Underground garage automatic driving safety control system based on friction coefficient perception

Abstract

The invention discloses an underground garage automatic driving safety control system based on friction coefficient perception, belongs to the technical field of automatic driving environment perception and control, and solves the problems of perception failure, inaccurate friction modeling, unreasonable path planning and the like in complex weather and scenes in the prior art. The environment sensing module is used for collecting road surface states, vehicle dynamic states and environment parameters, the multi-source data processing and fusion module is used for extracting key features and realizing cross-modal fusion, the friction heat map generation module is used for constructing a road surface friction coefficient space distribution model, and the path planning module is combined for generating a safety track avoiding a low-friction area. And finally, trajectory tracking and dynamic correction are realized through a self-adaptive control module. Besides, a multi-objective optimization algorithm and a feedback control mechanism are introduced, safety, comfort and driving efficiency are considered, the robustness and reliability of the autonomous vehicle in a complex scene of the underground garage in rainy and snowy weather are remarkably improved, and the method has good engineering practicability.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving environmental perception and control technology, and in particular relates to an autonomous driving safety control system for underground parking garages based on friction coefficient perception. Background Technology

[0002] Underground parking garages, as typical enclosed environments for autonomous vehicles, are prone to water accumulation, snow accumulation, and icy surfaces during rainy or snowy weather, leading to a significant decrease in the road surface friction coefficient. Existing technologies have clear limitations in this scenario:

[0003] In terms of environmental perception, ordinary visual sensors are affected by rain and snow and water reflection, resulting in a decrease in image signal-to-noise ratio of more than 30% and a misjudgment rate of up to 40% on slippery roads.

[0004] LiDAR reduces reflectivity by 50% on slippery surfaces, and snow scattering causes point cloud noise to account for up to 30%. Conventional millimeter-wave radar, due to its low spatial resolution (angular resolution of only 2°×2°), cannot distinguish the difference in water depth within 10cm, making it difficult for a single sensor to achieve comprehensive and reliable perception.

[0005] In terms of friction modeling, traditional empirical formulas based on vehicle speed and slip ratio (such as the Burckhardt model) do not incorporate key parameters such as ambient temperature and humidity. In icy scenarios, the estimation error exceeds 0.2, and it lacks spatial distribution representation, failing to reflect the frictional differences in local areas such as ramp corners and poorly drained areas within the garage.

[0006] At the path planning and control level, the traditional RRT* algorithm takes path length as the main optimization objective. In low-friction areas (μ<0.4), the safety redundancy is insufficient, which can easily lead to insufficient braking distance. The fixed-parameter PID controller increases the steering overshoot by 2-3 times on low-friction surfaces. Conventional braking strategies do not consider the dynamic changes in the friction coefficient, and the braking distance deviation is more than 15%.

[0007] Therefore, there is an urgent need to build an integrated technical solution of "perception-modeling-planning-control" to achieve safe and reliable autonomous driving in underground parking garages during rain and snow. Summary of the Invention

[0008] To address the problems of existing technologies, this invention proposes an autonomous driving safety control system for underground parking garages based on friction coefficient perception. This system is designed to solve problems such as perception failure, inaccurate friction modeling, and unreasonable path planning in complex weather and scenarios.

[0009] The technical solution protected by this invention is: an autonomous driving safety control system for underground parking garages based on friction coefficient perception, comprising an environmental perception module, a multi-source data processing and fusion module, a friction force heat map generation module, a path planning module, and an adaptive control module; the environmental perception module collects multi-dimensional data of the underground parking garage in rainy or snowy weather based on polarization cameras, lidar, millimeter-wave radar, IMU, wheel speed sensors, and environmental sensors; the multi-source data processing and fusion module preprocesses and fuses the collected multi-dimensional data to output a fused feature vector. The frictional thermal map generation module uses fused features The system generates a friction coefficient heatmap as input, the path planning module plans a safe path by improving the RRT algorithm, and finally the adaptive control module realizes trajectory tracking to achieve safe autonomous driving in underground parking garages during rain and snow.

