Control method, system, device and vehicle for a cleaning device in a vehicle

Abstract

The embodiment of the application provides a kind of control method, system, device and vehicle of cleaning device in vehicle, the method comprises: obtaining the image data of at least one windshield mounted in vehicle;Based on image data, generate the clean demand heat map of windshield, wherein the clean demand heat map is used to characterize the distribution state of stain on the surface of windshield at different positions;Call information recognition model, identify clean demand heat map, obtain the stain type of stain on windshield, and determine the control parameter of cleaning device based on stain type, wherein the information recognition model is obtained by training deep reinforcement learning model using image data sample of windshield sample in vehicle sample and clean demand heat map sample of windshield sample;Based on control parameter, control cleaning device to perform cleaning operation on windshield.The application solves the technical problem that the cleaning device of vehicle cannot be effectively controlled to clean.

Description

Technical Field

[0001] This application relates to the field of vehicles, and more specifically, to a control method, system, device, and vehicle for a cleaning device in a vehicle. Background Technology

[0002] Currently, vehicle cleaning devices (such as windshield wiper systems) often employ fixed arc-shaped mechanical paths and open-loop control strategies, relying on a single rain sensor located at the rearview mirror to trigger the cleaning operation. This fails to detect the actual cleaning effect, resulting in insufficient cleaning of critical blind spots such as the base of the A-pillar, the curved area at the edge of the glass, and areas where water accumulates due to high-speed wind pressure. Therefore, the technical problem of effectively controlling vehicle cleaning devices remains.

[0003] There is currently no good solution to the above problems. Summary of the Invention

[0004] This application provides a method, system, device, and vehicle for controlling a cleaning device in a vehicle, to at least solve the technical problem of the inability to effectively control the cleaning device in a vehicle for cleaning.

[0005] According to one aspect of the embodiments of this application, a control method for a cleaning device in a vehicle is provided, comprising: acquiring image data of at least one windshield installed in the vehicle; generating a cleaning demand heatmap of the windshield based on the image data, wherein the cleaning demand heatmap is used to characterize the distribution state of stains at different locations on the surface of the windshield; invoking an information recognition model to identify the cleaning demand heatmap to obtain the stain type of the stains on the windshield, and determining control parameters of the cleaning device based on the stain type, wherein the information recognition model is obtained by training a deep reinforcement learning model using image data samples of windshield samples in a vehicle sample and cleaning demand heatmap samples of windshield samples, and the control parameters include a set of control instruction parameters for the cleaning device to perform cleaning operations according to a cleaning path; and controlling the cleaning device to perform cleaning operations on the windshield based on the control parameters.

[0006] Furthermore, based on the image data, a heatmap of the cleanliness requirements for the windshield is generated, including: inputting the image data into a fully convolutional neural network to obtain a classification probability map for each pixel in the image data, wherein the classification probability map is used to represent the confidence distribution corresponding to the cleaning intensity of each pixel on the windshield; and generating a heatmap of cleanliness requirements based on the classification probability map and the vehicle's bus data, wherein the bus data is used to represent the dynamic operating condition information of the vehicle under the current driving state.

[0007] Furthermore, based on the classification probability map and the vehicle's bus data, a cleanliness requirement heatmap is generated, including: spatially aligning the classification probability map and the vehicle's bus data to obtain aligned classification probability map and bus data, wherein the bus data is used to represent the dynamic operating condition information of the vehicle under the current driving state; weightedly fusing the aligned classification probability map and bus data to obtain the cleanliness requirement value corresponding to the pixel region of each pixel; and generating a cleanliness requirement heatmap based on the cleanliness requirement value.

[0008] Furthermore, the aligned classification probability map and bus data are weighted and fused to obtain the cleanliness requirement value corresponding to the pixel region of each pixel. This includes: associating the confidence score of the cleaning intensity of each pixel in the classification probability map with the parameters reflecting the current driving state in the bus data to obtain the weight information of the cleaning intensity, wherein the weight information is used to represent the importance of cleaning the windshield under different types of stains; and summing the confidence score and the weight information in a weighted manner to obtain the cleanliness requirement value. The method also includes: adjusting the weight information in response to detecting that the number of cleaning operations performed is greater than a threshold and the cleanliness requirement value is greater than a threshold value.

[0009] Furthermore, based on the cleanliness requirement values, a cleanliness requirement heatmap is generated, including: normalizing the cleanliness requirement values ​​to obtain normalized cleanliness requirement values; filtering the normalized cleanliness requirement values ​​to obtain filtered cleanliness requirement values; converting the filtered cleanliness requirement values ​​into a two-dimensional heatmap according to their position in the windshield coordinate system; and defining the two-dimensional heatmap as the cleanliness requirement heatmap.

[0010] Further, the stain types include at least one of the following: dry stain type, uniform water film stain type, discrete water droplet stain type, viscous stain type, and oil film stain type. Based on the stain type, the control parameters of the cleaning device are determined, including: nonlinear feature extraction of the spatial distribution of the cleaning demand heatmap to obtain the demand region, wherein the demand region is used to characterize the region where the priority of cleaning the windshield is higher than the priority threshold; based on the demand region and the stain type, control parameters are generated, wherein the control parameters include the starting position information and ending position information of the cleaning device, the velocity profile curve along the cleaning path, and the dynamic pressure curve of the wiper arm of the cleaning device as a function of stroke. The starting position information is used to indicate the physical position where the wiper arm begins cleaning in the horizontal direction of the windshield, the ending position information is used to indicate the physical position where the wiper arm ends cleaning in the horizontal direction of the windshield, the velocity profile curve is used to indicate the instantaneous movement speed change of the wiper arm at different spatial positions along the cleaning path, and the dynamic pressure curve is used to indicate the distribution characteristics of the clamping force applied to the windshield surface by the wiper blade as a function of position during the movement along the cleaning path.

[0011] Further, acquiring image data of at least one windshield installed in the vehicle includes: acquiring initial image data collected by a data acquisition device in the vehicle, wherein the initial image data includes visual representations of optical reflection, stain distribution, and ambient lighting conditions on the windshield surface; correcting and / or performing perspective correction on the initial image data to obtain image data, wherein the image data is a bird's-eye view image based on the windshield.

[0012] Furthermore, based on the control parameters, the cleaning device is controlled to perform a cleaning operation on the windshield, including: converting the control parameters to obtain control commands, wherein the control commands include at least one of the following: displacement and time sequence commands for the servo motor of the cleaning device, speed and time commands for the rotary motor of the cleaning device, and pressure and displacement commands for the servo motor; and performing a cleaning operation on the windshield based on the control commands.

[0013] Furthermore, based on control commands, a cleaning operation is performed on the windshield, including: in response to displacement and time sequence commands, controlling the wiper arm of the cleaning device to move to a target position; in response to speed and time commands, driving the wiper arm to move along the surface of the windshield; in response to pressure and displacement commands, adjusting the normal pressure between the wiper blade and the surface of the windshield; and performing a cleaning operation on the windshield according to the target position, movement operation, and normal pressure.

[0014] Furthermore, the method also includes: in response to the cleaning device completing the cleaning operation, evaluating the information recognition model to obtain the evaluation result; and in response to the vehicle being in a parked charging state or an idle state, updating the information recognition model based on the evaluation result.

[0015] According to another aspect of the embodiments of this application, a control system for a cleaning device in a vehicle is also provided, comprising: a vision perception unit for acquiring image data of at least one windshield installed in the vehicle; a central processing and decision-making unit for generating a cleaning demand heat map of the windshield based on the image data, wherein the cleaning demand heat map is used to characterize the distribution state of stains at different locations on the surface of the windshield; invoking an information recognition model to identify the cleaning demand heat map to obtain the stain type of the stains on the windshield, and determining control parameters of the cleaning device based on the stain type, wherein the information recognition model is obtained by training a deep reinforcement learning model using image data samples of windshield samples in a vehicle sample and clean demand heat map samples of windshield samples, and the control parameters include a set of control instruction parameters for the cleaning device to perform cleaning operations according to a cleaning path; and an intelligent wiper actuator for controlling the cleaning device to perform cleaning operations on the windshield based on the control parameters.

[0016] According to another aspect of the embodiments of this application, a control device for a cleaning device in a vehicle is also provided, comprising: an acquisition unit for acquiring image data of at least one windshield installed in the vehicle; a generation unit for generating a cleaning demand heat map of the windshield based on the image data, wherein the cleaning demand heat map is used to characterize the distribution state of stains at different locations on the surface of the windshield; an identification unit for calling an information identification model to identify the cleaning demand heat map, obtain the stain type of the stains on the windshield, and determine control parameters of the cleaning device based on the stain type, wherein the information identification model is obtained by training a deep reinforcement learning model using image data samples of windshield samples in a vehicle sample and cleaning demand heat map samples of windshield samples, and the control parameters include a set of control instruction parameters for the cleaning device to perform cleaning operations according to a cleaning path; and a control unit for controlling the cleaning device to perform cleaning operations on the windshield based on the control parameters.

[0017] According to another aspect of the embodiments of this application, a vehicle is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0018] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0019] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0020] According to another aspect of the embodiments of this application, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the methods in various embodiments of this application.

[0021] According to another aspect of the embodiments of this application, a computer program is also provided, which, when executed by a processor, implements the methods of the various embodiments of this application.

