Automotive control system and method

Abstract

The present application relates to the technical field of automobile control, and particularly relates to an automobile control system and method, which first collects millimeter wave radar data, uses a radio frequency enhancement component to improve lateral spatial resolution, and generates a two-dimensional obstacle heat map; secondly, according to the vehicle-obstacle distance, an emergency level is divided, a time domain analysis window is dynamically intercepted, and a mapping library is established in combination with the obstacle material, historical failure records and speed peak value; subsequently, rain signal attenuation compensation is fused, and a collision risk index is output; further, according to the risk index and the speed peak value, a steering mechanism damping response function is reconstructed, and a collision avoidance path correction is performed; finally, steering motor vibration characteristics are extracted in real time to back-propagate torque distribution, and the damping function is closed-loop corrected. Through high-precision perception, dynamic risk assessment and full-closed-loop feedback control, the method effectively overcomes environmental interference, and realizes precise, smooth and high-robust intelligent obstacle avoidance in complex scenarios.

Description

Technical Field

[0001] This invention relates to the field of automotive control technology, and more particularly to an automotive control system and method. Background Technology

[0002] With the rapid development of intelligent driving technology, automatic parking systems, as an important component of advanced driver assistance systems (ADAS), have been widely used in mid-to-high-end passenger vehicles. Automatic parking enables the vehicle to automatically identify parking spaces and complete parking maneuvers without the driver needing to manually operate the steering wheel, accelerator, and brake, greatly improving driving convenience and safety. However, in complex urban parking environments, vehicles are often surrounded by static obstacles (such as pillars, walls, and low curbs) and dynamic disturbances (such as pedestrians, non-motorized vehicles, and other moving vehicles). These obstacles are diverse in shape and their movement is unpredictable, posing significant challenges to path planning and obstacle avoidance control during automatic parking.

[0003] Existing automatic parking systems mostly rely on the fusion of ultrasonic radar and cameras to perceive environmental information, but they have significant limitations in terms of perception accuracy, anti-interference ability, and environmental adaptability. For example, ultrasonic sensors have short detection ranges and low angular resolution, making it difficult to accurately identify small obstacles; cameras are easily affected by factors such as changes in lighting and rainy / foggy weather, leading to recognition failures. In addition, traditional control systems typically use fixed thresholds to judge collision risk, lacking the ability to comprehensively model the material characteristics, movement trends, and external environmental disturbances of obstacles, resulting in delayed or overreacting obstacle avoidance responses, affecting parking smoothness and safety.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide an automotive control system and method that aims to solve the technical problems of existing automatic parking systems, which suffer from insufficient perception accuracy, weak anti-interference ability, and lack of comprehensive modeling and closed-loop adaptive adjustment mechanisms for complex environmental factors. These problems result in delayed or misjudged obstacle avoidance responses in dynamic and changing scenarios, making it difficult to balance parking safety and smoothness.

[0006] To achieve the above objectives, the present invention provides an automobile control method, the method comprising: The system collects millimeter-wave radar data containing the outline and shape features of obstacles and the movement information of surrounding vehicles during the automatic parking process. The spatial resolution of the millimeter-wave radar data in the lateral direction of the parking path is improved by the radio frequency enhancement component, and a two-dimensional obstacle thermal distribution map of the parking area is generated. Based on the real-time relative distance between the vehicle and surrounding obstacles, the emergency level of the corresponding obstacle avoidance behavior is divided, and a time-domain analysis window corresponding to the obstacle heat map is dynamically extracted according to the emergency level. Establish a mapping relationship library between obstacle material type and historical parking obstacle avoidance failure records, and associate it with the peak speed of surrounding moving obstacles to obtain the associated mapping relationship library; Based on the mapping relationship library, the signal baseline offset of the obstacle thermal distribution map is compensated by integrating the attenuation of electromagnetic waves in rainy weather. The compensated obstacle thermal distribution map and the time domain analysis window are processed by the rule decision module, and the collision risk index of the obstacle is output. Based on the collision risk index and the peak speed, the damping response function of the parking steering mechanism is reconstructed, and the collision avoidance path correction action of the automatic parking process is executed. The vibration characteristics of the steering motor are extracted in real time during the parking and steering process. The torque distribution between the wheels and the steering mechanism is then derived. The damping response function is corrected in a closed loop. The intelligent control method for obstacle avoidance in the automatic parking process is realized through closed-loop control of all steps.

[0007] Optionally, the step of reconstructing the damping response function of the parking steering mechanism based on the collision risk index and the peak speed, and executing the collision avoidance path correction action in the automatic parking process, includes: The collision risk index is obtained to determine the risk level. When the risk level is greater than a first preset threshold, the parking speed is reduced. When the risk level is less than a second preset threshold, the normal parking speed is maintained. The peak speed is obtained, and the response delay of the steering mechanism is adjusted according to a preset adjustment rule based on the magnitude of the peak speed. The collision risk index and the peak speed are input into a predefined function to generate a new damping coefficient. The motion resistance of the steering mechanism is updated using the new damping coefficient, and the damping response function of the steering mechanism is reconstructed. Based on the reconstructed damping response function, the steering motor of the automatic parking system performs a collision avoidance path correction action.

[0008] Optionally, the step of extracting the vibration characteristics of the steering motor in real time during parking and steering, back-deriving the torque distribution between the wheels and the steering mechanism, and correcting the damping response function in a closed loop includes: During the parking and steering process, the vibration characteristics of the motor, including the amplitude and frequency of the vibration, are collected in real time by a torque sensor. The vibration characteristics are then input into a pre-stored correspondence model, and the real-time torque distribution value of the wheel steering is output. Based on the torque distribution value, regions with uneven torque or overload are identified, and the damping coefficient of the damping response function is corrected according to the preset adjustment rules. The modified damping response function is updated and applied in real time to control the parking and steering motion in a closed loop.

[0009] Optionally, the step of improving the spatial resolution of the millimeter-wave radar data in the lateral direction of the parking path through the radio frequency enhancement component to generate a two-dimensional obstacle thermal distribution map of the parking area includes: The focusing antenna and filtering module with radio frequency enhancement components are used to process the acquired millimeter-wave radar data, which includes the outline and morphological features of obstacles and information on surrounding moving obstacles. By adjusting the radio frequency transmission and reception path, the difference in echo signal strength between lateral adjacent points of the parking path is increased, thereby improving the spatial resolution of the processed millimeter-wave radar data. The millimeter-wave radar data with improved spatial resolution is converted into a two-dimensional grid representation, and the echo intensity value of each location is displayed along the lateral direction of the parking path to generate a two-dimensional obstacle thermal distribution map of the parking area.

[0010] Optionally, the step of dynamically extracting a time-domain analysis window corresponding to the obstacle thermal distribution map based on the emergency level includes: Based on the classification of emergency levels, obstacle course movement is divided into emergency movement level or normal movement level; When the emergency level is emergency movement level, a short time-domain analysis window is selected; when the emergency level is normal movement level, a long time-domain analysis window is selected. The time-domain analysis window covers the time-domain change data of the obstacle heat distribution map, and the window length determines the time range for analyzing the obstacle heat distribution map.

[0011] Optionally, the step of establishing a mapping relationship library between obstacle material types and historical parking obstacle avoidance failure records, and associating it with the peak speeds of surrounding moving obstacles to obtain the associated mapping relationship library, includes: Collect obstacle material type data and historical parking obstacle avoidance failure records, and associate the material type with the historical parking obstacle avoidance failure records to establish an initial mapping relationship library; Add the peak velocity of the surrounding moving obstacles as an additional dimension to the initial mapping relation library; For different obstacle material types and peak speeds, the frequency of corresponding obstacle avoidance failures is associated with and stored in the initial mapping relationship library, and the associated mapping relationship library containing the material type, the peak speed, and the historical parking obstacle avoidance failure records is output.

[0012] Optionally, the step of processing the compensated obstacle thermal distribution map and the time-domain analysis window using the rule-based decision module to output the obstacle collision risk index includes: The rule-based decision-making module extracts the echo intensity change value in the lateral direction of the parking path from the compensated obstacle thermal distribution map; Extract the characteristics of the rate of change of echo intensity within the corresponding time range from the time domain analysis window; The system analyzes the changes in echo intensity and the rate of change using preset rules. Based on the results of the analysis, it adjusts the collision risk index by increasing or decreasing it, and finally outputs the adjusted collision risk index representing the probability of a collision.

[0013] Furthermore, to achieve the above objectives, the present invention also provides an automotive control system, the system comprising: The perception enhancement module is used to collect millimeter-wave radar data containing the outline and shape features of obstacles and the movement information of surrounding vehicles during the automatic parking process. The spatial resolution of the millimeter-wave radar data in the lateral direction of the parking path is improved by the radio frequency enhancement component, and a two-dimensional obstacle thermal distribution map of the parking area is generated. The dynamic window module is used to classify the emergency level of the obstacle avoidance behavior based on the real-time relative distance between the vehicle and surrounding obstacles, and dynamically extract the time domain analysis window corresponding to the obstacle heat distribution map according to the emergency level. The mapping and association module is used to establish a mapping relationship library between obstacle material type and historical parking obstacle avoidance failure records, and to associate the peak speed of surrounding moving obstacles to obtain the associated mapping relationship library; The compensation decision module is used to compensate for the signal baseline shift of the obstacle thermal distribution map by integrating the attenuation of electromagnetic waves in rainy weather based on the mapping relationship library. The rule decision module processes the compensated obstacle thermal distribution map and the time domain analysis window, and outputs the collision risk index of the obstacle. The damping reconstruction module is used to reconstruct the damping response function of the parking steering mechanism based on the collision risk index and the peak speed, and to perform the anti-collision path correction action in the automatic parking process. The closed-loop correction module is used to extract the vibration characteristics of the steering motor in real time during the parking and steering process, back-derive the torque distribution of the wheels and steering mechanism, and correct the damping response function in a closed loop. Through closed-loop control of all steps, an intelligent control method for obstacle avoidance in the automatic parking process is realized.

