is in an idle state.

[0156] The determination module 302 is also used to perform recursive Bayesian filtering calculation based on the confidence level in the current confidence field, the first observation probability and the second observation probability to obtain the confidence level corresponding to the second point cloud data;

[0157] The determination module 302 is also used to obtain the second confidence field based on the confidence level corresponding to the second point cloud data.

[0158] In one possible implementation, the scanning module 305 is further configured to perform connected component processing on the region to be updated before the start of each round of scanning to obtain at least one region to be probed.

[0159] The scanning module 305 is also used to determine, for each of the at least one region to be detected, a set of feasible viewpoints that the second robotic arm can scan without collision.

[0160] The scanning module 305 is also used to determine the detection viewpoint based on the set of feasible viewpoints, so as to control the second robotic arm to scan the area to be detected based on the detection viewpoint.

[0161] In one possible implementation, the scanning module 305 is also used to calculate the total information gain corresponding to each feasible viewpoint in the feasible viewpoint set based on each region to be detected, and obtain multiple total information gains.

[0162] The scanning module 305 is also used to take the feasible viewpoint corresponding to the maximum total information gain among multiple total information gains as the detection viewpoint for each region to be detected.

[0163] In one possible implementation, the scanning module 305 is also used to control the point cloud acquisition device set on the second robotic arm to scan the area to be updated at a corresponding scanning speed; the scanning speed is proportional to the confidence level corresponding to the area to be updated.

[0164] In one possible implementation, the generation module 303 is further configured to employ a path planning algorithm to perform path planning on regions in the first confidence field where the confidence level is greater than or equal to the first confidence threshold, thereby generating a first cleaning trajectory.

[0165] In one possible implementation, the control module 304 is further configured to perform path allocation processing on the second cleaning trajectory after the second robotic arm completes at least one round of scanning of the area to be updated, and to determine the cleaning path corresponding to the first robotic arm and the cleaning path corresponding to the second robotic arm respectively.

[0166] The control module 304 is also used to control the first robotic arm and the second robotic arm to clean the vehicle according to their respective cleaning paths.

[0167] The vehicle cleaning control device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0168] This application provides a vehicle cleaning control system, including a control device and a car wash device, wherein the control device is communicatively connected to the car wash device; the control device is used to send control commands to the car wash device; the car wash device or the control device is used to execute the method provided above, determine a second cleaning trajectory of the vehicle, and generate control commands based on the second cleaning trajectory to control the cleaning of the vehicle.

[0169] This application provides a car wash device, including a main body, a first robotic arm, a second robotic arm, a memory, and a processor; the main body is connected to the first robotic arm; the main body is connected to the second robotic arm; the first robotic arm is used to clean the surface of a vehicle; the second robotic arm is used to carry a point cloud acquisition device to scan areas in a first confidence field with a confidence level less than a first confidence threshold; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory, causing the processor to perform the method provided above.

[0170] Figure 4 This is a structural diagram of the car wash equipment provided in this application. Figure 4 As shown, the car wash equipment 40 provided in this embodiment includes at least one processor 401 and a memory 402. Optionally, the car wash equipment 40 also includes a communication component 403. The processor 401, memory 402, and communication component 403 are connected via a bus 404.

[0171] In the specific implementation process, at least one processor 401 executes computer execution instructions stored in memory 402, causing at least one processor 401 to execute the above-described vehicle cleaning control method.

[0172] The specific implementation process of processor 401 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0173] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0174] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0175] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0176] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0177] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0178] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0179] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0180] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0181] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0182] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0183] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0184] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0185] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A vehicle cleaning control method, characterized in that, Applied to car wash equipment, the car wash equipment including a first robotic arm and a second robotic arm, the method includes: Acquire the first point cloud data of the vehicle's outer surface; Based on the first point cloud data, determine the first confidence field; Based on the first confidence field, a first cleaning trajectory is generated, and the first robotic arm is controlled to clean the vehicle based on the first cleaning trajectory. Using the point cloud acquisition device installed on the second robotic arm, at least one round of scanning is performed on the area to be updated in the first confidence field where the confidence level is less than the first confidence threshold, and the first cleaning trajectory is updated based on the scanned second point cloud data to obtain the second cleaning trajectory. Based on the second cleaning trajectory, the first robotic arm is controlled to clean the vehicle.

2. The method according to claim 1, characterized in that, The step involves using a point cloud acquisition device mounted on the second robotic arm to scan at least one round of the region to be updated in the first confidence field where the confidence level is less than the first confidence threshold, and updating the first cleaning trajectory based on the scanned second point cloud data to obtain a second cleaning trajectory, including: For each round of scanning, the confidence field obtained in the previous round of scanning is updated based on the second point cloud data obtained from the scanning to obtain the second confidence field; For a first region in the region to be updated where the confidence level in the second confidence field is greater than or equal to the first confidence threshold, a cleaning path corresponding to the first region is determined. The cleaning path corresponding to the first area is merged with the first cleaning trajectory to generate the second cleaning trajectory; The regions in the second confidence field of the region to be updated whose confidence level is less than the first confidence threshold are identified as new regions to be updated.