[0022] In this embodiment, image data of at least one windshield installed in a vehicle is acquired, and a cleaning demand heatmap is generated based on the image data, representing the distribution of stains at different locations on the windshield surface. Then, an information recognition model is invoked to identify the cleaning demand heatmap, determining the stain type on the windshield, and based on the stain type, determining the control parameters of the cleaning device. Since the control parameters include a set of control instruction parameters for the cleaning device to perform cleaning operations according to the cleaning path, the cleaning device can be controlled to perform cleaning operations on the windshield based on these parameters. This overcomes the limitations of related technologies that only use fixed arc-shaped mechanical paths and open-loop control strategies, resulting in poor windshield cleaning effects. Therefore, it solves the technical problem of not being able to effectively control the vehicle's cleaning device, achieving the technical effect of effectively controlling the vehicle's cleaning device. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0024] Figure 1 This is a flowchart of a control method for a cleaning device in a vehicle according to an embodiment of this application;

[0025] Figure 2 This is a schematic diagram of an intelligent windshield wiper system deeply integrated into the automotive electronic and electrical architecture according to an embodiment of this application;

[0026] Figure 3 This is a schematic diagram of an intelligent wiper actuator assembly according to an embodiment of this application;

[0027] Figure 4This is a flowchart of a smart windshield wiper method deeply integrated into the automotive electronic and electrical architecture according to an embodiment of this application;

[0028] Figure 5 This is a schematic diagram of a control system for a cleaning device in a vehicle according to an embodiment of this application;

[0029] Figure 6 This is a schematic diagram of a control device for a cleaning device in a vehicle according to an embodiment of this application. Detailed Implementation

[0030] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0031] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0032] According to an embodiment of this application, an embodiment of a control method for a cleaning device in a vehicle is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0033] This embodiment provides a control method for a cleaning device in a vehicle. Figure 1 This is a flowchart of a control method for a cleaning device in a vehicle according to an embodiment of this application, such as... Figure 1 As shown, the process includes the following steps.

[0034] Step S102: Acquire image data of at least one windshield installed in the vehicle.

[0035] In the technical solution provided by step S102 of this application, in order to fully perceive the cleanliness of the entire windshield, image data of at least one windshield installed in the vehicle can be obtained.

[0036] In this embodiment, the aforementioned image data may include visual features such as water film distribution, raindrop shape, oil film reflection, shellac residue, mud accumulation, and aging areas of the coating on the windshield surface, as well as dynamic phenomena such as water flow trajectory, edge accumulation, and local dry areas caused by vehicle speed and wind pressure.

[0037] Optionally, image data (e.g., optical images) of the windshield surface can be continuously acquired by a forward-facing advanced driver assistance system (ADAS) camera. This camera can be located behind the rearview mirror and has been pre-calibrated for the perspective distortion of the vehicle's windshield. It has high dynamic range and low-light adaptability and can stably image under complex lighting conditions such as rain, fog, backlight, and night.

[0038] Optionally, the aforementioned image data can be transmitted in real time to the vehicle's domain controller (e.g., intelligent driving domain controller or high-performance body domain controller) via in-vehicle Ethernet or a high-speed flexible data rate (CAN FD) bus.

[0039] In this embodiment of the application, by acquiring image data of at least one windshield installed in a vehicle, a data foundation is provided for the subsequent windshield cleaning demand heat map, which can significantly enhance the active safety capabilities and user experience of cleaning devices (such as wiper systems) under complex operating conditions.

[0040] Step S104: Based on the image data, generate a heat map of the cleanliness requirements of the windshield.

[0041] In the technical solution provided in step S104 of this application, the cleanliness requirement heat map can be used to characterize the distribution of stains at different locations on the surface of the windshield. The cleanliness requirement heat map can also be called a glass cleanliness requirement heat map.

[0042] In this embodiment, the aforementioned cleanliness demand heatmap is a two-dimensional spatial distribution map used to characterize the urgency and necessity of cleaning operations for each physical location point on the windshield surface. The value range of the cleanliness demand heatmap is 0 to 1, where 0 indicates no cleaning is required (e.g., dry areas) and 1 indicates a need for deep cleaning (e.g., high-concentration oil film or stubborn stains).

[0043] Optionally, the aforementioned heat map of cleanliness requirements does not simply reflect the presence or absence of stains, but rather is a multi-dimensional weighted assessment result that comprehensively considers stain type, spatial location, current vehicle speed, wind pressure disturbance, and driver visibility safety priority.

[0044] Optionally, perspective correction and coordinate mapping are performed on the raw images captured by the vehicle's forward-facing camera to transform them into bird's-eye view images based on the physical plane of the glass, i.e., obtaining image data. Subsequently, the image data can be input into a lightweight fully convolutional network (FCN). This FCN outputs five semantic classification probabilities for each pixel, such as dry glass, uniform water film, discrete water droplets, viscous stains, and oil film. These classification probabilities can then be fused with vehicle speed, yaw rate, and raw rainfall signals from the CAN FD bus, and dynamically weighted for different areas. For example, discrete water droplets at the base of the A-pillar are significantly weighted due to their impact on lateral visibility at high speeds, while oil film directly in front of the driver's seat is given the highest cleaning requirement value because it directly obstructs the view. Finally, through pixel-level weighted superposition, a grayscale heatmap corresponding one-to-one with the physical coordinates of the windshield is generated, i.e., a cleanliness requirement heatmap.

[0045] In this embodiment, a heat map of the windshield's cleanliness requirements is generated based on image data, realizing the intelligent control foundation for on-demand cleaning and precise coverage, enabling subsequent recognition based on deep reinforcement learning models to perform actions according to real visual needs rather than mechanical presets.

[0046] Step S106: Call the information recognition model to identify the heat map of cleanliness requirements, obtain the stain type of the stain on the windshield, and determine the control parameters of the cleaning device based on the stain type.

[0047] In the technical solution provided in step S106 of this application, the information recognition model is obtained by training a deep reinforcement learning model using image data samples of windshield samples in the vehicle sample and cleanliness requirement heat map samples of windshield samples. The control parameters may include a set of control command parameters for the cleaning device to perform cleaning operations according to the cleaning path.

[0048] In this embodiment, the information recognition model can be an online decision-maker based on Deep Reinforcement Learning (DRL), employing an improved Soft Actor-Critic (SAC) algorithm framework. The information recognition model is one that is trained offline and fine-tuned and inferred online.

[0049] Optionally, the input to the above information recognition model can be a real-time generated heat map of cleaning demand H(t). After identifying the type of stains on the windshield by combining multi-dimensional state variables such as vehicle speed, wiper blade life, and historical action status, the model outputs complete control parameters required for a complete cleaning operation (e.g., wiping action) based on the stain type. These control parameters include the starting position information and ending position information of the cleaning device, the speed profile curve along the cleaning path, and the dynamic pressure curve of the wiper blade pressure of the cleaning device as a function of stroke.

[0050] Optionally, the cleanliness demand heatmap and vehicle operating state vector are input into the SAC strategy network. The SAC strategy network extracts spatial features from the high-demand area distribution, edge aggregation trend, and abrupt changes in dirt density in the cleanliness demand heatmap through its internal neural layers. Combined with empirical patterns learned during pre-training, such as "high speed + edge water droplets → extended path + low speed pressurization" and "oil film concentration → fixed-point slow sweeping + high pressure maintenance," it outputs a better control parameter curve in the continuous motion space. This control parameter curve may include the X-axis start and end point offset (corresponding to the start and end position information of the cleaning device), the nonlinear function V(s) of speed changing with stroke (corresponding to the speed profile curve along the cleaning path), and the adaptive curve P(s) of pressure changing with displacement (corresponding to the dynamic pressure curve of the wiper arm of the cleaning device changing with stroke).

[0051] In the embodiments of this application, through the above steps, a better cleaning path (e.g., motion trajectory), i.e. control parameters, can be automatically generated for different vehicle windshield curvatures, different regional stain types (e.g., southern mold, northern dust), and different wiper blade aging states, thus solving the problem of not being able to effectively control the vehicle's cleaning device to perform cleaning operations on the windshield.

[0052] Step S108: Based on the control parameters, control the cleaning device to perform a cleaning operation on the windshield.

[0053] In the technical solution provided by step S108 of this application, the control parameters are a dynamic instruction set output by the information recognition model, including start position information, end position information (i.e., the start and end coordinates of the wiper arm in the X-axis direction), the speed profile curve along the cleaning path (i.e., the speed profile curve V(s) along the arc trajectory), and the dynamic pressure curve of the wiper arm blade pressure changing with the stroke (i.e., the normal pressure curve P(s) of the contact surface between the wiper blade and the windshield).

[0054] In this embodiment, the cleaning device is controlled to perform cleaning operations on the windshield based on control parameters. This enables precise execution driven by visual perception, allowing each wiping action to accurately match the current distribution of dirt on the windshield and the driving environment, rather than relying on a preset fixed pattern.

[0055] Optionally, the vehicle's domain controller can transmit the output control parameters to the vehicle's trajectory planning submodule. The trajectory planning submodule generates a high-precision four-dimensional control sequence of time, position, speed, and pressure, which is then distributed to three execution units: The X-axis servo motor, based on the displacement-time sequence command, can precisely drive the wiper arm to move steplessly in the horizontal direction, allowing the start and end points to exceed traditional mechanical limits and cover traditional blind spots such as the base of the A-pillar; The rotary motor, based on the speed-time command, uses vector control to achieve non-uniform wiping. For example, in high-demand areas of the cleanliness heatmap (such as oil film areas and edge liquid accumulation areas), it automatically reduces the speed to 1.5 Hz for fine wiping, and accelerates to 3.0 Hz in clean areas to achieve rapid sweeping, avoiding unnecessary power consumption; The Z-axis servo motor, based on the pressure-displacement mapping command, adjusts the vertical pressure applied by the wiper blade to the windshield in real time. In edge areas where the curvature of the windshield changes drastically, it automatically increases the pressure by 15%–25% to ensure the rubber strip adheres, and in flat areas, it returns to the reference pressure to reduce noise and rubber strip wear.