[0014] In addition, to achieve the above objectives, the present invention also provides an automobile control device, the device comprising: a memory, a processor, and an automobile control program stored in the memory and executable on the processor, the automobile control program being configured to implement the steps of the automobile control method as described in any one of the above descriptions.

[0015] In addition, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a vehicle control program, which, when executed by a processor, implements the steps of the vehicle control method as described in any one of the above descriptions.

[0016] This invention provides a vehicle control method that, by introducing a radio frequency enhancement component, specifically improves the spatial resolution of millimeter-wave radar in the lateral direction of the parking path, enabling clearer identification of obstacle contours and generating high-precision two-dimensional obstacle thermal distribution maps. Simultaneously, addressing the electromagnetic wave attenuation problem caused by inclement weather such as rain, a signal baseline offset compensation mechanism based on a mapping relation library is established, effectively overcoming environmental interference and ensuring the accuracy and reliability of perception data under low visibility or severe weather conditions. This method abandons the traditional fixed threshold judgment mode, instead dynamically classifying the urgency level based on the real-time distance between the vehicle and obstacles, and adaptively adjusting the length of the time-domain analysis window accordingly (gradual urgency processing). Combining a multi-dimensional mapping relation library of obstacle material type, historical failure records, and peak speeds of surrounding vehicles, the collision risk index can be quantified more accurately. This multi-dimensional fusion analysis method prevents overreaction to stationary or slow-moving objects while ensuring rapid response to high-speed approaching objects, significantly reducing the false alarm and false negative rates of the obstacle avoidance system. Based on the calculated collision risk index and peak speed, the system can reconstruct the damping response function of the parking steering mechanism in real time. This means that the vehicle's steering resistance is no longer fixed, but dynamically adjusted according to the level of risk. For example, damping is increased to suppress sharp turns in high-risk situations, and damping is reduced to ensure smoothness in low-risk situations. This proactive damping adjustment strategy makes path correction actions more delicate and smooth, avoiding the sudden braking and turning phenomena common in traditional systems, significantly improving passenger comfort and vehicle control stability. Innovatively, closed-loop feedback based on the vibration characteristics of the steering motor is introduced into the execution phase. By extracting vibration characteristics in real time to infer the torque distribution between the wheels and the steering mechanism, the system can instantly detect deviations between actual execution and the expected model, such as sudden torque changes caused by uneven road surfaces, and accordingly correct the damping response function. This fully closed-loop control logic of perception, decision-making, execution, feedback, and re-correction greatly enhances the system's adaptability to unknown disturbances and mechanical nonlinear factors, ensuring high reliability and safety of the automatic parking process under extreme conditions. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating an embodiment of the vehicle control method of the present invention; Figure 2 This is a structural block diagram of an embodiment of the vehicle control system of the present invention.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] Reference Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the vehicle control method of the present invention, which presents an embodiment of the vehicle control method of the present invention.

[0021] In one embodiment, the vehicle control method includes: Step S100: Collect millimeter-wave radar data containing obstacle outline features and surrounding vehicle movement information during automatic parking. Improve the spatial resolution of the millimeter-wave radar data in the lateral direction of the parking path through the radio frequency enhancement component, and generate a two-dimensional obstacle thermal distribution map of the parking area.

[0022] Millimeter-wave radar can be an active sensor that uses millimeter-wave electromagnetic waves to detect the distance, speed, and angle of obstacles in the surrounding environment. It can be used to collect environmental perception data containing obstacle contour features and surrounding vehicle motion information. For example, millimeter-wave radar can include, but is not limited to, one or more of 77GHz automotive millimeter-wave radar, 24GHz short-range millimeter-wave radar, and MIMO architecture millimeter-wave radar. Obstacle contour features can be a comprehensive representation of the obstacle's geometric boundaries and surface reflection characteristics in space, which can be used to distinguish static obstacle types such as pillars, walls, and curbs, improving recognition accuracy. Surrounding vehicle motion information can be the speed, acceleration, and trajectory trend information of nearby moving obstacles, which can be used to assess the potential collision threat of dynamic interference sources. Millimeter-wave radar data can be the raw echo signal collected by millimeter-wave radar or a set of distance-velocity-angle information after preliminary processing, which can be used as the original source of obstacle contour features and surrounding vehicle motion information. The lateral direction of the parking path can be a horizontal axis perpendicular to the vehicle's direction of travel, i.e., the left and right sides of the vehicle, which can be used as a key dimension for optimizing spatial resolution by radio frequency enhancement components, directly affecting the ability to recognize lateral obstacles. Spatial resolution is the minimum distance a sensor can distinguish between two adjacent targets in a specific direction, and it can be used to determine whether millimeter-wave radar can accurately distinguish small or dense obstacles.

[0023] The radio frequency (RF) enhancement component can be a hardware or algorithm module used to enhance the RF signal processing capabilities of millimeter-wave radar. Specifically optimized for lateral spatial resolution performance, it can improve the lateral spatial resolution of millimeter-wave radar in the parking path, thereby more clearly identifying obstacle contours. Furthermore, the RF enhancement component can enhance the lateral echo signal at the millimeter-wave radar receiver by introducing phase modulation, multi-channel synthesis, or beamforming techniques. In an exemplary embodiment, the RF enhancement component can work in conjunction with the millimeter-wave radar to enhance the lateral resolution of its raw output data; its output serves as the basic input for generating a two-dimensional obstacle thermal distribution map of the parking area. Exemplarily, the RF enhancement component can include, but is not limited to, one or more of a multi-channel beam combiner, a lateral phase compensation unit, and an RF signal gain regulator. The two-dimensional obstacle thermal distribution map of the parking area can be a graphical data structure representing the probability or intensity distribution of obstacles within the parking area in a two-dimensional plane. It can be used to visually present the spatial distribution and contour of obstacles within the parking area, providing basic perceptual input for subsequent risk assessment. In one specific embodiment, a two-dimensional obstacle thermal distribution map of the parking area can be generated through spatial mapping and density clustering based on high-resolution millimeter-wave radar data processed by radio frequency enhancement components.

[0024] Collecting millimeter-wave radar data containing obstacle contour features and surrounding vehicle motion information during automatic parking can be achieved by continuously activating the millimeter-wave radar and reading its raw or pre-processed data streams after automatic parking is initiated. This operation can be further implemented through continuous polling of the radar interface or an event-driven approach, thereby obtaining highly timely environmental perception input to support subsequent obstacle recognition and risk assessment. Improving the spatial resolution of millimeter-wave radar data in the lateral direction of the parking path using radio frequency enhancement components can be achieved by performing lateral phase correction or multi-channel synthesis processing on the received millimeter-wave radar signal. This operation can be further achieved by using digital beamforming technology to coherently superimpose echoes from multiple antenna arrays, or by introducing lateral frequency modulation to expand the effective aperture to improve angular resolution, thereby enhancing the ability to distinguish small or dense obstacles and overcoming the insufficient angular resolution of traditional ultrasonic radar. Generating a two-dimensional obstacle thermal distribution map of the parking area can be achieved by mapping high-resolution millimeter-wave radar data onto a two-dimensional grid in the vehicle coordinate system and assigning values ​​according to reflection intensity or presence probability. Furthermore, this operation can be achieved by using density clustering algorithms (such as DBSCAN) to merge regions of the point cloud to generate a heatmap, or by smoothing the original point cloud with convolutional kernels to generate a continuous probability distribution map, thereby providing a structured spatial distribution representation of obstacles, which is convenient for subsequent time-domain analysis and risk quantification.

[0025] Step S200: Based on the real-time relative distance between the vehicle and surrounding obstacles, classify the emergency level of the corresponding obstacle avoidance behavior, and dynamically extract the time domain analysis window corresponding to the obstacle heat distribution map according to the emergency level.

[0026] The real-time relative distance between the vehicle and surrounding obstacles can be the Euclidean distance or projected distance between the vehicle and each obstacle at the current moment, which can be used as a direct basis for classifying the urgency level of obstacle avoidance behavior. The urgency level of obstacle avoidance behavior can be a dynamic classification of obstacle avoidance response priority based on relative distance, which can be used to determine the length of the time domain analysis window to achieve tiered processing of urgency levels. The time domain analysis window can be the length of the time segment used to analyze the dynamic behavior of obstacles, which can be adaptively adjusted according to the urgency level. High urgency corresponds to a short window for rapid response, and low urgency corresponds to a long window for stable judgment.

[0027] Based on the real-time relative distance between the vehicle and surrounding obstacles, the urgency level of the obstacle avoidance behavior is classified. This can be achieved by comparing the relative distance with a preset threshold range and assigning it to high, medium, or low urgency levels. Furthermore, this operation can be implemented by using a piecewise linear function to map the distance to continuous urgency values, or by dynamically adjusting the level classification boundaries in conjunction with obstacle movement trends. This allows for differentiated responses to obstacle avoidance strategies, avoiding response lag or overreaction caused by uniform thresholds. A time-domain analysis window corresponding to the obstacle heat map is dynamically extracted based on the urgency level. This can be achieved by selecting historical data windows of different lengths for dynamic behavior analysis based on the urgency level. Further, this operation can be implemented by using a 100ms window for high urgency levels to capture instantaneous changes and a 500ms window for low urgency levels to filter noise, or by making the window length inversely proportional to the obstacle speed, with shorter windows for faster speeds. This ensures both response speed and judgment stability, achieving urgency-level classification processing.

[0028] Step S300: Establish a mapping relationship library between obstacle material type and historical parking obstacle avoidance failure records, and associate it with the speed peak of surrounding moving obstacles to obtain the associated mapping relationship library.