[0010] Furthermore, the specific process of the environmental perception module is as follows:

[0011] S11. Acquire road surface image data through four channels of a polarization camera (0°, 45°, 90°, 135°) to obtain polarization characteristics of watermarks, snow accumulation, and ice areas;

[0012] S12. Use a 128-line lidar to collect three-dimensional point cloud data and record information on road surface unevenness, water depth and snow thickness.

[0013] S13. Collect road surface reflection signals using 77GHz millimeter-wave radar to distinguish the differences in reflection characteristics between dry and wet areas;

[0014] S14. Use IMU and wheel speed sensors to collect vehicle dynamic data, including vehicle speed, acceleration, heading angle and wheel slip ratio;

[0015] S15. Collect environmental data, including road surface temperature, air humidity and dew point temperature, through an infrared thermometer and humidity sensor;

[0016] S16. Organize the collected data into standardized datasets, including image datasets, point cloud datasets, radar signal datasets, vehicle dynamic datasets, and environmental parameter datasets.

[0017] Furthermore, the specific process of the multi-source data processing and fusion module is as follows:

[0018] S21. Image Preprocessing: A U-Net-based rain and snow removal network is used to remove image noise. The formula is as follows:

[0019]

[0020] in, The original image containing rain and snow. As a rain / snow mask, a binary matrix is ​​used to label the rain / snow regions in the image. For rain and snow kernel functions, use a convolution template to simulate rain and snow patterns. For convolution operations, this formula is used to separate rain and snow noise from effective road surface features;

[0021] S22. Point cloud preprocessing: Noise points are removed using statistical filtering and voxel filtering, as shown in the following formula:

[0022]

[0023] in, The original point cloud, The mean of the neighborhood points is the average of the coordinates of the 16 nearest neighbors of a given point. For Euclidean distance, The standard deviation of the neighborhood point coordinates is used to eliminate rain and snow noise points that deviate from the normal distribution.

[0024] S23. Radar signal processing: The constant false alarm rate (CFAR) algorithm is used to filter rain and snow clutter. The formula is as follows:

[0025]

[0026] in, For the detection threshold, This is the threshold coefficient, ranging from 2.5 to 4.0. The mean of background noise is the average clutter power obtained through sliding window statistics. This formula is used to distinguish between effective road surface reflection signals and rain and snow interference signals.

[0027] S24. Feature Extraction:

[0028] Image features: Road surface state features are extracted using the MobileNetV3 network, as shown in the following formula:

[0029]

[0030] in, For the attention mechanism module, the feature weights of key areas such as watermarks and snow accumulation are enhanced, and a 256-dimensional road surface semantic feature vector is output.

[0031] Point cloud features: 3D morphological features are extracted using the PointNet network, as shown in the following formula:

[0032]

[0033] Among them, SA is the sampling and grouping module, which aggregates local features based on ball query and outputs a 512-dimensional three-dimensional structural feature vector;

[0034] Dynamic characteristics: Time-series data is processed through a bidirectional LSTM network, as shown in the following formula:

[0035]

[0036] in, It is a bidirectional long short-term memory network that can simultaneously capture past and future features of time-series data, outputting a 128-dimensional vehicle dynamic feature vector. Vehicle speed, unit: m / s Acceleration, unit: m / s² Slip ratio;

[0037] S25. Multimodal fusion: Multi-source features are fused using a Transformer encoder, as shown in the following formula:

[0038]

[0039] in, Radar characteristics, 256-dimensional reflection intensity characteristics. As environmental features, 64-dimensional temperature and humidity features are used to learn cross-modal associations through a self-attention mechanism, outputting a 1024-dimensional fused feature vector.

[0040] Furthermore, the specific process of the friction force heat map generation module is as follows:

[0041] S31. Construct a friction coefficient prediction network: using fused features As input, the friction coefficient is output through a 3-layer fully connected network, as shown in the following formula:

[0042]

[0043] in, These are fully connected layers, with dimensions of 1024→512→256→1 respectively. The activation function outputs the friction coefficient value of a single sampling point, ranging from 0 to 1.