[0056] Optionally, the three execution units mentioned above achieve millisecond-level coordination under the synchronous clock drive of the Electronic Control Unit (ECU), enabling the wiper arm to perform continuous, dynamic, and adaptive wiping by moving in a spatially variable cleaning path, speed, and pressure composite motion while closely adhering to the windshield surface.

[0057] In this embodiment of the application, the above steps realize three-dimensional dynamic control of the "perception-decision-execution" closed loop, which can effectively improve the cleaning coverage and realize effective control of the vehicle's cleaning device to clean the windshield.

[0058] Through steps S102 to S108, image data of at least one windshield installed in the vehicle is acquired. Based on the image data, a cleaning demand heatmap representing the distribution of stains at different locations on the windshield surface is generated. Then, an information recognition model is invoked to identify the cleaning demand heatmap, determining the stain type on the windshield. Based on the stain type, control parameters for the cleaning device are determined. Since the control parameters include a set of control instruction parameters for the cleaning device to perform cleaning operations according to the cleaning path, the cleaning device can be controlled to perform cleaning operations on the windshield based on these parameters. This overcomes the limitations of related technologies that only use fixed arc-shaped mechanical paths and open-loop control strategies, resulting in poor windshield cleaning effects. Therefore, it solves the technical problem of not being able to effectively control the vehicle's cleaning device, achieving the technical effect of effectively controlling the vehicle's cleaning device.

[0059] The above-mentioned method of this application will be further described below.

[0060] As an optional implementation, step S104, generating a heat map of windshield cleanliness requirements based on image data, includes: inputting image data into a fully convolutional neural network to obtain a classification probability map of each pixel in the image data, wherein the classification probability map is used to represent the confidence distribution corresponding to the cleaning intensity of each pixel on the windshield; generating a heat map of cleanliness requirements based on the classification probability map and vehicle bus data, wherein the bus data is used to represent the dynamic operating condition information of the vehicle under the current driving state.

[0061] In this embodiment, during the process of generating a heatmap of windshield cleanliness requirements based on image data, the image data can be input into a lightweight fully convolutional neural network (FCN). This FCN is trained offline on hundreds of thousands of images covering different weather conditions, lighting, glass coating states, and types of contaminants, and can output five semantic classification probabilities for each pixel in the image data: dry glass, uniform water film, discrete water droplets, viscous stains (e.g., shellac or mud), and oil film, thereby forming a classification probability map spatially aligned with the image data.

[0062] Optionally, the aforementioned classification probability map can reflect the confidence distribution of each pixel belonging to each state. For example, a certain area may simultaneously have a "water film" probability of 0.7 and an "oil film" probability of 0.3, indicating that the area is a mixed oil-water contamination zone. Subsequently, this classification probability map can be fused with the vehicle's bus data in a multimodal manner to generate a cleanliness requirement heatmap.

[0063] Optionally, the bus data mentioned above can be dynamic operating condition information such as vehicle speed, yaw rate, raw rain sensor signal, and ambient light intensity on the CAN FD. This is just an example and no specific limitation is made here.

[0064] Optionally, the aforementioned bus data can be used to model wind pressure disturbances, visibility safety priorities, and the dynamic evolution of dirt under current driving conditions. For example, when the vehicle speed exceeds 100 km / h, the weight of "discrete water droplets" at the base of the A-pillar and the upper edge of the side windows can be automatically increased, as high-speed airflow easily forms water film accumulation in these areas, obstructing lateral visibility. When a high probability of oil film is detected and the vehicle speed is low, the cleaning requirement for the area directly in front of the driver is strengthened, as this area directly affects forward visibility.

[0065] Optionally, by fusing the classification probability map and the vehicle's bus data, a two-dimensional cleanliness demand heatmap H(t) that is strictly aligned with the physical coordinates of the windshield can be generated, where each pixel value in the cleanliness demand heatmap is a continuous demand intensity between 0 and 1, representing the urgency of cleaning at that location.

[0066] Optionally, overcoming the inherent limitations of traditional rain sensors that can only sense local water volume and cannot distinguish stain types or spatial distribution, a global cleaning demand model with spatial resolution, semantic understanding capabilities, and environmental adaptability can be constructed.

[0067] In this embodiment, a cleanliness requirement heatmap is generated based on the classification probability map and the vehicle's bus data, enabling the wiper system to have human-like perception capabilities such as "seeing the type of stain, understanding dangerous areas, and sensing the impact of vehicle speed". This allows subsequent control parameters to no longer blindly rely on preset algorithms, but to accurately respond to real visual and operational requirements, thereby significantly improving the cleaning coverage of key visual areas with the same energy consumption and time, and effectively reducing dynamic blind spots caused by wind pressure, curvature, and stain mixing.

[0068] As an optional implementation, a cleanliness requirement heatmap is generated based on the classification probability map and vehicle bus data, including: spatially aligning the classification probability map and vehicle bus data to obtain aligned classification probability map and bus data, wherein the bus data is used to represent the dynamic operating condition information of the vehicle under the current driving state; weightedly fusing the aligned classification probability map and bus data to obtain the cleanliness requirement value corresponding to the pixel region of each pixel; and generating a cleanliness requirement heatmap based on the cleanliness requirement value.

[0069] In this embodiment, during the process of generating a cleanliness requirement heatmap based on the classification probability map and vehicle bus data, the classification probability map and vehicle bus data (e.g., vehicle speed, yaw rate, longitudinal acceleration, ambient light, and raw analog signals from the rain sensor) can be spatially aligned. This alignment process can accurately map the pixel coordinates in the classification probability map to the two-dimensional physical plane of the windshield based on the physical coordinate calibration parameters of the camera-windshield completed before the vehicle leaves the factory. At the same time, the environmental variables in the bus data are transformed into a two-dimensional spatial field consistent with the image space through a preset wind pressure model and a field-of-view safety weight function, so that each piece of bus data can correspond to a specific area on the windshield surface.

[0070] Optionally, after mapping each bus data point to a specific area on the windshield surface, the aligned classification probability map and the bus data can be weighted and fused at the pixel level. For example, dynamic weighting coefficients can be set for each type of stain (e.g., water film, oil film, mud) and dynamic operating condition information (e.g., high speed, turning). For instance, if the probability of detecting "discrete water droplets" in the A-pillar root area exceeds 0.4 during high-speed driving, the cleanliness requirement value is amplified to 1.8 times the original probability; if both "oil film" and "low light" conditions exist in the main field of vision area, the cleanliness requirement value is superimposed and enhanced to form a comprehensive risk assessment.

[0071] Optionally, the above weighted fusion process can be implemented using a learnable weighting function or a rule engine to ensure that the cleaning priority of critical safety areas is highlighted even under complex operating conditions. Subsequently, each pixel outputs a continuous cleanliness requirement value between 0 and 1, forming a two-dimensional cleanliness requirement heatmap H(t) that perfectly matches the physical surface of the windshield.

[0072] In this embodiment of the application, the above steps enable the wiper system to have the intelligence of environmental perception and safety priority judgment, which significantly improves the ability to maintain visibility in complex scenarios such as high-speed lane changes, driving on curves, strong wind disturbances, and oil film accumulation, and avoids energy waste and excessive wear of wiper blades caused by blindly wiping the entire width.

[0073] As an optional implementation, the aligned classification probability map and bus data are weighted and fused to obtain the cleanliness requirement value corresponding to the pixel region of each pixel. This includes: associating the confidence score of the cleaning intensity of each pixel in the classification probability map with the parameters reflecting the current driving state in the bus data to obtain the weight information of the cleaning intensity, wherein the weight information is used to represent the importance of cleaning the windshield under different types of stains; and summing the confidence score and the weight information in a weighted manner to obtain the cleanliness requirement value. The method further includes: adjusting the weight information in response to detecting that the number of cleaning operations performed is greater than a threshold and the cleanliness requirement value is greater than a threshold value.

[0074] In this embodiment, during the weighted fusion of the aligned classification probability map and bus data to obtain the cleanliness requirement value corresponding to the pixel region of each pixel, the confidence scores of various stains corresponding to each pixel in the classification probability map output by the fully convolutional neural network (such as "water film" 0.6, "oil film" 0.3, and "stain" 0.1) can be semantically correlated with the parameters reflecting the current dynamic operating conditions in the vehicle's bus data. For example, when the vehicle speed is higher than 80km / h, the cleaning weight of "discrete water droplets" and "uniform water film" in the A-pillar root area is automatically enhanced; when a large yaw rate is detected (during a turn), the weight of "water droplets" on the upper edge of the side window is temporarily increased to ensure the turning visibility; when the ambient light is lower than the ambient light threshold and the probability of detecting "oil film" is high, the cleaning weight of the area directly in front of the driver's seat is given the highest priority; these correlations can be realized through a preset dynamic weight function to form a cleaning importance score for each stain type and operating condition combination, i.e., weight information.

[0075] Optionally, the aforementioned weight information is not a fixed constant, but rather an environmentally sensitive parameter that changes in real time with the driving scenario. After obtaining the weight information, the confidence level of the original stain of each pixel can be multiplied by the corresponding weight information, and then the weighted results of multiple types of stains can be normalized and superimposed to finally generate the cleanliness requirement value of that pixel. This cleanliness requirement value changes continuously within the range of 0 to 1, truly reflecting the urgency of the need to clean this location under the current spatiotemporal conditions.