[0029] The obstacle material type can be the category of the physical composition of the obstacle surface, affecting millimeter-wave reflection characteristics. It can be used as a key input dimension of the mapping relation library to correct perception errors and risk misjudgments. Historical parking obstacle avoidance failure records can be event logs of system failures due to perception or decision-making errors during past parking processes. These can be used to train the mapping relation library and improve robustness to error-prone scenarios. The peak speed of surrounding moving obstacles can be the maximum speed value reached by dynamic obstacles within the observation window. It can be used to reflect potential collision kinetic energy and serve as an important factor for risk quantification. The mapping relation library can be a data structure storing the association rules between obstacle material type, historical parking obstacle avoidance failure records, and peak speed of surrounding moving obstacles. It can be used to support the accurate quantification of the collision risk index, avoiding overreaction to low-risk static obstacles while ensuring rapid response to high-speed approaching targets. In a specific embodiment, the mapping relation library can be constructed through a combination of offline training and online learning, establishing statistical or logical mappings between multi-dimensional parameters based on a large amount of parking scenario data. Furthermore, the mapping relation library can provide multi-dimensional input for the rule decision module; its content participates in signal baseline offset compensation and collision risk index calculation. For example, the mapping relationship library may include, but is not limited to, one or more of the following: material-reflectivity mapping table, failure mode-scene context index, speed-risk weight lookup table, etc.

[0030] Establishing a mapping database between obstacle material types and historical parking obstacle avoidance failure records can be achieved by collecting millimeter-wave reflection characteristics of obstacles of different materials in multiple scenarios and corresponding obstacle avoidance failure cases, thus constructing a related database. Furthermore, this operation can be achieved by establishing a material-reflectivity-failure mode triplet through offline clustering analysis, or by continuously updating the mapping relationship using online reinforcement learning, thereby improving the system's robustness in recognizing obstacles of specific materials (such as low rubber curbs). Associating the peak speeds of surrounding moving obstacles to obtain a mapped database can be done by adding the peak speeds as a new dimension to the existing mapping database, forming a three-dimensional associated structure. Further, this operation can be achieved by constructing a three-dimensional lookup table of speed-material-failure probability, or by using a regression model to predict the failure risk of obstacles of specific materials at different speeds, thereby enhancing the ability to model the dynamic risks of high-speed moving obstacles.

[0031] Step S400: Based on the mapping relationship library, the signal baseline offset of the obstacle thermal distribution map is compensated by integrating the electromagnetic wave attenuation in rainy weather. The compensated obstacle thermal distribution map and time domain analysis window are processed by the rule decision module, and the collision risk index of the obstacle is output.

[0032] Among these, electromagnetic wave attenuation due to rain can be attributed to signal energy loss during millimeter wave propagation caused by rainfall, which can lead to a decrease in the strength of the sensed signal and requires correction through a compensation mechanism. Signal baseline offset can be caused by a systematic shift in the overall amplitude or phase of the millimeter-wave radar echo signal due to environmental interference, which, if not compensated, could lead to misjudgments of obstacle location or existence. The rule-based decision module is a software functional unit that fuses and judges multi-source sensed information based on preset logical rules and outputs a structured risk index. It can be used to comprehensively process the compensated obstacle thermal distribution map and time-domain analysis window to output the obstacle collision risk index. In an exemplary embodiment, the rule-based decision module can use an expert system, fuzzy logic, or a lightweight neural network to map multi-dimensional inputs to the collision risk index. For example, the rule-based decision module can include, but is not limited to, one or more of the following: a fuzzy inference engine, a threshold-based risk assessor, and a multi-condition logic judge. The collision risk index can be a quantitative indicator representing the probability and severity of obstacle collisions, calculated by comprehensively considering multiple factors. It can be used to drive the reconstruction of the damping response function to achieve differentiated steering control.

[0033] Compensating for the signal baseline offset of the obstacle thermal distribution map by integrating electromagnetic wave attenuation during rain can be achieved by retrieving the corresponding baseline offset from a mapping database based on current meteorological conditions and applying additive or multiplicative correction to the thermal map values. Furthermore, this operation can be implemented by selecting a pre-stored attenuation compensation curve based on rain sensor input, or by using the multi-band attenuation differences of millimeter-wave signals to invert rainfall intensity and dynamically calculate the compensation amount, thereby ensuring the accuracy of sensing data under severe weather conditions and preventing missed detections due to signal attenuation. The rule-based decision module processes the compensated obstacle thermal distribution map and the time-domain analysis window, which can be achieved by inputting the compensated thermal map and the obstacle trajectory within the dynamic window into the rule engine for logical judgment. Further, this operation can be achieved by using a fuzzy rule set to jointly evaluate obstacle approach rate and distance, or by using a weighted summation method to fuse multi-dimensional features to generate a risk score, thereby enabling structured risk assessment under multi-source information fusion. The collision risk index of the obstacle is output, which can be calculated and output by the rule-based decision module as a standardized risk value or level label, thus providing quantitative input for the reconstruction of the damped response function.

[0034] Step S500: Based on the collision risk index and peak speed, reconstruct the damping response function of the parking steering mechanism and execute the collision avoidance path correction action during the automatic parking process.

[0035] The parking steering mechanism can be an electromechanical integrated actuator that executes steering wheel rotation commands during automatic parking. It can be used to perform path correction actions based on the reconstructed damping response function to achieve vehicle steering control. For example, the parking steering mechanism can include, but is not limited to, one or more of an electric power steering system (EPS), a steer-by-wire actuator, and a hydraulically assisted steering unit. The damping response function can be a dynamic functional relationship describing the steering resistance required by the parking steering mechanism under different risk levels. It can be used to intelligently adjust the steering resistance according to the risk level, suppressing aggressive steering in high-risk situations and ensuring smoothness in low-risk situations. In one specific embodiment, the damping response function can be reconstructed in real time driven by both the collision risk index and the peak speed, exhibiting a nonlinear mapping between steering torque and steering angular velocity. Further, the damping response function can be reconstructed driven by the collision risk index and the peak speed; its output acts on the parking steering mechanism; and its execution deviation is corrected secondaryly through feedback of the steering motor vibration characteristics. For example, the damping response function can include, but is not limited to, one or more of a high-risk damping curve, a low-risk damping curve, and a transitional damping interpolation function.

[0036] Based on the collision risk index and peak speed, the damping response function of the parking steering mechanism is reconstructed. This can be achieved by using the risk index and peak speed as parameters to adjust the functional relationship between steering damping torque and steering angular velocity in real time. Furthermore, this operation can be performed using a lookup table method: interpolating the current function from multiple pre-stored damping curves, or by parametric modeling: using a nonlinear function that modulates the damping coefficient with the risk index. This allows for dynamic adaptation of steering resistance, balancing safety and comfort. The collision avoidance path correction action during automatic parking can be performed by the parking steering mechanism executing steering commands based on the reconstructed damping response function, adjusting the vehicle trajectory to complete the actual obstacle avoidance operation and prevent contact with obstacles.

[0037] In step S600, the vibration characteristics of the steering motor are extracted in real time during the parking and steering process, and the torque distribution of the wheels and steering mechanism is obtained by reverse calculation. The damping response function is corrected in a closed loop, and an intelligent control method for obstacle avoidance in the automatic parking process is realized through closed-loop control of all steps.

[0038] The vibration characteristics of the steering motor can be the frequency and time domain characteristics of the mechanical vibration signal generated by the steering motor during operation due to external disturbances or internal nonlinearities. These characteristics can be used to infer the torque distribution between the wheels and the steering mechanism, and to detect deviations between actual execution and the expected model. In an exemplary embodiment, the steering motor vibration characteristics can be extracted in real time using an acceleration sensor mounted on the steering motor or current harmonic analysis. Furthermore, the steering motor vibration characteristics can be used to correct the damping response function in a closed loop, forming an execution feedback loop. Exemplarily, the steering motor vibration characteristics can include, but are not limited to, one or more of vibration spectrum characteristics, transient impact response, and harmonic distortion components. The torque distribution between the wheels and the steering mechanism can be the spatial and temporal distribution of the mechanical torque borne by each component in the steering system. This can be used to reflect the deviation between the actual execution state and the ideal model, and for closed-loop correction. Closed-loop control can be a complete control loop including sensing, decision-making, execution, feedback, and re-correction, and can be used to enhance the system's adaptability to unknown disturbances and mechanical nonlinear factors.

[0039] Real-time extraction of steering motor vibration characteristics during parking and steering can be achieved through motor current harmonic analysis or by acquiring high-frequency vibration signals and extracting features using built-in vibration sensors. Further, this operation can be performed by extracting characteristic frequency band energy through FFT transformation of the motor phase current, or by directly measuring housing vibration using a MEMS accelerometer and performing time-frequency analysis, thereby obtaining physical state feedback of the execution process for deviation detection. The torque distribution of the wheels and steering mechanism can be obtained by inversely estimating the actual torque at each joint based on vibration characteristics and known mechanical transfer functions. Further, this operation can be achieved by using a pre-calibrated vibration-torque mapping model for table lookup, or by constructing a dynamic model and estimating the torque state using Kalman filtering, thus revealing the source of deviation between actual execution and the expected model. Closed-loop correction of the damping response function can be achieved by comparing the inversely obtained torque distribution with the expected value and adjusting the damping response function parameters to compensate for the deviation. Furthermore, this operation can be achieved by fine-tuning the damping coefficient using a proportional-integral controller, or by triggering local reprogramming and recalculating the damping function, thereby improving the system's adaptability to disturbances such as uneven road surfaces and changes in tire friction.

[0040] Taking automatic parking in an underground parking lot during rainy weather as an example, the vehicle control method in this embodiment can be as follows: when a vehicle enters the underground parking lot on a wet and slippery surface, the millimeter-wave radar signal is attenuated due to the rain. The radio frequency enhancement component improves the lateral resolution and clearly identifies low concrete pillars; the system performs baseline offset compensation on the heat map based on the historical failure records of the material in the mapping relationship library and the current rainy weather conditions; at the same time, an electric vehicle drives past quickly from the side, and the system classifies it into a high emergency level based on its real-time distance, shortens the time domain analysis window, and matches a high-risk mode in the mapping relationship library based on its speed peak, and the rule decision module outputs a high collision risk index; the damping response function is then reconstructed into a high-damping mode, and the steering action becomes more cautious; during the steering process, the road surface joint causes a sudden change in the vibration of the steering motor, the system extracts the back thrust torque distribution of this feature, finds that the actual steering resistance is lower than expected, and immediately corrects the damping function a second time, increasing the compensation torque to ensure that the path correction is accurately completed.