[0044] S32. Grid map modeling: Divide the underground parking garage scene into 10cm×10cm grid units and associate them with static structure coordinates;

[0045] S33. Spatial Mapping of Friction Coefficient: The predicted values ​​are assigned to corresponding grid cells to generate a friction coefficient heatmap, as shown in the following formula:

[0046]

[0047] in, For grid The coordinates of the center point, in meters. The rasterized friction coefficient matrix has H×W dimensions;

[0048] S34. Heatmap Optimization: Combine historical data and environmental parameters to correct outliers and use Gaussian filtering to smooth the spatial distribution.

[0049] Furthermore, the specific process of the path planning module is as follows:

[0050] S41. Define the path cost function:

[0051]

[0052] in, Here, μ represents the average friction coefficient along the path, and μ is the mean value of the path-covered raster. This is the slope value, in degrees (°). Path length, in meters (m). The rate of change of curvature, in rad / m. These are the weighting coefficients, with initial values ​​of 0.5, 0.2, 0.2, and 0.1 respectively; one number was missing here and has been added.

[0053] S42. Improved RRT algorithm: Adopts a friction coefficient bias sampling strategy, prioritizing the selection of... The formula for the region is as follows:

[0054]

[0055] in, For the coordinates of the sampling point (two-dimensional coordinates in the underground parking garage scene) ), To retrieve the variable corresponding to the maximum value, For grid The coefficient of friction at that location This is a random perturbation term, ranging from 0 to 0.1, which is used to increase the sampling probability of high-friction areas.

[0056] S43. Uphill and downhill path optimization: The maximum safe vehicle speed is calculated based on the slope-friction coupling model, using the following formula:

[0057]

[0058] in, It is the acceleration due to gravity. For safety, a factor of 1.5 is used to output the maximum permissible speed for each road segment;

[0059] S44. Multi-objective optimization: The cost function weights are dynamically adjusted using the particle swarm optimization algorithm, as shown in the following formula:

[0060]

[0061] in, As a learning factor, It is a random number. For the individual's optimal, It is the global optimal solution.

[0062] Furthermore, the specific process of the adaptive control module is as follows:

[0063] S1. Model-based predictive control of vehicle speed adjustment, with the following objective function:

[0064]

[0065] in, The safe speed curve output by the route planning system is used as a reference speed. To control the input, acceleration commands, For time-domain prediction, 5 steps are taken, corresponding to 1 second;

[0066] S2. Adaptive PID steering control, which adjusts the proportional coefficient based on the friction coefficient, as shown in the following formula:

[0067]

[0068] in, , The basic proportional coefficient, The trajectory tracking error is expressed in meters (m), and the steering angle command is output.

[0069] S3, Emergency Braking Control: When slip ratio When ABS is activated, the formula is as follows:

[0070]

[0071] in, Wheel speed (unit: m / s) The anti-lock braking system is activated at the safety threshold.

[0072] The present invention has the following advantages compared with the prior art.

[0073] 1. This invention collects road surface conditions, vehicle dynamics and environmental parameters through an environmental perception module, extracts key features through a multi-source data processing and fusion module and achieves cross-modal fusion, constructs a spatial distribution model of road surface friction coefficient using a friction heat map generation module, generates a safe trajectory to avoid low friction areas by combining a path planning module, and finally achieves trajectory tracking and dynamic correction through an adaptive control module.

[0074] 2. This invention introduces a multi-objective optimization algorithm and feedback control mechanism, which takes into account safety, comfort and driving efficiency, and significantly improves the robustness and reliability of autonomous vehicles in complex scenarios such as underground parking garages in rainy and snowy weather, and has good engineering applicability.

[0075] 3. This invention is applicable to the entry, exit, driving and parking processes of L4 and above level autonomous vehicles in underground garages in rainy and snowy weather. By dynamically sensing the road surface friction coefficient and optimizing the driving trajectory, it solves the safety hazards caused by slippery road surfaces and provides core technical support for autonomous driving in complex weather and scenarios. Attached Figure Description

[0076] The present invention will now be described in further detail with reference to the accompanying drawings.

[0077] Figure 1 This is a system framework diagram of the present invention.