[0076] Optionally, when the cumulative number of cleaning operations exceeds a threshold (e.g., 50 consecutive scrapes), and under the same or similar operating conditions, the cleanliness requirement value of certain areas in the cleanliness requirement heatmap consistently exceeds the cleanliness requirement value threshold (e.g., consistently above 0.7), an online weight adjustment mechanism can be automatically activated. For example, based on the changes in the cleanliness requirement heatmap before and after cleaning recorded in experience playback and the feedback from the actual reward function, the association weight between stain type and operating condition can be fine-tuned. For example, if water droplets are found to remain in the root area of ​​the left A-pillar after multiple scrapes, the weight coefficient of this area under the "high speed + water droplet" combination can be gradually increased, and even triggering conditions for local pressure enhancement or micro-vibration strategies can be introduced.

[0077] In this embodiment, not only is accurate scene perception and demand modeling achieved in the initial stage, but the cleaning decision logic is also continuously optimized during long-term use. This allows it to automatically adapt to the aging characteristics of the vehicle's windshield coating, common local stains (such as mold in the south and dust in the north), and even the driver's driving habits, greatly improving cleaning efficiency and user satisfaction. Simultaneously, through adaptive adjustment of weights, it can effectively identify and overcome stubborn cleaning blind spots, gradually eliminating the user experience pain point of repeated cleaning failures, and achieving a leap from intelligent execution to autonomous learning.

[0078] As an optional implementation method, a cleanliness requirement heatmap is generated based on the cleanliness requirement value, including: normalizing the cleanliness requirement value to obtain a normalized cleanliness requirement value; filtering the normalized cleanliness requirement value to obtain a filtered cleanliness requirement value; converting the filtered cleanliness requirement value into a two-dimensional heatmap according to the position of the cleanliness requirement value in the coordinate system of the windshield; and determining the two-dimensional heatmap as the cleanliness requirement heatmap.

[0079] In this embodiment, during the process of generating a cleanliness requirement heatmap based on cleanliness requirement values, the cleanliness requirement value calculated for each pixel can be normalized to obtain a normalized cleanliness requirement value. That is, the cleanliness requirement values ​​are uniformly mapped to a standard range of 0 to 1, eliminating numerical drift caused by differences in operating conditions, sensor sensitivity, or neural network output scale, ensuring that the cleanliness requirement heatmap has consistent semantic comparability across all weather conditions and multiple scenarios.

[0080] Optionally, after obtaining the normalized cleanliness requirement value, spatial filtering can be applied to it. For example, nonlinear filtering algorithms such as Gaussian smoothing or median filtering can be used to suppress pixel-level noise while preserving the gradient edge features of key areas. For instance, a clear transition is maintained at the boundary between oil film accumulation areas and clean areas to avoid false triggering caused by image sensor jitter or instantaneous fluctuations in lighting. The filtered cleanliness requirement value still retains its spatial location information in the physical coordinate system of the windshield. Based on this, the processed cleanliness requirement values ​​of each pixel can be rearranged according to their corresponding physical coordinates to construct a two-dimensional heat map that strictly corresponds to the windshield surface. The color or grayscale depth in this two-dimensional heat map intuitively reflects the cleaning urgency of each area. For example, red or dark gray represents high-demand areas (such as oil film at the base of the A-pillar or high-speed wind pressure accumulation areas), while blue or light gray represents low-demand or dry areas. Finally, this two-dimensional heat map can be identified as a cleanliness requirement heat map and used as input for subsequent information recognition models.

[0081] In this embodiment of the application, the above steps solve the problems of noise interference, numerical drift and spatial discontinuity in the original pixel-level cleanliness requirement values, ensuring that the cleanliness requirement heatmap has a high signal-to-noise ratio, strong robustness and good spatial continuity, so that the subsequent information recognition model does not need to process data anomalies.

[0082] As an optional implementation, the stain type includes at least one of the following: dry stain type, uniform water film stain type, discrete water droplet stain type, viscous stain type, and oil film stain type. Step S106 involves determining the control parameters of the cleaning device based on the stain type, including: extracting nonlinear features from the spatial distribution of the cleaning demand heatmap to obtain a demand region, wherein the demand region is used to characterize areas where the priority for cleaning the windshield is higher than a priority threshold; and generating control parameters based on the demand region and the stain type, wherein the control parameters include the starting position information and ending position information of the cleaning device. The system includes stop position information, velocity profile curve along the cleaning path, and dynamic pressure curve showing the change in wiper blade pressure with stroke. The start position information indicates the physical position where the wiper arm begins cleaning in the horizontal direction of the windshield, and the end position information indicates the physical position where the wiper arm ends cleaning in the horizontal direction of the windshield. The velocity profile curve indicates the instantaneous velocity change of the wiper arm at different spatial positions along the cleaning path, and the dynamic pressure curve indicates the distribution characteristics of the clamping force applied to the windshield surface by the wiper blade as it moves along the cleaning path, with varying position.

[0083] In this embodiment, stain types include dry stain types (e.g., dry glass), uniform water film stain types (e.g., uniform water film), discrete water droplet stain types (e.g., discrete water droplets), viscous stain types (e.g., shellac / mud), and oil film stain types (e.g., oil film). These stain types are not simple image classification labels, but semantic units with clear physical meanings and mapping relationships to cleaning operations.

[0084] Optionally, based on the cleanliness requirement heatmap, nonlinear features can be extracted from the spatial distribution of the heatmap. Then, through convolutional-pooling layers or attention mechanisms in deep neural networks, regions with high gradient, high aggregation, and high persistence characteristics—these are the requirement regions. These requirement regions not only have an intensity higher than the priority threshold (e.g., requirement value > 0.6) but also exhibit a continuous or structured spatial distribution. Examples include the strip-shaped high-value area at the base of the A-pillar, the annular water film band at the upper edge of the windshield, and the oil film cloud in the main field of view. Essentially, this is the automatic identification of risk areas in key fields of view. Based on this, highly personalized control parameters can be generated by combining the type of stain with the spatial morphology of the requirement regions.

[0085] Optionally, the start and end position information, i.e., the start and end positions, no longer follow the fixed arc of mechanical limits, but are dynamically offset according to the left and right boundaries of the demand area. For example, when a high demand is detected at the base of the A-pillar due to a combination of "discrete water droplets + high-speed operation", the start position can be shifted 35mm to the left and the end position 25mm to the right, so that the cleaning path (e.g., the wiping trajectory) actively covers the traditional blind spot. The speed profile curve V(s) can be nonlinearly modulated according to the spatial distribution of the stain type. For example, in areas with dense oil film or viscous stains, the speed is reduced to 1.5Hz to achieve low-speed repeated wiping; in flat areas with only a uniform water film, the speed is increased to 3.0Hz to achieve high-speed sweeping to reduce energy consumption; the dynamic pressure curve P(s) of the wiper arm of the cleaning device, which varies with the stroke, is adjusted in coordination with the glass curvature and stain adhesion characteristics: in areas where the curvature of the windshield changes abruptly (e.g., near the A-pillar), the pressure is automatically increased by 15%–25% to ensure that the rubber strip adheres fully. In the flat area in the center of the windshield, the pressure is reduced to the baseline value to avoid excessive wear. These control parameters together constitute a four-dimensional control command set that changes continuously with spatial position, rather than a single gear command.

[0086] In this embodiment, high-dimensional visual perception information is transformed into executable, physically realizable multi-dimensional motion control commands, thus breaking away from the mechanical response mode of traditional windshield wiper systems that operate at a uniform speed and pressure across the entire width.

[0087] As an optional implementation, step S102, acquiring image data of at least one windshield installed in the vehicle, includes: acquiring initial image data collected by a data acquisition device in the vehicle, wherein the initial image data includes visual representations of optical reflection, stain distribution, and ambient lighting conditions on the windshield surface; correcting and / or performing perspective correction on the initial image data to obtain image data, wherein the image data is a bird's-eye view image based on the windshield.

[0088] In this embodiment, the vehicle's existing forward-facing ADAS camera can be used. This camera can be located behind the interior rearview mirror, and its field of view covers the entire windshield area.

[0089] Optionally, during the acquisition of image data for at least one windshield installed in the vehicle, initial image data (e.g., the original image) can be collected. This initial image data is not an idealized clean image, but rather contains multiple visual disturbances from the windshield surface. These include specular reflection caused by the aging of the windshield's coating, diffuse reflection caused by raindrops or water films, iridescent interference caused by oil films, overexposure or shadows caused by external light (such as direct sunlight or strong light at tunnel entrances and exits), and local texture abruptness caused by stains (such as insect remains, mud spots, or resin residue). These optical phenomena intertwine, severely obscuring the true state of the windshield in the initial image data.

[0090] Optionally, to restore the physical essence, rigorous geometric corrections and perspective transformations can be performed on the initial image data. For example, based on the proprietary calibration parameters completed before leaving the factory, nonlinear compensation can be applied to lens distortion (e.g., barrel distortion) to eliminate edge stretching. Then, using the known camera installation height, pitch angle, and three-dimensional geometric model of the windshield, the coordinate system of the initial image data can be precisely mapped to the physical plane coordinate system of the windshield, completing the spatial reconstruction from the "camera view" to the "windshield bird's-eye view." This process transforms the originally tilted, distorted, and perspective-compressed initial image data into orthogonal, distortion-free, and proportionally consistent image data (e.g., a two-dimensional top view), where each pixel coordinate directly corresponds to a physical location point on the windshield surface (e.g., 840mm from the bottom edge and 230mm from the left A-pillar).

[0091] In this embodiment of the application, by acquiring a bird's-eye view image based on the windshield, precise spatial anchor points are provided for subsequent generation of heat maps of stain types and cleaning needs, as well as planning of cleaning paths (e.g., trajectories).