[0041] In one embodiment, the damping response function of the parking steering mechanism is reconstructed based on the collision risk index and peak speed, and a collision avoidance path correction action is performed during the automatic parking process, including: The collision risk index is obtained to determine the risk level. When the risk level is greater than the first preset threshold, the parking speed is reduced, and when the risk level is less than the second preset threshold, the normal parking speed is maintained.

[0042] The risk level can be a discrete risk category based on the collision risk index, used to trigger different parking control strategies. In this embodiment, the risk level can be mapped to a finite number of levels, such as high, medium, and low, after comparing continuous collision risk indices with preset threshold intervals, serving as the decision basis for adjusting parking speed and steering response characteristics. For example, the risk level can include, but is not limited to, one or more of high-risk, medium-risk, and low-risk levels. The first preset threshold can be the upper limit boundary value of the collision risk index used to determine whether a high-risk state has been entered. When the risk level exceeds this threshold, speed reduction control is triggered to reserve a safety margin. The second preset threshold can be the lower limit boundary value of the collision risk index used to determine whether a low-risk state has been entered. When the risk level is below this threshold, the normal parking speed is maintained to ensure efficiency. The parking speed can be the linear velocity of the vehicle moving forward or backward along the planned trajectory during automatic parking, as a controlled variable, dynamically adjusted according to the risk level to balance safety and efficiency. The normal parking speed can be the standard parking speed adopted by the system when there is no significant obstacle threat, representing the parking efficiency benchmark under normal operating conditions.

[0043] Obtaining the collision risk index and determining the risk level can be achieved by reading the collision risk index value output by the rule decision module and comparing it with a preset threshold range to assign a risk level. Furthermore, this operation can be implemented by mapping the risk index to specific level labels, thus providing a tiered decision-making basis for subsequent speed and steering control. When the risk level is greater than a first preset threshold, reducing the parking speed can be achieved by sending a deceleration command to the vehicle's longitudinal controller, lowering the current parking speed below a safe limit. In an exemplary embodiment, this operation can be achieved by a step-wise speed reduction according to the risk level (e.g., reducing to 0.5 m / s for high risk, and 1.0 m / s for medium risk) or by using a continuous function to map the risk index to a specific target speed value, thereby extending the reaction time window and improving obstacle avoidance success rate in high-risk scenarios. When the risk level is less than a second preset threshold, maintaining the normal parking speed can be achieved by keeping the current longitudinal control command unchanged and not triggering additional deceleration logic, thus avoiding reduced parking efficiency due to excessive conservatism in low-risk scenarios.

[0044] The peak speed is obtained, and the response delay of the steering mechanism is adjusted according to the preset adjustment rules based on the magnitude of the peak speed.

[0045] The response delay can be the time lag between issuing a steering command and the actual start of the steering mechanism's action, affecting the timeliness of obstacle avoidance actions and requiring dynamic adjustment based on the obstacle's dynamic characteristics. In a specific embodiment, the response delay may include, but is not limited to, one or more of mechanical transmission delay, control signal processing delay, and actuator start-up delay. The preset adjustment rule can be a logical or functional relationship that maps the peak speed amplitude to the response delay adjustment amount. A correspondence table or formula between the peak speed and the optimal response delay is established through offline calibration or simulation to achieve rapid execution response to high-speed approaching obstacles. For example, the preset adjustment rule can adopt speed-delay lookup table rules, linear proportional adjustment rules, piecewise nonlinear mapping rules, etc. Obtaining the peak speed and adjusting the steering mechanism's response delay based on the amplitude of the peak speed according to the preset adjustment rule can involve reading the peak speed of surrounding moving obstacles, querying or calculating the corresponding response delay compensation amount, and applying it to the steering control timing. Furthermore, this operation can compensate by triggering the steering command earlier with higher speed and shorter response delay, or by dynamically adjusting the control cycle to use a higher frequency of steering update rate at high speeds, thereby improving the agility of the execution response to high-speed moving obstacles and reducing obstacle avoidance lag.

[0046] The collision risk index and peak speed are input into a predefined function to generate a new damping coefficient.

[0047] The predefined function can be a mathematical mapping relationship between the collision risk index and the peak velocity as input and output damping coefficients. It can be an analytical function, a neural network model, or an interpolation surface, determined through training or design, and used to generate real-time damping coefficients that match the current risk scenario. In an exemplary embodiment, the predefined function may include, but is not limited to, a bivariate weighted summation function, a radial basis function interpolator, or a multilayer perceptron regression model. The damping coefficient can be a proportional parameter describing the magnitude of the motion resistance of the steering mechanism, directly affecting the steering feel and response characteristics. As a core parameter of the damping response function, it determines the dynamic stiffness of the steering system. For example, the damping coefficient may include, but is not limited to, one or more of the following: static damping coefficient, dynamic damping gain, and nonlinear damping factor. Inputting the collision risk index and peak velocity into the predefined function to generate new damping coefficients can be achieved by calling a predefined function interface, passing in two parameters, and returning the calculated damping coefficient. Furthermore, this operation can be achieved by interpolating the damping coefficient in a two-dimensional grid using a lookup table method, or by generating the damping coefficient through online inference using a lightweight neural network, thereby enabling multi-dimensional dynamic adaptation of damping parameters, balancing safety and comfort.

[0048] The motion resistance of the steering mechanism is updated using a new damping coefficient, and the damping response function of the steering mechanism is reconstructed.

[0049] The motion resistance of the steering mechanism can be considered as the opposing torque experienced by the steering system during rotation. It is determined by both the damping coefficient and angular velocity, directly affecting the load on the steering motor and influencing the smoothness and stability of path correction. Updating the motion resistance of the steering mechanism with a new damping coefficient and reconstructing its damping response function can be achieved by writing the new damping coefficient into the damping model parameters of the steering control algorithm, replacing the original value, so that the steering system possesses resistance characteristics that match the current risk level in real time.

[0050] Based on the reconstructed damping response function, the steering motor of the automatic parking system performs a collision avoidance path correction action.

[0051] Based on the reconstructed damping response function, the steering motor driving the automatic parking system performs collision avoidance path correction. This can be achieved by the steering control module calculating the required torque according to the updated damping response function and outputting PWM or current commands to drive the steering motor. Furthermore, this operation can achieve precise steering output through closed-loop torque control, thus completing a closed-loop linkage from risk assessment to physical execution, ensuring accurate and smooth path correction.

[0052] Taking the example of a car encountering a fast-moving electric vehicle while parking in a narrow alley in a residential area, the vehicle control method in this embodiment can be as follows: When the vehicle performs automatic parking in a narrow alley, millimeter-wave radar detects an electric vehicle approaching from the side at a relatively high speed. The system calculates a high collision risk index, and if the risk level exceeds a first preset threshold, it immediately reduces the parking speed. At the same time, the peak speed of the electric vehicle is obtained as 3.5 m / s, and the steering response delay is shortened from 80 ms to 40 ms according to a preset adjustment rule. The collision risk index and the peak speed are input into a predefined function to generate a high damping coefficient, which updates the motion resistance of the steering mechanism. The steering motor performs a small path correction in high damping mode, resulting in smooth and abrupt movements. The entire process avoids collisions with the electric vehicle and prevents passenger discomfort caused by sudden braking and sharp turns.

[0053] In one embodiment, the vibration characteristics of the steering motor are extracted in real time during the parking and steering process, and the torque distribution between the wheels and the steering mechanism is derived by reverse calculation. The closed-loop modified damping response function includes: During the parking and steering process, the vibration characteristics of the motor, including the amplitude and frequency of the vibration, are collected in real time by the torque sensor. The vibration characteristics are input into the pre-stored correspondence model and the real-time torque distribution value of the wheel steering is output. The torque sensor can be a physical sensing device used to directly measure the mechanical torque on the output shaft of the steering motor or related transmission components. It can be used to collect amplitude and frequency information from the vibration characteristics of the steering motor in real time, serving as a high-fidelity input source for inferring the torque distribution. In this embodiment, the torque sensor can convert mechanical deformation or magnetic field changes into electrical signal output based on strain gauges, magnetoelastic effects, or the Hall effect. Furthermore, the torque sensor's output can be used as input to a correspondence model to support the generation of real-time torque distribution values ​​for wheel steering. Exemplarily, the torque sensor can include, but is not limited to, one or more of strain gauge torque sensors, magnetostrictive torque sensors, and non-contact optical torque sensors. The correspondence model can be a pre-established mathematical or data-driven model that maps the vibration characteristics (amplitude, frequency) of the steering motor to real-time torque distribution values ​​for wheel steering. It can be used to achieve indirect, high-precision estimation from measurable vibration signals to indirect, unobservable torque distributions. In an exemplary embodiment, the correspondence model can be obtained by simultaneously collecting vibration signals and real torque data under calibration conditions and training it using regression, neural networks, or system identification methods. Furthermore, the correspondence model can receive vibration characteristics collected by a torque sensor and output real-time torque distribution values ​​for wheel steering, providing a basis for torque anomaly identification. In a specific embodiment, the correspondence model can employ a linear transfer function model, a multilayer perceptron mapping model, a temporal convolution inversion model, etc.

[0054] The real-time torque distribution value of wheel steering can be a dynamic vector set describing the magnitude and direction of the torque borne by each key node in the steering system (such as the steering rack, tie rod, and wheel contact point) at the current moment. This can be used to identify areas of uneven torque or overload, supporting accurate correction of the damping coefficient. In this embodiment, the real-time torque distribution value of wheel steering can be calculated based on vibration characteristics using a corresponding relationship model, reflecting the actual execution load state. For example, the real-time torque distribution value of wheel steering may include, but is not limited to, one or more of the left front wheel steering torque component, right front wheel steering torque component, and the resultant torque at the center of the steering rack. Real-time acquisition of motor vibration characteristics, including the amplitude and frequency of vibration, is achieved through a torque sensor. This can be done by continuously reading the time-domain signal output by the torque sensor during parking and steering, and performing amplitude and spectrum analysis. Furthermore, this operation can replace or enhance traditional indirect estimation methods by obtaining high-fidelity, low-latency physical state representations of the execution end, thereby achieving the technical effect of improving the underlying state perception capability. The vibration features are input into a pre-stored correspondence model, which outputs the real-time torque distribution value of wheel steering. Alternatively, the extracted vibration amplitude and frequency can be used as input vectors to the correspondence model, which outputs a multi-dimensional torque distribution estimate. Furthermore, this operation can be achieved through real-time inference on the vehicle ECU using a lightweight neural network model, or through rapid estimation using lookup table interpolation combined with pre-stored vibration-torque calibration data. This enables a high-precision mapping from easily measurable vibration signals to difficult-to-measurable torque distributions, giving the system a stronger state reconstruction capability.