3. The method according to claim 2, characterized in that, The method further includes: After each round of scanning, determine whether the preset conditions are met; if the preset conditions are met, stop scanning. The preset conditions include reaching a preset number of scan rounds or the absence of regions with a confidence level less than the first confidence threshold.

4. The method according to claim 2, characterized in that, The step of updating the confidence field obtained from the previous scan based on the second point cloud data acquired by the scan to obtain the second confidence field includes: Based on the second point cloud data, at least one point cloud observation result is determined; the point cloud observation result is used to indicate whether the sensor has observed point cloud data. The at least one point cloud observation result is input into the sensor model to obtain the first observation probability and the second observation probability corresponding to each point cloud observation result; the first observation probability is the probability that the sensor observes point cloud data when the area to be updated is occupied by the vehicle's outer surface; the second observation probability is the probability that the sensor observes point cloud data when the area to be updated is in an idle state. Based on the confidence level in the current confidence field, the first observation probability, and the second observation probability, a recursive Bayesian filter is performed to calculate the confidence level corresponding to the second point cloud data. The second confidence field is obtained based on the confidence level corresponding to the second point cloud data.

5. The method according to claim 2, characterized in that, Before each round of scanning begins, the method further includes: Perform connected component processing on the region to be updated to obtain at least one region to be detected; For each of the at least one region to be detected, determine a set of feasible viewpoints that the second robotic arm can scan without collision; Based on the set of feasible viewpoints, a detection viewpoint is determined, and the second robotic arm is controlled to scan the area to be detected based on the detection viewpoint.

6. The method according to claim 5, characterized in that, The step of determining the detection viewpoint based on the set of feasible viewpoints includes: For each of the regions to be detected, the total information gain corresponding to each feasible viewpoint in the feasible viewpoint set is calculated to obtain multiple total information gains; The feasible viewpoint corresponding to the maximum total information gain among the multiple total information gains is used as the detection viewpoint for each region to be detected.

7. The method according to any one of claims 1 to 6, characterized in that, During the scanning of the area to be updated using the point cloud acquisition device mounted on the second robotic arm, the method further includes: The point cloud acquisition device mounted on the second robotic arm is controlled to scan the area to be updated at a corresponding scanning speed; the scanning speed is proportional to the confidence level of the area to be updated.

8. The method according to claim 1, characterized in that, The step of generating the first cleaning trajectory based on the first confidence field includes: A path planning algorithm is used to plan paths for regions in the first confidence field where the confidence level is greater than or equal to the first confidence threshold, thereby generating a first cleaning trajectory.

9. The method according to any one of claims 1 to 6, characterized in that, The step of controlling the first robotic arm to clean the vehicle based on the second cleaning trajectory includes: After the second robotic arm completes at least one round of scanning of the area to be updated, the second cleaning trajectory is processed for path allocation, and the cleaning path corresponding to the first robotic arm and the cleaning path corresponding to the second robotic arm are determined respectively. The first robotic arm and the second robotic arm are controlled to clean the vehicle according to their respective cleaning paths.

10. A vehicle cleaning control device, characterized in that, Applied to car wash equipment, the car wash equipment includes a first robotic arm and a second robotic arm, the device includes: The acquisition module is used to acquire the first point cloud data of the vehicle's outer surface. The determination module is used to determine a first confidence field based on the first point cloud data; The generation module is used to generate a first cleaning trajectory based on the first confidence field; The control module is used to control the first robotic arm to clean the vehicle based on the first cleaning trajectory; The scanning module is used to perform at least one round of scanning on the area to be updated in the first confidence field where the confidence level is less than the first confidence threshold, using the point cloud acquisition device set on the second robotic arm. The generation module is also used to update the first cleaning trajectory based on the scanned second point cloud data to obtain the second cleaning trajectory; The control module is also used to control the first robotic arm to clean the vehicle based on the second cleaning trajectory.

11. A vehicle cleaning control system, characterized in that, It includes control equipment and car wash equipment, wherein the control equipment is communicatively connected to the car wash equipment; The control device is used to send control commands to the car wash equipment; The car wash equipment or the control equipment is configured to perform the method as described in any one of claims 1-9, determine a second cleaning trajectory for the vehicle, and generate the control command based on the second cleaning trajectory to control the cleaning of the vehicle.

12. A car wash device, characterized in that, It includes a main body, a first robotic arm, a second robotic arm, a memory, and a processor; the main body is connected to the first robotic arm; the main body is connected to the second robotic arm. The first robotic arm is used to clean the surface of the vehicle; The second robotic arm is used to carry a point cloud acquisition device to scan the region in the first confidence field where the confidence level is less than the first confidence threshold. The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-9.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-9.

14. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-9.

Images