[0092] As an optional implementation, step S108, based on the control parameters, controls the cleaning device to perform a cleaning operation on the windshield, including: converting the control parameters to obtain control commands, wherein the control commands include at least one of the following: displacement and time sequence commands for the servo motor of the cleaning device, speed and time commands for the rotary motor of the cleaning device, and pressure and displacement commands for the servo motor; and performing a cleaning operation on the windshield based on the control commands.

[0093] In this embodiment, during the process of controlling the cleaning device to perform a cleaning operation on the windshield based on the control parameters, the control parameters can be converted to generate a sequence of low-level control instructions that the ECU can directly execute, that is, to obtain control instructions.

[0094] Optionally, the above transformation process is not a simple mapping, but rather a discretization of the continuous space-time function into a high-precision timing instruction package. For example, for an X-axis linear servo motor, it is possible to generate displacement and time point sequence instructions for the servo motor of the cleaning device, such as a set of precise point sequences containing displacement-timestamps. For instance, moving from the initial position X=150mm to X=185mm within 0.2 seconds, interpolating 50 control points in between to ensure smooth, shock-free motion.

[0095] Optionally, for the rotary motor, speed and time commands for the cleaning device's rotary motor can be generated based on the speed profile curve V(s), such as a corresponding speed-time command stream. For example, it can run at a constant speed of 2.8Hz for the first 30% of the stroke, then decrease to 1.5Hz exponentially for 1.2 seconds after entering the A-pillar area, and then accelerate back to the original speed, achieving an intelligent rhythm of "high-speed sweeping and low-speed fine wiping". For the Z-axis pressure servo motor, the dynamic pressure curve P(s) of the wiper arm of the cleaning device, which shows the change in wiper blade pressure with stroke, can be mapped to pressure and displacement commands for the servo motor, such as a displacement-pressure command pair. For example, when the wiper blade moves to the A-pillar root area where the windshield curvature is greatest (corresponding to displacement s=65%), the pressure linearly increases from a baseline of 4.5N to 5.8N and remains stable in this range. After entering the flat area, the pressure gradually decreases, ensuring that the wiper blade always fits the curved surface of the glass and avoids suspension or overpressure.

[0096] Optionally, the above three sets of control commands can be synchronously sent to the local ECU of the wiper actuator via high-speed CAN FD or vehicle Ethernet, and the multi-loop servo control system inside the ECU can respond in real time.

[0097] In this embodiment, the windshield is cleaned according to the displacement and time sequence commands of the servo motor of the cleaning device, the speed and time commands of the rotary motor of the cleaning device, and the pressure and displacement commands of the servo motor. This can ensure millimeter-level positioning accuracy and millisecond-level dynamic response in a coordinated manner.

[0098] As an optional implementation, a cleaning operation is performed on the windshield based on control commands, including: controlling the wiper arm of the cleaning device to move to a target position in response to displacement and time sequence commands; driving the wiper arm to move along the surface of the windshield in response to speed and time commands; adjusting the normal pressure between the wiper blade and the surface of the windshield in response to pressure and displacement commands; and performing a cleaning operation on the windshield according to the target position, movement operation, and normal pressure.

[0099] In this embodiment, in response to the X-axis displacement-time sequence command output by the trajectory planning module, the dual-axis linear servo motor can be driven to precisely control the physical displacement of the wiper arm in the horizontal direction, so that the starting point and ending point break through the fixed arc trajectory of traditional mechanical limit and dynamically position to the edge of blind spots that cannot be covered by traditional methods, such as the base of the A-pillar and the upper edge of the glass.

[0100] Optionally, in response to the speed-time command of the rotary motor, the output speed of the main rotary motor can be precisely adjusted using vector control, enabling the wiper arm to achieve non-uniform motion along the glass curvature according to a preset speed profile curve V(s). For example, in areas with dense oil film or sticky stains, the speed is actively reduced to extend the contact time for thorough cleaning; in clean or low-demand areas, the speed is increased to reduce ineffective power consumption and mechanical impact, thus forming an intelligent motion rhythm of "slow sweeping in key areas and fast sweeping in blank areas." At the same time, in response to the pressure-displacement command of the Z-axis servo motor, the normal pressure between the wiper blade and the glass surface can be dynamically adjusted in real time. When the wiper arm enters the A-pillar area or edge transition area where the glass curvature changes drastically, the pressure is automatically increased by 15%–25% to ensure that the rubber strip fully adheres to the complex curved surface and avoids water film residue due to suspension; when moving to the flat central area of ​​the glass, the pressure returns to the reference level, reducing rubber strip wear and operating noise.

[0101] Optionally, the three types of control commands mentioned above—spatial positioning, speed modulation, and pressure adaptation—are coordinated at the millisecond level under the synchronous clock drive of the actuator ECU, forming a highly coupled three-dimensional motion control flow. This ensures that the wiper arm always maintains optimal posture, speed, and pressure against the glass surface during its movement, completing an intelligent cleaning process that is cognitive, strategic, and responsive.

[0102] In this embodiment of the application, the above steps make the cleaning action no longer a mechanical reproduction of a preset program, but a real-time response to the current windshield condition and driving environment.

[0103] As an optional implementation, the method further includes: evaluating the information recognition model in response to the cleaning device completing the cleaning operation and obtaining the evaluation result; and updating the information recognition model based on the evaluation result in response to the vehicle being in a parked charging state or an idle state.

[0104] In this embodiment, after the cleaning device completes the cleaning operation, such as after the wiper actuator completes a wiping action, it does not immediately enter the next trigger wait, but instead initiates an "evaluation-learning" micro-loop. Specifically, the image data after wiping can be captured again to generate a new H(t+1); the actual reward value R_actual of this action can be calculated; and the experience (S_t, A_t, R_actual, S_t+1) can be stored as an experience data package in a finite-capacity cyclic experience replay buffer. Here, S_t can represent the state space at time t, A_t can represent the action space, R_actual can represent the reward function (e.g., the reward value), and S_t+1 can represent the state space at time t+1.

[0105] Optionally, when the vehicle is parked and charging or idle, and the domain controller's computing power is idle, an online learning thread can be started in the background. This thread samples a batch of data from the buffer and performs small gradient updates (fine-tuning) on ​​the policy network of the SAC decision-maker, thereby continuously adapting to the specific operating conditions of the vehicle (such as windshield coating characteristics and common local pollutants), achieving personalized adaptation that improves with use.

[0106] For example, after the wiper actuator completes a wiping action, a new round of visual acquisition can be triggered immediately. The forward-facing camera acquires image data of the wiped windshield, and this image data is input into the same lightweight fully convolutional neural network (FCN) used in the initial training, regenerating a cleanliness requirement heatmap H(t+1) after this operation. Then, this new cleanliness requirement heatmap H(t+1) can be compared pixel-by-pixel with H(t) before the operation. The reduction area of ​​stains in key visual areas (e.g., directly in front of the driver, at the base of the A-pillar, and at the upper edge of the side windows), changes in the distribution of residual stain types, and the degree of elimination in high-demand areas are calculated. This quantifies the actual cleaning effect of the wiping action, forming an objective evaluation result. Next, by determining whether the vehicle is currently in a parked charging or idle state, if the conditions are met, a background online learning thread can be started. This thread encapsulates complete information about the experience, such as the state H(t) before the action, the control parameters A_t, the actual reward R_actual obtained from the evaluation, and the state H(t+1) after the action, into an experience data package and stores it in a loop experience replay buffer. Once the buffer has accumulated enough samples, an improved soft actor-critic (SAC) algorithm can be invoked. Using this experience data from real-world driving scenarios as samples, the information recognition model (i.e., the FCN used to generate the cleanliness requirement heatmap) can be lightly fine-tuned. For example, this could enhance the ability to distinguish reflection interference caused by aging of specific windshield coatings, or optimize the confidence level for recognizing common local insect residue stains.

[0107] In the embodiments of this application, by continuously learning and adapting to the unique characteristics of each vehicle's windshield, such as the attenuation of light transmittance of different batches of coatings, the differences in the composition of pollutants in different regions (e.g., mold in the south, dust in the north), and even the adhesive properties of different brands of wiper blades, the generation of the cleanliness demand heat map can become closer and closer to the real demand, thereby driving the continuous optimization of subsequent decision-making and execution strategies.

[0108] The control method for a vehicle cleaning device in this application embodiment acquires image data of at least one windshield installed in the vehicle, and generates a cleaning demand heat map based on the image data, representing the distribution of stains at different locations on the windshield surface. Then, an information recognition model is invoked to identify the cleaning demand heat map, obtaining the stain type on the windshield, and determining the control parameters of the cleaning device based on the stain type. Since the control parameters include a set of control instruction parameters for the cleaning device to perform cleaning operations according to the cleaning path, the cleaning device can be controlled to perform cleaning operations on the windshield based on the control parameters. This overcomes the limitations of related technologies that only use fixed arc-shaped mechanical paths and open-loop control strategies, resulting in poor windshield cleaning effects. Therefore, it solves the technical problem of not being able to effectively control the vehicle's cleaning device, achieving the technical effect of effectively controlling the vehicle's cleaning device.

[0109] The above technical solutions of the embodiments of this application will be further illustrated below with reference to preferred embodiments.

[0110] Currently, vehicle wiper systems can be divided into two categories: manual speed control and rain-sensing automatic speed control. Rain-sensing systems use optical / capacitive sensors located at the base of the rearview mirror to monitor interference (such as raindrops) in localized areas of the glass, and then adjust the wiping interval and speed according to a preset map. Some vehicles have a wiper memory function, which can reset to the last used wiping setting. The wiper arm drive path of existing systems is strictly limited by mechanical limiters, resulting in a fixed arc trajectory.