[0055] Based on the torque distribution value, the region of uneven torque or overload is identified, and the damping coefficient of the damping response function is corrected according to the preset adjustment rules. Regions of uneven torque or overload can be localized locations in the torque distribution where there is significant asymmetry or the torque exceeds a safety threshold. These regions can be used as criteria to trigger corrections to the damping response function, indicating anomalies such as road disturbances, tire slippage, or mechanical failures. In one specific embodiment, regions of uneven torque or overload may include, but are not limited to, one or more of the following: a single-wheel overload zone, a steering mechanism center torque imbalance zone, and a high-frequency alternating torque fluctuation zone. The damping coefficient can be a proportional parameter in the damping response function used to adjust the steering resistance intensity. Its value directly determines the smoothness and suppression capability of steering actions, making it a core adjustable variable of the damping response function.

[0056] The preset adjustment rules can be a set of logical or algorithmic rules that specify how to modify the damping coefficient based on the type and degree of torque distribution anomalies. This can be used to ensure the physical rationality and control stability of the damping coefficient correction. In this embodiment, the preset adjustment rules can be formulated based on vehicle dynamics simulation and real-vehicle testing experience, and can be a lookup table method, piecewise function, or fuzzy control rule. Furthermore, the preset adjustment rules can act on the damping response function, guiding the update direction and magnitude of its damping coefficient. For example, the preset adjustment rules may include, but are not limited to, overload-damping gain mapping tables, torque imbalance compensation rules, and high-frequency vibration suppression strategies.

[0057] Based on torque distribution values, regions of uneven torque or overload can be identified by comparing each torque component with a preset balance threshold or historical baseline to determine if there are significant deviations or exceedances. Furthermore, this operation can be achieved by calculating the torque difference between the left and right wheels and determining if it exceeds the allowable range, or by detecting the variance or kurtosis of the torque signal to identify high-frequency impact overloads. This allows for precise location of the execution anomaly source, making subsequent damping adjustments more targeted. The damping coefficient of the damping response function is corrected according to preset adjustment rules. This can be done by querying or calculating the corresponding damping coefficient adjustment amount based on the identified anomaly type and degree, and updating the function parameters. Further, this operation can be achieved by locally increasing the steering damping coefficient on the detected side if a unilateral overload is detected, or by increasing the overall damping to suppress resonance if high-frequency torque oscillations are detected. This enables adaptive damping adjustment based on the scenario, avoiding the control performance degradation caused by blind increases or decreases.

[0058] The modified damping response function is updated and applied in real time to control the parking and steering motion in a closed loop.

[0059] The modified damping response function can be a new damping response function adjusted through closed-loop feedback, including updated damping coefficients. It can be used for real-time control of parking steering motion, improving adaptability to disturbances. Parking steering motion can be the process of the vehicle's front wheels changing steering angle according to the planned trajectory during automatic parking. It can be used as the final execution object of closed-loop control, and its smoothness and accuracy are directly affected by the modified damping response function. Real-time updating and application of the modified damping response function for closed-loop control of parking steering motion can be achieved by immediately loading the modified damping response function into the steering control module for command generation in the next control cycle. Furthermore, this operation can form a high-speed execution feedback closed loop, effectively suppressing steering jitter or response lag caused by road disturbances or mechanical nonlinearity.

[0060] Taking automatic parking on a wet and bumpy road as an example, the vehicle control method in this embodiment can be as follows: When a vehicle performs automatic parking on a potholed road after rain, one wheel drives over a puddle, causing a momentary decrease in adhesion and triggering a sudden change in steering torque. The torque sensor captures the sudden increase in motor vibration amplitude and frequency shift in real time, and the corresponding relationship model outputs that the torque of the left front wheel is significantly lower than that of the right front wheel. The system identifies the torque imbalance area and increases the damping coefficient of the left side according to a preset adjustment rule to compensate for the loss of grip. The corrected damping response function takes effect immediately, making the steering action more stable and avoiding path deviation caused by slippage. The entire process is completed within hundreds of milliseconds, and the passengers do not perceive any obvious vibration, and the parking trajectory maintains high accuracy.

[0061] In one embodiment, the spatial resolution of millimeter-wave radar data in the lateral direction of the parking path is improved by a radio frequency enhancement component, generating a two-dimensional obstacle thermal distribution map of the parking area, including: The focusing antenna and filtering module with radio frequency enhancement components are used to process the acquired millimeter-wave radar data, which includes the outline and morphological features of obstacles and information on surrounding moving obstacles. By adjusting the radio frequency transmission and reception path, the difference in echo signal strength between lateral adjacent points of the parking path is increased, thereby improving the spatial resolution of the processed millimeter-wave radar data. The millimeter-wave radar data with improved spatial resolution is converted into a two-dimensional grid representation, and the echo intensity value of each location is displayed along the lateral direction of the parking path to generate a two-dimensional obstacle thermal distribution map of the parking area.

[0062] The focusing antenna can be a millimeter-wave antenna structure with beam-focusing capabilities, used to enhance the concentration of electromagnetic wave energy in a specific direction. In this embodiment, the focusing antenna can be designed using a phased array or lens antenna to focus energy on the target area during the RF transmission or reception phase. Furthermore, as a component of the RF enhancement component, the focusing antenna works in conjunction with the filtering module to process millimeter-wave radar data; its focusing effect directly affects the generation of echo signal intensity differences. Exemplarily, the focusing antenna can be one or more of, including but not limited to, phased focusing arrays, dielectric lens antennas, and reflector focusing antennas. The filtering module can be an RF or digital signal processing unit used to suppress out-of-band noise or interference components in the millimeter-wave radar signal. In an exemplary embodiment, the filtering module can employ a bandpass filter, matched filter, or adaptive noise suppression algorithm to purify the echo signal. Furthermore, the filtering module and the focusing antenna together constitute the core processing link of the RF enhancement component, outputting optimized millimeter-wave radar data for subsequent spatial resolution improvement. In a specific embodiment, the filtering module can include, but is not limited to, digital matched filters, adaptive notch filters, and multi-stage bandpass filtering units.

[0063] Using a focusing antenna and filtering module with RF enhancement components, the acquired millimeter-wave radar data, which includes obstacle contour features and information about surrounding moving obstacles, can be processed by sequentially passing the raw millimeter-wave radar data through energy concentration processing of the focusing antenna and noise suppression processing of the filtering module. Furthermore, this operation can be achieved by first enhancing the target direction echo through the focusing antenna and then removing clutter through a digital filter, or by integrating analog filtering and phased-array focusing at the RF front end to achieve integrated preprocessing. This can improve the echo signal-to-noise ratio and directional selectivity, laying a signal quality foundation for subsequent spatial resolution enhancement. The RF transmit / receive path can be the complete RF channel configuration experienced by the millimeter-wave radar system from signal transmission to echo reception, including phase, delay, and gain parameters. In a specific embodiment, the RF transmit / receive path can include, but is not limited to, a phase-adjustable transmit path, a delay-differentiated receive channel, and a gain gradient configuration path. By adjusting the RF transmit / receive path, the difference in echo signal strength between lateral adjacent points on the parking path can be increased. This can be achieved by dynamically configuring the transmit phase or receive channel delay, so that the echoes from adjacent lateral sampling points produce significant amplitude distinctions after synthesis. Furthermore, this operation can amplify edge response differences by adjusting the transmit-receive combination weight of the virtual aperture in the MIMO radar architecture, or by introducing a lateral non-uniform sampling strategy in conjunction with path parameter modulation, so as to break through the traditional angular resolution limitation without increasing the number of physical antennas, and make the outline of small obstacles distinguishable at the signal level.

[0064] The echo signal intensity difference can be the difference in millimeter-wave echo amplitude corresponding to adjacent spatial sampling points along the parking path. In an exemplary embodiment, the echo signal intensity difference may include, but is not limited to, edge gradient intensity difference, amplitude jumps at consecutive points, and local peak contrast. The echo signal intensity difference serves as a key indicator for improving spatial resolution; a larger difference is more beneficial for distinguishing the boundaries of nearby obstacles. Improving the spatial resolution of the processed millimeter-wave radar data can be achieved by reconstructing a higher-density distribution of lateral spatial sampling points based on the enhanced echo signal intensity difference, thereby enabling effective identification of obstacles that are easily missed by traditional systems, such as low curbs and thin pillars. The two-dimensional grid representation can be a data organization form that discretizes continuous space into a regular row and column structure, with each cell corresponding to a physical location. In a specific embodiment, the two-dimensional grid representation may include, but is not limited to, polar coordinate grids, Cartesian coordinate grids, and adaptive density grids. The two-dimensional grid representation provides a structured representation basis for millimeter-wave radar data, facilitating the generation of visual heat map distribution. Converting the millimeter-wave radar data with improved spatial resolution into a two-dimensional grid representation can be achieved by mapping high-resolution point clouds or echo sequences onto a two-dimensional planar grid centered on the vehicle. Furthermore, this operation can assign point clouds to grid cells through nearest neighbor interpolation, or generate a continuous probability field through kernel density estimation and then discretize it into a grid, thereby providing a structured and regular data format, which is convenient for subsequent heat map generation and risk analysis.