[0111] In summary, the relevant technologies have the following shortcomings.

[0112] There is a disconnect between perception and execution. The rain sensor only provides a trigger frequency signal and cannot assess the actual cleaning effect after each wipe (e.g., whether a water film or blind spots remain), resulting in open-loop control.

[0113] Fixed path leads to blind spots. Fixed curved paths cannot clean the base of the A-pillar, areas where the curvature of the glass edge changes abruptly, and special water accumulation areas caused by high-speed wind pressure (such as the upper part of the side window). These areas are precisely the key visibility areas for drivers when changing lanes or turning.

[0114] Lacks adaptive optimization capabilities. The wiper system cannot adjust its wiping strategy (e.g., start and end points, speed, pressure) in real time based on dynamic environmental parameters such as individual glass differences (e.g., coating aging), wiper blade wear, vehicle speed, and wind direction to achieve optimal cleaning performance and minimum power consumption / wear.

[0115] To address the aforementioned issues, this application provides an intelligent windshield wiper system and method deeply integrated into the automotive electronic and electrical architecture. This aims to solve the problems of large blind spots, low cleaning efficiency, and inability to adapt to complex dynamic conditions caused by fixed mechanical paths and open-loop control in existing windshield wiper systems. Specifically, it provides an intelligent system and method capable of real-time observation of wiping performance and, based on this, autonomously and continuously optimizing the start and end positions, speed profile, and blade adhesion pressure of each wiping action. This dynamically maximizes the coverage of the effective visual area of ​​the windshield under various operating conditions.

[0116] Figure 2 This is a schematic diagram of an intelligent windshield wiper system deeply integrated into the automotive electronic and electrical architecture according to an embodiment of this application, such as... Figure 2 As shown, the intelligent wiper system 200 may include a visual perception unit 201, a central processing and decision-making unit 202, and an intelligent wiper actuator assembly 203.

[0117] The visual perception unit 201 can reuse or be dedicated to a forward-facing ADAS camera (which can be located behind the rearview mirror), with a resolution of no less than 1.28 million pixels and a frame rate of over 30fps in rain mode. This camera must have good low-light performance and high dynamic range performance, and its perspective distortion has been calibrated for the windshield to ensure that the image pixel coordinates can be accurately mapped to the physical coordinates of the glass.

[0118] The central processing and decision-making unit 202 is integrated into the vehicle's intelligent driving domain controller or high-performance body domain controller. Hardware-wise, it relies on an internal multi-core system-on-a-chip (SoC) and utilizes neural network processing unit resources.

[0119] The intelligent wiper actuator assembly 203 includes a dual-axis drive mechanism, a rotary drive motor, and an integrated electronic control unit.

[0120] Figure 3 This is a schematic diagram of an intelligent wiper actuator assembly according to an embodiment of this application, as shown below. Figure 3 As shown, the intelligent wiper actuator assembly 300 includes a dual-axis drive mechanism 301, a rotary drive motor 302, and an integrated electronic control unit 303.

[0121] The dual-axis drive mechanism 301 employs two high-precision linear servo motors (e.g., a brushless DC motor paired with a ball screw) to control the movement of the wiper arm in the X-axis (parallel to the lower edge of the glass) and Z-axis (perpendicular to the glass plane), respectively. The X-axis motor is responsible for the stepless adjustment of the start and end points of the wiper arm (the stroke covers the entire reach of the wiper arm), while the Z-axis motor is responsible for dynamically adjusting the pressure of the wiper blade on the glass.

[0122] The rotary drive motor 302 retains but upgrades the traditional rotary motor, adopting a brushless motor with vector control, which can realize arbitrary stepless speed adjustment during the brushing process (not limited to low speed / high speed).

[0123] The integrated electronic control unit 303, which acts as the local controller of the actuator, receives high-level instructions from the domain controller (such as the target trajectory point sequence, speed curve, and pressure curve) and is responsible for the closed-loop control of the underlying servo loop (position loop, speed loop, and current loop) to ensure precise execution of the action.

[0124] Figure 4 This is a flowchart of a smart wiper method deeply integrated into the automotive electronic and electrical architecture according to an embodiment of this application, such as... Figure 4 As shown, the method may include the following steps.

[0125] Step S401: Perform system initialization and calibration.

[0126] In this embodiment, system initialization and calibration can be performed first.

[0127] Step S402: Perform real-time sensing and status analysis.

[0128] In this embodiment, real-time perception and state analysis can be performed through a multimodal fusion perception and state modeling module. This module not only processes images but also integrates vehicle bus data (e.g., vehicle speed, yaw rate, and raw signals from the rain sensor on the CAN FD). The core output is a two-dimensional matrix of a heatmap showing glass cleanliness requirements. The output process is as follows.

[0129] The system acquires raw frames from the camera and applies distortion correction and perspective transformation based on calibration parameters to convert the image to a bird's-eye view with the glass as the reference.

[0130] Run a lightweight fully convolutional neural network (FCN). This network takes the corrected image as input and outputs a classification probability map for each pixel, with categories including: dry glass, uniform water film, discrete water droplets, viscous stains (shellac / mud), and oil film. This network has been pre-trained in the cloud using hundreds of thousands of images covering different weather conditions, lighting, and glass coating states.

[0131] The classification probability is fused with environmental parameters from the bus (e.g., wind pressure model influenced by vehicle speed) to assign a "cleanliness requirement value" between 0 and 1 to each pixel region. For example, the requirement value of "discrete water droplets" above the A-pillar side is weighted and amplified at high speeds; the requirement value of "oil film" in the area directly in front of the driver's seat has the highest weight. Finally, a "demand heatmap" H(t) is generated that corresponds one-to-one with the physical location of the glass.

[0132] Step S403: Does the scraping requirement meet the requirements?

[0133] In this embodiment, it can be determined whether the wiping requirement is met. If yes, then step S404 is executed; otherwise, then step S402 is executed.

[0134] Step S404: Perform dynamic path planning and optimization.

[0135] In this embodiment, an improved soft actor-critic (SAC) algorithm framework can be used for dynamic path planning and optimization. The improved soft actor-critic (SAC) algorithm framework is a model that is trained offline and fine-tuned and inferred online.

[0136] The state space (S) is composed of the current time H(t), the estimated remaining lifespan of the wiper blades, the current vehicle speed, and the historical state of the wiper system since its last action.

[0137] Action Space (A): The output is a tuple of control parameters for a complete scraping action. It is not a simple instruction, but a set of curves that change with time or position.

[0138] Start and end point coordinates: (X_start, Z_start_pressure), (X_end, Z_end_pressure). The Z-coordinate is directly associated with the initial / end pressure. (X_start, Z_start_pressure) represents the starting position information, and (X_end, Z_end_pressure) represents the ending position information.

[0139] Speed ​​profile curve V(s): Defines how the speed of the wiper arm changes with the stroke s as it moves from the start to the end point. For example, "low-speed fine wiping" is used in areas with high demand, while "high-speed sweeping" is used in clean areas.

[0140] Dynamic pressure curve P(s): Defines how the wiper pressure changes with the stroke s. For example, in edge areas with large glass curvature, the pressure is automatically increased by 15-25% to ensure wiper adhesion; in flat areas, the standard pressure is restored to reduce wear and noise.

[0141] Reward Function (R) Design: The reward function guides AI learning. The design principle is comprehensive optimization, including: positive rewards from the reduction in area of ​​high-demand areas (increased coverage) and improved cleanliness of key viewing areas (defined by regulations) in H(t+1) after the current wiping. Negative penalties come from the mechanical wear and tear of the action (such as the weighted sum of high-speed, high-pressure actions), energy consumption, and the continued presence of stubborn stains that were not removed. By adjusting the weighting coefficients of these rewards and penalties, the AI ​​learns to find an optimal balance between "cleaning effectiveness," "energy consumption," and "mechanical wear and tear."

[0142] Step S405: Generate and issue control commands.

[0143] In this embodiment, the action tuple output by the decision-maker can be transformed into a detailed sequence of instructions executable by the trajectory planning submodule. The control instructions include displacement-time sequences for the X-axis servo motor, speed-time commands for the rotary motor, and pressure-displacement mapping commands for the Z-axis servo motor. These instructions are transmitted to the wiper actuator ECU via automotive Ethernet or high-speed CAN FD.

[0144] Step S406: The intelligent actuator performs the action.

[0145] In this embodiment, after the control command is issued, the intelligent actuator performs the scraping action.

[0146] Step S407: Conduct effect evaluation and experience storage.

[0147] In this embodiment, after each scraping action, the system does not immediately enter the next trigger wait, but instead initiates an "evaluation-learning" micro-loop. Specifically, the image after scraping is captured again, generating a new H(t+1). The actual reward value R_actual of this action is calculated. The experience (S_t, A_t, R_actual, S_t+1) is stored as an "experience data package" in a limited-capacity circular experience replay buffer. When the vehicle is in a parked charging or idle state, and the domain controller's computing power is idle, an online learning thread is started in the background. This thread samples a batch of data from the buffer and performs small gradient updates (fine-tuning) on ​​the policy network of the SAC decision-maker. This allows the system to continuously adapt to the specific operating conditions of the vehicle (such as glass coating characteristics and common local pollutants), achieving personalized adaptation that "gets better with use."

[0148] Step S408: Does the online learning condition meet?

[0149] In this embodiment, it can be determined whether the conditions for online learning are met. If yes, then step S409 is executed; otherwise, step S402 is executed.

[0150] Step S409: Perform incremental learning of the background model.

[0151] In this embodiment, if the conditions for online learning are met, incremental learning of the background model will be performed.

[0152] Through the above steps, the embodiments of this application achieve the following technical effects.

[0153] Eliminating dynamic blind spots and enhancing active safety: By dynamically adjusting the start and end points, it can effectively cover critical blind spots such as the base of the A-pillar and the edge of the glass that cannot be reached by traditional fixed paths. Especially when changing lanes at high speeds or driving on curves, it can provide drivers with a more complete lateral view, directly improving driving safety.

[0154] Maximizing coverage and cleaning efficiency: The system can identify the density of dirt distribution and perform targeted cleaning in a single wipe using variable speed and pressure, avoiding wasted effort in clean areas. Real-world testing shows that, with the same time and energy consumption, it can effectively increase cleaning coverage by 20-35% for irregularly distributed moderate to heavy rain.

[0155] Extending system lifespan and reducing operating costs: Dynamic pressure regulation avoids applying excessively high pressure over long periods in flat areas, reducing the probability of abnormal blade wear (such as rubber strip flipping and localized blade aging) by approximately 40%. The optimized speed curve also reduces impact and fatigue on the motor and transmission mechanism.

[0156] With adaptive and personalized capabilities: The online learning mechanism enables the system to self-optimize, adapt to individual vehicle differences (such as glass curvature and wiper brand) and the road conditions commonly encountered by car owners (such as dusty and insect-infested areas), and provide a continuously evolving user experience.

[0157] Greater system integration and cost advantages: Existing ADAS cameras and domain controller hardware are deeply reused; the main incremental cost lies in the upgraded intelligent actuators. Compared to the resulting safety and user experience benefits, it offers excellent cost-effectiveness.

[0158] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0159] According to an embodiment of this application, a control system embodiment for a cleaning device in a vehicle is provided. Figure 5 This is a schematic diagram of a control system for a cleaning device in a vehicle according to an embodiment of this application, such as... Figure 5As shown, the control system 500 of the cleaning device in the vehicle may include: a vision perception unit 502, a central processing and decision-making unit 504, and an intelligent wiper actuator 506.

[0160] The visual perception unit 502 is used to acquire image data of at least one windshield installed in the vehicle.

[0161] The central processing and decision-making unit 504 is used to generate a heat map of the cleanliness requirements of the windshield based on image data. The heat map of cleanliness requirements is used to characterize the distribution of stains at different locations on the surface of the windshield. The unit calls an information recognition model to identify the type of stains on the windshield and, based on the stain type, determines the control parameters of the cleaning device. The information recognition model is obtained by training a deep reinforcement learning model using image data samples of windshield samples from vehicle samples and cleanliness requirements heat map samples of windshield samples. The control parameters include a set of control command parameters for the cleaning device to perform cleaning operations according to the cleaning path.

[0162] The intelligent wiper actuator 506 is used to control the cleaning device to perform a cleaning operation on the windshield based on control parameters.

[0163] In this embodiment, image data of at least one windshield installed in a vehicle is acquired, and a cleaning demand heatmap is generated based on the image data, representing the distribution of stains at different locations on the windshield surface. Then, an information recognition model is invoked to identify the cleaning demand heatmap, determining the stain type on the windshield, and based on the stain type, determining the control parameters of the cleaning device. Since the control parameters include a set of control instruction parameters for the cleaning device to perform cleaning operations according to the cleaning path, the cleaning device can be controlled to perform cleaning operations on the windshield based on these parameters. This overcomes the limitations of related technologies that only use fixed arc-shaped mechanical paths and open-loop control strategies, resulting in poor windshield cleaning effects. Therefore, it solves the technical problem of not being able to effectively control the vehicle's cleaning device, achieving the technical effect of effectively controlling the vehicle's cleaning device.

[0164] According to an embodiment of this application, a control device for a cleaning device in a vehicle is provided. It should be noted that the control device 600 for the cleaning device in the vehicle can be used to execute the control method for the cleaning device in the vehicle described above.

[0165] Figure 6 This is a schematic diagram of a control device for a cleaning device in a vehicle according to an embodiment of this application, such as... Figure 6 As shown, the control device 600 of the cleaning device in the vehicle may include: an acquisition unit 602, a generation unit 604, an identification unit 606, and a control unit 608.

[0166] Acquisition unit 602 is used to acquire image data of at least one windshield installed in a vehicle.

[0167] The generation unit 604 is used to generate a heat map of the cleanliness requirements of the windshield based on image data, wherein the heat map of cleanliness requirements is used to characterize the distribution of stains at different locations on the surface of the windshield.

[0168] The identification unit 606 is used to call the information identification model to identify the cleanliness requirement heatmap, obtain the stain type of the stain on the windshield, and determine the control parameters of the cleaning device based on the stain type. The information identification model is obtained by training a deep reinforcement learning model using image data samples of windshield samples in the vehicle sample and cleanliness requirement heatmap samples of the windshield samples. The control parameters include a set of control instruction parameters for the cleaning device to perform cleaning operations according to the cleaning path.

[0169] The control unit 608 is used to control the cleaning device to perform a cleaning operation on the windshield based on control parameters.

[0170] Furthermore, the generation unit 604 includes: an input subunit for inputting image data into a fully convolutional neural network to obtain a classification probability map of each pixel in the image data, wherein the classification probability map is used to represent the confidence distribution corresponding to the cleaning intensity of each pixel on the windshield; and a generation subunit for generating a cleaning demand heatmap based on the classification probability map and the vehicle's bus data, wherein the bus data is used to represent the dynamic operating condition information of the vehicle under the current driving state.

[0171] Furthermore, the generation subunit includes: an alignment subunit, used to spatially align the classification probability map and the vehicle's bus data to obtain aligned classification probability map and bus data, wherein the bus data is used to represent the dynamic operating condition information of the vehicle under the current driving state; a fusion subunit, used to perform weighted fusion of the aligned classification probability map and bus data to obtain the cleanliness requirement value corresponding to the pixel region of each pixel; and a first generation subunit, used to generate a cleanliness requirement heatmap based on the cleanliness requirement value.

[0172] Furthermore, the fusion subunit includes: an association subunit, used to associate the confidence level of the cleaning intensity of each pixel in the classification probability map with parameters reflecting the current driving state in the bus data to obtain the weight information of the cleaning intensity, wherein the weight information is used to represent the importance of cleaning the windshield under different types of stains; a summation subunit, used to perform a weighted summation of the confidence level and the weight information to obtain the cleanliness requirement value; the control device 600 of the cleaning device in the vehicle also includes: an adjustment subunit, used to adjust the weight information in response to detecting that the number of cleaning operations performed is greater than the number threshold and the cleanliness requirement value is greater than the cleanliness requirement value threshold.

[0173] Further, the first generation subunit includes: a processing subunit for normalizing the cleanliness requirement value to obtain a normalized cleanliness requirement value; a filtering subunit for filtering the normalized cleanliness requirement value to obtain a filtered cleanliness requirement value; a conversion subunit for converting the filtered cleanliness requirement value into a two-dimensional heat map according to the position of the cleanliness requirement value in the coordinate system of the windshield; and a determination subunit for determining the two-dimensional heat map as a cleanliness requirement heat map.

[0174] Further, the stain types include at least one of the following: dry stain type, uniform water film stain type, discrete water droplet stain type, viscous stain type, and oil film stain type. The identification unit 606 includes: an extraction subunit, used to extract nonlinear features from the spatial distribution of the cleaning demand heatmap to obtain a demand region, wherein the demand region is used to characterize the region where the priority of cleaning the windshield is higher than the priority threshold; and a second generation subunit, used to generate control parameters based on the demand region and the stain type, wherein the control parameters include the starting position information and ending position information of the cleaning device, the velocity profile curve along the cleaning path, and the dynamic pressure curve of the wiper arm of the cleaning device as the stroke changes. The starting position information is used to indicate the physical position where the wiper arm begins cleaning in the horizontal direction of the windshield, the ending position information is used to indicate the physical position where the wiper arm ends cleaning in the horizontal direction of the windshield, the velocity profile curve is used to indicate the instantaneous movement speed change of the wiper arm at different spatial positions along the cleaning path, and the dynamic pressure curve is used to indicate the distribution characteristics of the clamping force applied to the windshield surface by the wiper blade as the position changes during the movement along the cleaning path.

[0175] Furthermore, the acquisition unit 602 includes: a subunit for acquiring initial image data collected by the acquisition device in the vehicle, wherein the initial image data includes visual representations of optical reflection, stain distribution, and ambient lighting conditions on the windshield surface; and a correction subunit for correcting and / or performing perspective correction on the initial image data to obtain image data, wherein the image data is a bird's-eye view image based on the windshield.

[0176] Furthermore, the control unit 608 includes: a conversion subunit for converting control parameters to obtain control instructions, wherein the control instructions include at least one of the following: a displacement and time sequence instruction for the servo motor of the cleaning device, a speed and time instruction for the rotary motor of the cleaning device, and a pressure and displacement instruction for the servo motor; and a first execution subunit for performing a cleaning operation on the windshield based on the control instructions.

[0177] Further, the first execution subunit includes: a control subunit for controlling the wiper arm of the cleaning device to move to a target position in response to displacement and time sequence commands; a drive subunit for driving the wiper arm to move along the surface of the windshield in response to speed and time commands; an adjustment subunit for adjusting the normal pressure between the wiper blade and the surface of the windshield in response to pressure and displacement commands; and a second execution subunit for performing a cleaning operation on the windshield according to the target position, movement operation, and normal pressure.