[0065] Echo intensity values ​​can be quantized values ​​of the reflected signal amplitude received by millimeter-wave radar at a specific spatial location. These values ​​serve as the basis for assigning values ​​to each grid cell in the heatmap, reflecting the probability of obstacle presence or surface reflection characteristics. Displaying the echo intensity value at each location along the lateral direction of the parking path can be achieved by filling the corresponding echo intensity values ​​in order of physical location along the lateral axis of a two-dimensional grid, thus visually presenting the distribution density and boundary characteristics of obstacles on both sides of the vehicle. Generating a two-dimensional obstacle heatmap of the parking area can be based on the echo intensity values ​​of each cell in the two-dimensional grid, using color or grayscale mapping to form a visualized heatmap. This allows for the output of high-fidelity, structured environmental perception results, supporting subsequent multi-dimensional risk modeling and dynamic obstacle avoidance decisions.

[0066] Taking the scenario of automatic parking in a narrow parking space with a low curb as an example, the vehicle control method in this embodiment can be that the vehicle drives into a perpendicular parking space with a width only slightly larger than the vehicle body, and there is a rubber curb with a height of less than 10cm on the right. Traditional ultrasonic radar cannot identify the curb due to blind spots, while in this solution, the millimeter-wave radar concentrates energy to illuminate the curb area through the focusing antenna of the radio frequency enhancement component, and the filtering module suppresses ground clutter; the system actively adjusts the radio frequency transmission and reception path, so that the lateral sampling points on both sides of the curb edge produce a significant difference in echo intensity, and the originally blurred boundary forms a clear gradient at the signal level; the high-resolution data is converted into a two-dimensional grid, and each lateral position is assigned a precise echo intensity value. The final generated heat map clearly shows that there is a continuous low-intensity obstacle zone on the right. Based on this, the system determines that it is an impassable area, and the planned path shifts to the left, successfully completing the safe parking.

[0067] In one embodiment, the time-domain analysis window corresponding to the obstacle thermal distribution map is dynamically extracted based on the emergency level, including: Based on the classification of emergency levels, obstacle course movement is divided into emergency movement level or normal movement level; When the emergency level is emergency movement level, a short time-domain analysis window is selected; when the emergency level is normal movement level, a long time-domain analysis window is selected. The time-domain analysis window covers the change data of the obstacle heat distribution map in the time domain, and the window length determines the time range for analyzing the obstacle heat distribution map.

[0068] The emergency movement level can be a dynamic behavior category in the obstacle avoidance behavior emergency level that represents high risk and requires rapid response. It can be used to trigger a short-duration time-domain analysis window, focusing on recent changes in the thermal distribution of obstacles and improving the response sensitivity to rapidly approaching targets. In an exemplary embodiment, the emergency movement level can be explained in context, i.e., by mapping the pre-defined emergency levels to a preset binary movement behavior classification system, risk-oriented classification of obstacle dynamic behavior is achieved. Further, the emergency movement level can be a trigger source in the system, used to initiate a short-duration time-domain analysis window. For example, the emergency movement level can include, but is not limited to, one or more of high-speed approach, trajectory change, and lateral intrusion types. The regular movement level can be a dynamic or static behavior category in the obstacle avoidance behavior emergency level that represents low risk and can be stably tracked. It can be used to trigger a long-duration time-domain analysis window, integrating historical thermal evolution trends to enhance the stability of judgment on slowly moving or stationary obstacles. In a specific embodiment, the regular movement level can be explained in context, i.e., it is classified into this category based on the continuous value or threshold of the emergency level. Furthermore, the regular motion level can be a trigger condition in the analysis to enable a long-duration time-domain analysis window. For example, the regular motion level can include, but is not limited to, one or more of the following: stationary dwell, uniform moving away, and low-speed parallel movement.

[0069] Short-duration time-domain analysis windows can be used for data analysis spanning shorter time periods in emergency movement scenarios. They can limit the analysis scope to recent data, reduce historical noise interference, and accelerate risk assessment. In this embodiment, the operating principle of a short-duration time-domain analysis window can be explained in context: after determining an emergency movement level, a data segment from the time series data of the obstacle thermal distribution map is selected. Furthermore, a short-duration time-domain analysis window can be an execution unit triggered by an emergency movement level, covering recent changes in the obstacle thermal distribution map in the time domain. Long-duration time-domain analysis windows can be used for data analysis spanning longer time periods in normal movement scenarios. They can incorporate richer historical thermal distribution evolution information, improving robustness in identifying slow changes or periodic behaviors. In an exemplary embodiment, the acquisition method of a long-duration time-domain analysis window can be explained in context: after determining a normal movement level, a thermal distribution map sequence containing a longer historical span is selected. Furthermore, a long-duration time-domain analysis window can be an execution unit within the system, triggered by regular motion levels, covering long-term changes in obstacle thermal distribution data in the time domain.

[0070] The temporal variation data of the obstacle heat map can be a dynamic evolution record of the two-dimensional obstacle heat map of the parking area in a continuous time series. It can be used to reflect the changing trends of obstacle location, shape, or probability of existence over time, providing a temporal basis for dynamic risk assessment. The window length can be a time span parameter covered by the time-domain analysis window, determining the time range used to analyze the obstacle heat map and directly affecting response speed and judgment stability. The time range for analyzing the obstacle heat map can be the time interval of historical data of the heat map selected in the current time-domain analysis window, defining the boundaries of historical perception data upon which risk assessment relies. Based on the classified urgency levels, obstacle movement can be divided into emergency movement levels or normal movement levels, mapping the classified urgency levels to a preset binary movement behavior classification system. Furthermore, this operation can be implemented by using threshold segmentation: if the emergency level is higher than a set value, it is classified as an emergency movement level; otherwise, it is classified as a normal movement level. Alternatively, fuzzy membership can be introduced: the probability of belonging to two categories is calculated based on the continuous values ​​of the emergency level, and the category with the largest membership is selected. This can achieve risk-oriented classification of the dynamic behavior of obstacles and provide a clear basis for the selection of the time-domain window length.

[0071] When the emergency level is "emergency movement," a short time-domain analysis window can be selected. This can be achieved by choosing a recent data segment from the time-series data of the obstacle heat map after the emergency movement level is determined. Furthermore, this operation can be implemented by directly selecting the latest data with a fixed short window length (e.g., 100ms); or by dynamically fine-tuning the short window length based on the obstacle approach rate—the higher the rate, the shorter the window length. This allows focusing on recent abrupt changes, avoiding the dilution of current high-risk signals by historical data, and improving response timeliness. When the emergency level is "normal movement," a long time-domain analysis window can be selected. This can be achieved by selecting a heat map sequence containing a longer historical span after the normal movement level is determined. Furthermore, this operation can be implemented by retrospectively tracing historical data with a fixed long window length (e.g., 500ms); or by adaptively extending the window length based on environmental complexity, appropriately shortening it when obstacles are dense to avoid confusion. This allows using temporal redundancy to suppress transient noise and enhance confidence in the steady state. The temporal variation data of the thermal distribution map of the covered obstacles can ensure that the captured temporal analysis window contains a continuous sequence of thermal distribution map frames, thereby providing a complete temporal input for dynamic behavior modeling and supporting trajectory prediction and risk evolution analysis.

[0072] For example, in a scenario where a pedestrian suddenly crosses the side of a shopping mall's underground parking garage, the vehicle control method in this embodiment could involve the millimeter-wave radar detecting a pedestrian rapidly approaching from the side and rear during the vehicle's automatic parking process. The system classifies the pedestrian's urgency level as high based on the real-time relative distance, further categorizing it as an emergency movement level. A short 100ms time-domain analysis window is then captured, analyzing only the recent heatmap frames showing a sudden increase in the intensity of the pedestrian's area. Based on this, the rule-making module quickly outputs a high collision risk index, triggering a high-damping steering response. Simultaneously, a slow-moving shopping cart ahead is determined to be at a normal movement level. The system allocates a 400ms long window to it, comprehensively analyzing multiple heatmap frames to confirm its slow rightward movement trend, avoiding misjudging a threat due to single-frame jitter, and maintaining smooth steering under low-risk conditions.

[0073] In one embodiment, a mapping database is established between obstacle material types and historical parking obstacle avoidance failure records, and the peak speeds of surrounding moving obstacles are correlated to obtain the correlated mapping database, including: Collect obstacle material type data and historical parking obstacle avoidance failure records, and associate and store the material type with the historical parking obstacle avoidance failure records to establish an initial mapping relationship library; The initial mapping relationship library can be a basic data structure containing only the association between obstacle material types and historical parking obstacle avoidance failure records. This structure can serve as the foundation for building a multi-dimensional mapping relationship library, supporting subsequent expansion with the speed dimension. In an exemplary embodiment, the initial mapping relationship library can collect offline event logs of obstacles of different materials causing obstacle avoidance failures during parking and establish an index mapping between material categories and failure records. Furthermore, the initial mapping relationship library can receive input of obstacle material types and historical parking obstacle avoidance failure records; serving as an intermediate data structure before adding the peak speed dimension. For example, the initial mapping relationship library can include, but is not limited to, one or more of the following: a material-failure frequency table, a scene context annotation library, and a reflection characteristic-false detection association set.

[0074] Collecting obstacle material type data and historical parking obstacle avoidance failure records can be achieved by extracting structured data containing obstacle material tags and corresponding obstacle avoidance failure events from vehicle cloud logs or local storage. Furthermore, collecting obstacle material type data and historical parking obstacle avoidance failure records can be accomplished by inverting material types from millimeter-wave radar echo features and matching failure events, or by combining obstacle material annotations with camera semantic segmentation results and then associating failure records. This allows for the construction of a data foundation for a mapping relationship library, ensuring the representativeness of the statistical correlation between materials and failures. Associating and storing material types with historical parking obstacle avoidance failure records to establish an initial mapping relationship library can be done by using material types as keys and aggregating corresponding failure records to form an index structure. In a specific embodiment, this operation can be achieved by using a hash table to quickly look up materials in the failure event list, or by using a graph database to establish a multi-hop association between materials, scenes, and failures, thus forming a preliminary material sensitivity model to identify obstacle types that are easily missed.