[0178] Furthermore, the control device 600 of the cleaning device in the vehicle also includes: an evaluation subunit, used to evaluate the information recognition model and obtain evaluation results in response to the cleaning device completing the cleaning operation; and an update subunit, used to update the information recognition model based on the evaluation results in response to the vehicle being in a parking charging state or an idle state.

[0179] In the control device of the vehicle cleaning device in this embodiment, the acquisition unit 602 acquires image data of at least one windshield installed in the vehicle; the generation unit 604 generates a cleaning requirement heat map of the windshield based on the image data, wherein the cleaning requirement heat map is used to characterize the distribution state of stains at different positions on the surface of the windshield; the identification unit 606 calls the information identification model to identify the cleaning requirement heat map, obtain the stain type of the stains on the windshield, and determine the control parameters of the cleaning device based on the stain type, wherein the information identification model is obtained by training a deep reinforcement learning model using image data samples of windshield samples in the vehicle sample and cleaning requirement heat map samples of windshield samples, and the control parameters include a set of control instruction parameters for the cleaning device to perform cleaning operations according to the cleaning path; the control unit 608 controls the cleaning device to perform cleaning operations on the windshield based on the control parameters, thereby solving the technical problem of not being able to effectively control the cleaning device of the vehicle to perform cleaning, and achieving the technical effect of effectively controlling the cleaning device of the vehicle to perform cleaning.

[0180] Embodiments of this application also provide a vehicle, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods described in various embodiments of this application when it runs.

[0181] Embodiments of this application also provide a computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0182] Embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0183] Embodiments of this application also provide a computer program product, including a non-volatile computer-readable storage medium for storing a computer program that, when executed by a processor, implements the methods in various embodiments of this application.

[0184] Embodiments of this application also provide a computer program that, when executed by a processor, implements the methods described in the various embodiments of this application.

[0185] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0186] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0187] 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 units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0188] Furthermore, the functional units in the various embodiments of this application 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. The integrated unit can be implemented in hardware or as a software functional unit.

[0189] If the integrated unit 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 application, in essence, or the part that contributes to the prior art, or all or 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 described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0190] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A control method for a cleaning device in a vehicle, characterized in that, include: Acquire image data of at least one windshield installed in the vehicle; Based on the image data, a heat map of the cleanliness requirements of the windshield is generated, wherein the heat map of the cleanliness requirements is used to characterize the distribution of stains at different locations on the surface of the windshield. The information recognition model is invoked to identify the cleanliness requirement heatmap, thereby obtaining the stain type of the stain on the windshield, and based on the stain type, the control parameters of the cleaning device are determined. The information recognition model is obtained by training a deep reinforcement learning model using image data samples of windshield samples in the vehicle sample and cleanliness requirement heatmap samples of the windshield samples. The control parameters include a set of control instruction parameters for the cleaning device to perform cleaning operations according to the cleaning path. Based on the control parameters, the cleaning device is controlled to perform a cleaning operation on the windshield.

2. The method according to claim 1, characterized in that, Based on the image data, a heat map of the cleanliness requirements of the windshield is generated, including: The image data is input into a fully convolutional neural network to obtain a classification probability map of each pixel in the image data, wherein the classification probability map is used to represent the confidence distribution corresponding to the cleaning intensity of each pixel on the windshield; Based on the classification probability map and the vehicle's bus data, a cleanliness requirement heatmap is generated, wherein the bus data is used to represent the dynamic operating condition information of the vehicle under the current driving state.

3. The method according to claim 2, characterized in that, Based on the classification probability map and the vehicle's bus data, the cleanroom demand heatmap is generated, including: The classification probability map and the vehicle bus data are spatially aligned to obtain the aligned classification probability map and the bus data, wherein the bus data is used to represent the dynamic operating condition information of the vehicle in the current driving state. The aligned classification probability map and the bus data are weighted and fused to obtain the cleanliness requirement value corresponding to the pixel region of each pixel; Based on the cleanliness requirement value, a cleanliness requirement heatmap is generated.

4. The method according to claim 3, characterized in that, The aligned classification probability map and the bus data are weighted and fused to obtain the cleanliness requirement value corresponding to the pixel region of each pixel, including: The confidence level of the cleaning intensity of each pixel in the classification probability map is associated with the parameters reflecting the current driving state in the bus data to obtain the weight information of the cleaning intensity, wherein the weight information is used to represent the importance of cleaning the windshield under different types of stains. The confidence level and the weight information are weighted and summed to obtain the cleanliness requirement value; The method further includes: In response to the detection that the number of cleaning operations performed exceeds a threshold and the cleanliness requirement value exceeds the cleanliness requirement value threshold, the weight information is adjusted.

5. The method according to claim 3, characterized in that, Based on the cleanliness requirement value, a cleanliness requirement heatmap is generated, including: The cleanliness requirement value is normalized to obtain the normalized cleanliness requirement value. The normalized cleanliness requirement value is filtered to obtain the filtered cleanliness requirement value. Based on the position of the cleanliness requirement value in the coordinate system of the windshield, the filtered cleanliness requirement value is converted into a two-dimensional heat map. The two-dimensional heat map is determined as the cleanliness requirement heat map.

6. The method according to claim 1, characterized in that, The stain type includes at least one of the following: dry stain type, uniform water film stain type, discrete water droplet stain type, viscous stain type, and oil film stain type. Based on the stain type, the control parameters of the cleaning device are determined, including: Nonlinear feature extraction is performed on the spatial distribution of the cleanliness demand heatmap to obtain the demand region, wherein the demand region is used to characterize the region where the priority of cleaning the windshield is higher than the priority threshold. Based on the required area and the type of stain, the control parameters are generated. These control parameters include the starting position information and ending position information of the cleaning device, a velocity profile curve along the cleaning path, and a dynamic pressure curve showing the change in wiper blade pressure of the cleaning device's wiper arm with its stroke. The starting position information indicates the physical position at which the wiper arm begins cleaning in the horizontal direction of the windshield. The ending position information indicates the physical position at which the wiper arm ends cleaning in the horizontal direction of the windshield. The velocity profile curve indicates the instantaneous velocity change of the wiper arm at different spatial positions along the cleaning path. The dynamic pressure curve indicates the distribution characteristics of the clamping force applied to the windshield surface by the wiper blade as it moves along the cleaning path, varying with position.

7. The method according to claim 1, characterized in that, Acquiring image data of at least one windshield installed in the vehicle, including: Acquire initial image data collected by the acquisition device in the vehicle, wherein the initial image data includes visual representations of optical reflection, stain distribution, and ambient lighting conditions on the windshield surface; The initial image data is corrected and / or perspective is adjusted to obtain the image data, wherein the image data is a bird's-eye view image based on the windshield.

8. The method according to claim 1, characterized in that, Based on the control parameters, the cleaning device is controlled to perform a cleaning operation on the windshield, including: The control parameters are converted to obtain control commands, wherein the control commands include at least one of the following: displacement and time sequence commands for the servo motor of the cleaning device, speed and time commands for the rotary motor of the cleaning device, and pressure and displacement commands for the servo motor; Based on the control command, a cleaning operation is performed on the windshield.

9. The method according to claim 8, characterized in that, Based on the control command, a cleaning operation is performed on the windshield, including: In response to the displacement and time sequence command, the wiper arm of the cleaning device is controlled to move to the target position; In response to the speed and time commands, the wiper arm is driven to move along the surface of the windshield. In response to the pressure and displacement commands, the normal pressure between the wiper blade and the surface of the windshield is adjusted. The windshield is cleaned according to the target position, the movement operation, and the normal pressure.

10. The method according to any one of claims 1 to 9, characterized in that, The method further includes: In response to the cleaning device completing the cleaning operation, the information recognition model is evaluated to obtain the evaluation result; In response to the vehicle being in a parked charging state or an idle state, the information recognition model is updated based on the evaluation results.

11. A control system for a cleaning device in a vehicle, characterized in that, include: A visual sensing unit is used to acquire image data of at least one windshield installed in the vehicle; A central processing and decision-making unit is used to generate a heat map of the cleanliness requirements of the windshield based on the image data, wherein the heat map of cleanliness requirements is used to characterize the distribution of stains at different locations on the surface of the windshield; to call an information recognition model to identify the heat map of cleanliness requirements to obtain the stain type of the stains on the windshield, and to determine the control parameters of the cleaning device based on the stain type, wherein the information recognition model is obtained by training a deep reinforcement learning model using image data samples of windshield samples in vehicle samples and cleanliness requirements heat map samples of windshield samples; the control parameters include a set of control instruction parameters for the cleaning device to perform cleaning operations according to the cleaning path. A smart wiper actuator is used to control the cleaning device to perform a cleaning operation on the windshield based on the control parameters.

12. A control device for a cleaning system in a vehicle, characterized in that, include: The acquisition unit is used to acquire image data of at least one windshield installed in the vehicle; The generation unit is used to generate a heat map of the cleanliness requirements of the windshield based on the image data, wherein the heat map of cleanliness requirements is used to characterize the distribution of stains at different locations on the surface of the windshield. The identification unit is used to call the information identification model to identify the cleanliness requirement heatmap, obtain the stain type of the stain on the windshield, and determine the control parameters of the cleaning device based on the stain type. The information identification model is obtained by training a deep reinforcement learning model using image data samples of windshield samples in the vehicle sample and cleanliness requirement heatmap samples of the windshield samples. The control parameters include a set of control instruction parameters for the cleaning device to perform cleaning operations according to the cleaning path. The control unit is used to control the cleaning device to perform a cleaning operation on the windshield based on the control parameters.

13. A vehicle, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the method according to any one of claims 1 to 10.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the storage medium is located to perform the method according to any one of claims 1 to 10.

Images