[0075] Add the peak velocity of surrounding moving obstacles as an additional dimension to the initial mapping relation library; Adding the peak speeds of surrounding moving obstacles as an additional dimension to the initial mapping database can be achieved by supplementing each failure record with the maximum relative speed values ​​of the surrounding moving obstacles at the time of the failure and reorganizing the data structure. For example, this operation can be implemented by adding a speed column to the existing two-dimensional table to form a triplet storage, or by constructing speed interval buckets, discretizing continuous speed values ​​and assigning them to the corresponding buckets for statistical analysis. This introduces a dynamic dimension, extending risk modeling from static material perception to dynamic-static coupled scenarios.

[0076] For different obstacle material types and peak speeds, the frequency of corresponding obstacle avoidance failures is associated and stored in the initial mapping database. The output is a mapping database containing associated material types, peak speeds, and historical parking obstacle avoidance failure records.

[0077] The obstacle avoidance failure frequency can be the statistical number of times the system fails to avoid obstacles under specific material type and peak speed combinations. This can be used to quantify the risk level under different material-speed combinations, providing empirical evidence for the collision risk index. In this embodiment, the obstacle avoidance failure frequency can be calculated by clustering historical parking logs, grouping failure events by material and speed range. Furthermore, the obstacle avoidance failure frequency can be stored as a key field in a mapped relation library to support risk assessment by the rule-based decision-making module. For example, the obstacle avoidance failure frequency can include, but is not limited to, one or more of the following: static obstacle failure frequency, low-speed dynamic obstacle failure frequency, and high-speed approach obstacle failure frequency. The mapped relation library, which associates material type, peak speed, and historical parking obstacle avoidance failure records, can be a three-dimensional relational database that integrates obstacle material type, peak speed of surrounding moving obstacles, and corresponding obstacle avoidance failure frequencies. This database can provide multi-dimensional prior knowledge to the rule-based decision-making module, enabling differentiated risk assessment of obstacles with different physical properties and dynamic states. In one specific embodiment, the mapping relationship library can be built upon an initial mapping relationship library by adding peak velocity as a new dimension and organizing and storing it according to triples (material, velocity, failure frequency). Furthermore, this mapping relationship library can replace the original mapping relationship library in signal baseline offset compensation and collision risk index calculation; its content directly affects the reconstruction logic of the damping response function. For example, the mapping relationship library may include, but is not limited to, one or more of the following: high reflectivity-high speed-low failure library, low reflectivity-low speed-high failure library, mixed material-variable speed-medium failure library, etc.

[0078] For different obstacle material types and peak speeds, the frequency of corresponding obstacle avoidance failures is associated and stored in an initial mapping database. This can be achieved by grouping and counting failure events according to material-speed combinations and updating the frequency field in the mapping database. Furthermore, this operation can be implemented by using a weighted average of failure frequencies within a sliding time window to reflect recent trends, or by introducing a confidence factor and performing Bayesian smoothing on the frequencies of low-sample-size combinations, thereby achieving a quantitative expression of risk and avoiding the coarse judgment caused by relying solely on binary failure labels. The output includes a mapping database containing associated material type, peak speed, and historical parking obstacle avoidance failure records. This can be achieved by encapsulating the constructed three-dimensional associated data structure into a callable interface or file format for use by the rule-based decision-making module. In an exemplary embodiment, this operation can be implemented by serializing the data in JSON or Protobuf format and loading it into an in-vehicle embedded system, or by providing direct access to the real-time decision engine through memory mapping, thus providing a structured and queryable prior knowledge base for subsequent multi-dimensional fusion decision-making.

[0079] Taking the automatic parking system in a shopping mall's underground parking garage encountering a soft obstacle and a delivery vehicle as an example, the vehicle control method in this embodiment can be as follows: During automatic parking, millimeter-wave radar detects a low-reflectivity obstacle ahead. The system identifies its material type as "foam plastic" and checks the initial mapping database, finding that this material has caused parking vehicles to run over obstacles multiple times due to its weak reflection. At this time, a low-speed moving obstacle passes by on the right at a speed of 8 km / h. The system extracts its peak speed and queries the associated mapping database, finding that the combination of "foam material + low speed" has an extremely high failure frequency, while the combination of "metal body + medium speed," although at a higher speed, has a lower failure frequency. Based on this, the rule decision module determines that the soft obstacle is a high-risk target, increases the collision risk index, and triggers high-damping steering; at the same time, it maintains a moderate response to the delivery vehicle to avoid sudden braking. This differentiated strategy effectively prevents missed detection of soft static obstacles and avoids overreaction to slow-moving vehicles.

[0080] In one embodiment, a rule-based decision module processes the compensated obstacle thermal distribution map and time-domain analysis window to output the obstacle collision risk index, including: The rule-based decision-making module extracts the echo intensity change value in the lateral direction of the parking path from the compensated obstacle thermal distribution map; The echo intensity variation value can be the difference or gradient of millimeter-wave echo intensity between adjacent spatial locations in the obstacle thermal distribution map along the lateral direction of the parking path. It can be used to reflect the degree of abrupt change in the obstacle's outline or material boundaries, and to distinguish between static obstacle types such as continuous walls and isolated pillars. In this embodiment, the echo intensity variation value can be obtained by performing a first-order or higher-order difference operation on the compensated obstacle thermal distribution map along the lateral direction. For example, the echo intensity variation value can be obtained by using the Sobel or Prewitt operator for lateral edge detection, or by calculating the sliding window standard deviation of the thermal map's lateral profile to characterize the degree of local variation. This operation can be performed by a rule-based decision module to extract the echo intensity variation value in the lateral direction of the parking path from the compensated obstacle thermal distribution map, which involves calculating the intensity difference between adjacent pixels or grid cells along an axis perpendicular to the vehicle's direction of travel in the compensated thermal distribution map. Furthermore, this operation can be achieved by using the Sobel or Prewitt operator to perform lateral edge detection to obtain the intensity gradient, or by calculating the sliding window standard deviation of the heatmap lateral profile to characterize the degree of local change. This can quantify the spatial discontinuity of obstacles in the lateral direction and help identify the boundaries of small or low obstacles.

[0081] Extract the characteristics of the rate of change of echo intensity within the corresponding time range from the time domain analysis window; The echo intensity change rate characteristic within a time range can be the rate or trend characteristic of echo intensity change over time at a specific spatial location within a time-domain analysis window. This can be used to characterize the dynamic behavior of obstacles approaching or moving away, and to identify high-speed moving targets and slowly drifting interference sources. In an exemplary embodiment, the echo intensity change rate characteristic within a time range is obtained by extracting dynamic characteristics through time derivative estimation, slope fitting, or frequency domain analysis of the thermodynamic value sequence of the same spatial unit within the time-domain window. This operation can be to extract the echo intensity change rate characteristic within the corresponding time range from the time-domain analysis window, which involves performing time-dimensional analysis on the thermodynamic value sequence of the same spatial location within the time-domain window to extract its rate of change or acceleration characteristics. Furthermore, this operation can be achieved by performing linear regression on the time series using the slope as an indicator of the rate of change, or by calculating the mean absolute value of the difference between adjacent frames' thermodynamic values ​​as an indicator of dynamic activity, thereby capturing the obstacle's movement trend and distinguishing between stationary objects and approaching threats.

[0082] The system analyzes the changes in echo intensity and the rate of change using preset rules. Based on the results of the analysis, it adjusts the collision risk index by increasing or decreasing it, and finally outputs the adjusted collision risk index, which represents the probability of a collision.

[0083] The preset rules can be a set of logical judgment conditions defined based on the combination of echo intensity change value and the rate of change characteristics. These rules can guide the rule decision module to adjust the initial collision risk index, achieving risk calibration driven by physical semantics. In one specific embodiment, the preset rules are obtained through a multi-dimensional feature-risk adjustment mapping relationship established by expert experience induction or data-driven methods. Furthermore, the preset rules can include, but are not limited to, one or more of the following: static high-intensity low-change rules, dynamic fast-increasing high-change rules, and weak signal slow-change ignoring rules. This operation can involve applying the preset rules to analyze the echo intensity change value and the rate of change characteristics. This involves inputting the two extracted features into the preset rule set and matching the corresponding risk adjustment strategy. Further, this operation can be achieved by using an if-else rule tree: if the change value is large and fast, the risk index is significantly increased; if the change value is large but slow, the risk index is slightly decreased. Alternatively, a fuzzy rule system can be used to fuzzify the two features and then synthesize them through reasoning. This allows for multi-dimensional fusion judgment based on physical echo characteristics, avoiding misjudgments caused by a single distance threshold.

[0084] This operation can also adjust the collision risk index by increasing or decreasing it based on the analysis results. It applies a positive or negative offset to the initial collision risk index according to the rule matching results. Furthermore, this operation can be achieved by setting a fixed increment (e.g., +0.3 or -0.2) according to the rule level, or by continuously adjusting the adjustment range according to the feature intensity (e.g., the larger the rate of change, the higher the increment). This allows the collision risk index to better reflect the actual physical scenario, improving the rationality of subsequent control decisions. Alternatively, this operation can output an adjusted collision risk index representing the probability of a collision. This is achieved by standardizing the adjusted value and outputting it to the damping response function reconstruction module. The probability of a collision can be a probabilistic description of the actual physical contact between the obstacle and the vehicle, quantified by the collision risk index, and can serve as a direct basis for damping response function reconstruction and path correction. This operation enables the output of the adjusted collision risk index representing the probability of a collision to provide precise, dynamic, and semantically rich risk input for steering control.

[0085] For example, in a scenario where a pedestrian crosses the road during temporary parking, the vehicle control method in this embodiment could be as follows: The vehicle is automatically parking on a narrow street, with a continuous brick wall (high intensity, low lateral change) on the right, and a pedestrian suddenly crossing rapidly on the left. The rule decision module extracts the small change in echo intensity on the right side from the compensated heatmap, identifying it as a continuous stationary obstacle; simultaneously, it observes from the time-domain window that the echo intensity in a certain area on the left increases rapidly within 200ms, showing a significant difference in the rate of change. The preset rule matching "high change, rapid increase" mode significantly increases the collision risk index for that area; while the risk of the right wall being triggered by the "high intensity, low change" rule is reduced. Finally, a differentiated risk index is output, and the system only applies high-damping cautious steering in the direction of the pedestrian to avoid excessive avoidance of the wall, which could lead to parking failure.

[0086] In addition, refer to Figure 2 To achieve the above objectives, the present invention also provides an automotive control system, the system comprising: The perception enhancement module 10 is used to collect millimeter-wave radar data containing the outline and shape features of obstacles and the movement information of surrounding vehicles during the automatic parking process. The spatial resolution of the millimeter-wave radar data in the lateral direction of the parking path is improved by the radio frequency enhancement component, and a two-dimensional obstacle thermal distribution map of the parking area is generated. The dynamic window module 20 is used to classify the emergency level of the obstacle avoidance behavior based on the real-time relative distance between the vehicle and the surrounding obstacles, and dynamically extract the time domain analysis window corresponding to the obstacle heat distribution map according to the emergency level. The mapping and association module 30 is used to establish a mapping relationship library between obstacle material type and historical parking obstacle avoidance failure records, and to associate the speed peak of surrounding moving obstacles to obtain the associated mapping relationship library; The compensation decision module 40 is used to compensate for the signal baseline shift of the obstacle thermal distribution map by integrating the attenuation of electromagnetic waves in rainy weather based on the mapping relationship library, and to process the compensated obstacle thermal distribution map and the time domain analysis window using the rule decision module to output the collision risk index of the obstacle. The damping reconstruction module 50 is used to reconstruct the damping response function of the parking steering mechanism based on the collision risk index and the peak speed, and to perform the anti-collision path correction action in the automatic parking process. The closed-loop correction module 60 is used to extract the vibration characteristics of the steering motor in real time during the parking and steering process, back-derive the torque distribution of the wheels and steering mechanism, and correct the damping response function in a closed loop. Through closed-loop control of all steps, the intelligent control method for obstacle avoidance in the automatic parking process is realized.

[0087] Other embodiments or specific implementations of the vehicle control system described in this invention can be found in the above-described method embodiments, and will not be repeated here.

[0088] In addition, to achieve the above objectives, the present invention also provides an automobile control device, the device comprising: a memory, a processor, and an automobile control program stored in the memory and executable on the processor, the automobile control program being configured to implement the steps of the automobile control method as described in any one of the above descriptions.

[0089] In addition, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a vehicle control program, which, when executed by a processor, implements the steps of the vehicle control method as described in any one of the above descriptions.

[0090] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A vehicle control method, characterized in that, The method includes: The system collects millimeter-wave radar data containing the outline and shape features of obstacles and the movement information of surrounding vehicles during the automatic parking process. The spatial resolution of the millimeter-wave radar data in the lateral direction of the parking path is improved by the radio frequency enhancement component, and a two-dimensional obstacle thermal distribution map of the parking area is generated. Based on the real-time relative distance between the vehicle and surrounding obstacles, the emergency level of the corresponding obstacle avoidance behavior is divided, and a time-domain analysis window corresponding to the obstacle heat map is dynamically extracted according to the emergency level. Establish a mapping relationship library between obstacle material type and historical parking obstacle avoidance failure records, and associate it with the peak speed of surrounding moving obstacles to obtain the associated mapping relationship library; Based on the mapping relationship library, the signal baseline offset of the obstacle thermal distribution map is compensated by integrating the attenuation of electromagnetic waves in rainy weather. The compensated obstacle thermal distribution map and the time domain analysis window are processed by the rule decision module, and the collision risk index of the obstacle is output. Based on the collision risk index and the peak speed, the damping response function of the parking steering mechanism is reconstructed, and the collision avoidance path correction action of the automatic parking process is executed. The vibration characteristics of the steering motor are extracted in real time during the parking and steering process. The torque distribution between the wheels and the steering mechanism is then derived. The damping response function is corrected in a closed loop. The intelligent control method for obstacle avoidance in the automatic parking process is realized through closed-loop control of all steps.

2. The vehicle control method as described in claim 1, characterized in that, The step of reconstructing the damping response function of the parking steering mechanism based on the collision risk index and the peak speed, and executing the collision avoidance path correction action in the automatic parking process, includes: The collision risk index is obtained to determine the risk level. When the risk level is greater than a first preset threshold, the parking speed is reduced. When the risk level is less than a second preset threshold, the normal parking speed is maintained. The peak speed is obtained, and the response delay of the steering mechanism is adjusted according to a preset adjustment rule based on the magnitude of the peak speed. The collision risk index and the peak speed are input into a predefined function to generate a new damping coefficient. The motion resistance of the steering mechanism is updated using the new damping coefficient, and the damping response function of the steering mechanism is reconstructed. Based on the reconstructed damping response function, the steering motor of the automatic parking system performs a collision avoidance path correction action.

3. The vehicle control method as described in claim 1, characterized in that, The process of extracting the vibration characteristics of the steering motor in real time during parking and steering, back-deriving the torque distribution between the wheels and the steering mechanism, and then correcting the damping response function in a closed loop includes: During the parking and steering process, the vibration characteristics of the motor, including the amplitude and frequency of the vibration, are collected in real time by a torque sensor. The vibration characteristics are then input into a pre-stored correspondence model, and the real-time torque distribution value of the wheel steering is output. Based on the torque distribution value, regions with uneven torque or overload are identified, and the damping coefficient of the damping response function is corrected according to the preset adjustment rules. The modified damping response function is updated and applied in real time to control the parking and steering motion in a closed loop.

4. The vehicle control method as described in claim 1, characterized in that, The step of enhancing the spatial resolution of the millimeter-wave radar data in the lateral direction of the parking path using radio frequency enhancement components to generate a two-dimensional obstacle thermal distribution map of the parking area includes: The focusing antenna and filtering module with radio frequency enhancement components are used to process the acquired millimeter-wave radar data, which includes the outline and morphological features of obstacles and information on surrounding moving obstacles. By adjusting the radio frequency transmission and reception path, the difference in echo signal strength between lateral adjacent points of the parking path is increased, thereby improving the spatial resolution of the processed millimeter-wave radar data. The millimeter-wave radar data with improved spatial resolution is converted into a two-dimensional grid representation, and the echo intensity value of each location is displayed along the lateral direction of the parking path to generate a two-dimensional obstacle thermal distribution map of the parking area.

5. The vehicle control method as described in claim 1, characterized in that, The step of dynamically extracting a time-domain analysis window corresponding to the obstacle thermal distribution map based on the emergency level includes: Based on the classification of emergency levels, obstacle course movement is divided into emergency movement level or normal movement level; When the emergency level is emergency movement level, a short time-domain analysis window is selected; when the emergency level is normal movement level, a long time-domain analysis window is selected. The time-domain analysis window covers the time-domain change data of the obstacle heat distribution map, and the window length determines the time range for analyzing the obstacle heat distribution map.

6. The vehicle control method as described in claim 1, characterized in that, The process involves establishing a mapping database between obstacle material types and historical parking obstacle avoidance failure records, and then associating this database with the peak speeds of surrounding moving obstacles to obtain the associated mapping database, which includes: Collect obstacle material type data and historical parking obstacle avoidance failure records, and associate the material type with the historical parking obstacle avoidance failure records to establish an initial mapping relationship library; Add the peak velocity of the surrounding moving obstacles as an additional dimension to the initial mapping relation library; For different obstacle material types and peak speeds, the frequency of corresponding obstacle avoidance failures is associated with and stored in the initial mapping relationship library, and the associated mapping relationship library containing the material type, the peak speed, and the historical parking obstacle avoidance failure records is output.

7. The vehicle control method as described in claim 1, characterized in that, The rule-based decision-making module processes the compensated obstacle thermal distribution map and the time-domain analysis window to output the obstacle collision risk index, including: The rule-based decision-making module extracts the echo intensity change value in the lateral direction of the parking path from the compensated obstacle thermal distribution map; Extract the characteristics of the rate of change of echo intensity within the corresponding time range from the time domain analysis window; The system analyzes the changes in echo intensity and the rate of change using preset rules. Based on the results of the analysis, it adjusts the collision risk index by increasing or decreasing it, and finally outputs the adjusted collision risk index representing the probability of a collision.

8. A vehicle control system, characterized in that, The system includes: The perception enhancement module is used to collect millimeter-wave radar data containing the outline and shape features of obstacles and the movement information of surrounding vehicles during the automatic parking process. The spatial resolution of the millimeter-wave radar data in the lateral direction of the parking path is improved by the radio frequency enhancement component, and a two-dimensional obstacle thermal distribution map of the parking area is generated. The dynamic window module is used to classify the emergency level of the obstacle avoidance behavior based on the real-time relative distance between the vehicle and surrounding obstacles, and dynamically extract the time domain analysis window corresponding to the obstacle heat distribution map according to the emergency level. The mapping and association module is used to establish a mapping relationship library between obstacle material type and historical parking obstacle avoidance failure records, and to associate the peak speed of surrounding moving obstacles to obtain the associated mapping relationship library; The compensation decision module is used to compensate for the signal baseline shift of the obstacle thermal distribution map by integrating the attenuation of electromagnetic waves in rainy weather based on the mapping relationship library. The rule decision module processes the compensated obstacle thermal distribution map and the time domain analysis window, and outputs the collision risk index of the obstacle. The damping reconstruction module is used to reconstruct the damping response function of the parking steering mechanism based on the collision risk index and the peak speed, and to perform the anti-collision path correction action in the automatic parking process. The closed-loop correction module is used to extract the vibration characteristics of the steering motor in real time during the parking and steering process, back-derive the torque distribution of the wheels and steering mechanism, and correct the damping response function in a closed loop. Through closed-loop control of all steps, an intelligent control method for obstacle avoidance in the automatic parking process is realized.

9. A vehicle control device, characterized in that, The device includes: a memory, a processor, and a vehicle control program stored in the memory and executable on the processor, the vehicle control program being configured to implement the steps of the vehicle control method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a vehicle control program, which, when executed by a processor, implements the steps of the vehicle control method as described in any one of claims 1 to 7.

Images