[0078] Figure 2 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0079] To make the objectives, features, and advantages of the present invention readily apparent, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0080] like Figure 1 As shown, the underground parking garage autonomous driving safety control system based on friction coefficient sensing includes an environmental perception module, a multi-source data processing and fusion module, a friction force heatmap generation module, a path planning module, and an adaptive control module. Its specific working process is as follows:

[0081] Step S1: The environmental perception module collects multi-dimensional data from the underground parking garage during rain and snow based on a polarization camera, LiDAR, millimeter-wave radar, IMU, wheel speed sensor, and environmental sensors. This multi-dimensional data includes: road surface images acquired by the polarization camera (camera), point cloud data acquired by the LiDAR, transmission data acquired by the millimeter-wave radar, vehicle motion states such as vehicle speed, acceleration, heading angle, and wheel slip ratio acquired by the IMU, and environmental data such as road surface temperature, air humidity, and dew point temperature collected by the environmental sensors. The specific process of the environmental perception module is as follows:

[0082] S11. Acquire road surface image data using a polarization camera (four channels: 0°, 45°, 90°, 135°) to obtain polarization characteristics of watermarks, snow accumulation, and icing areas.

[0083] S12. Use a 128-line lidar to collect three-dimensional point cloud data and record information on road surface unevenness, water depth, and snow thickness.

[0084] S13. Use 77GHz millimeter-wave radar to collect road surface reflection signals and distinguish the differences in reflection characteristics between dry and wet areas.

[0085] S14. Use IMU and wheel speed sensors to collect vehicle dynamic data, including vehicle speed, acceleration, heading angle and wheel slip ratio.

[0086] S15. Collect environmental data, including road surface temperature, air humidity and dew point temperature, through an infrared thermometer and humidity sensor.

[0087] S16. Organize the collected data into standardized datasets, including image datasets, point cloud datasets, radar signal datasets, vehicle dynamic datasets, and environmental parameter datasets, and output them to the multi-source data processing and fusion module after synchronization with timestamps.

[0088] Step S2: The multi-source data processing and fusion module will preprocess and fuse the collected multi-dimensional data to output a fused feature vector. The data that needs preprocessing includes image data. Point cloud data Reflected data. Different preprocessing methods are used for different data types, and corresponding feature extraction is performed on the preprocessed data to form their respective features. Image features, Point cloud features, dynamic features radar characteristics Environmental characteristics Finally, these features are fused to form a fused feature vector. .

[0089] like Figure 1 As shown, the specific process of the multi-source data processing and fusion module is as follows:

[0090] S21, Image Data Preprocessing: The U-Net architecture rain and snow removal network is used to remove image noise, as shown in the following formula:

[0091]

[0092] in, The original image containing rain and snow. This is a rain / snow mask (a binary matrix that marks rain / snow regions in an image). This is the rain / snow kernel function (a convolution template that simulates the shape of rain and snow). For convolution operations, the above formula is used to separate rain and snow noise from effective road surface features.

[0093] S22, point cloud data Preprocessing: Noise points are removed using statistical filtering and voxel filtering, as shown in the following formula:

[0094]

[0095] in, The original point cloud, It is the mean of the neighborhood points (the average coordinates of the 16 neighboring points around a given point). For Euclidean distance, denoted as the standard deviation of the neighborhood point coordinates, this formula is used to eliminate rain and snow noise points that deviate from the normal distribution.

[0096] S23. Preprocessing of radar signal reflection data: The constant false alarm rate (CFAR) algorithm is used to filter rain and snow clutter, as shown in the following formula:

[0097]

[0098] in, For the detection threshold, This is the threshold coefficient (usually taken as 2.5-4.0). The mean background noise (the average clutter power obtained through sliding window statistics) is used to distinguish between effective road surface reflection signals and rain / snow interference signals.

[0099] S24. Feature Extraction:

[0100] Image features: Road surface state features are extracted using the MobileNetV3 network, as shown in the following formula:

[0101]

[0102] in, For the attention mechanism module, the feature weights of key areas such as watermarks and snow accumulation are enhanced, and a 256-dimensional road surface semantic feature vector is output.

[0103] Point cloud features: 3D morphological features are extracted using the PointNet network, as shown in the following formula:

[0104]

[0105] Among them, SA is the sampling and grouping module, which aggregates local features based on ball query and outputs a 512-dimensional three-dimensional structural feature vector.

[0106] Dynamic characteristics: Time-series data is processed through a bidirectional LSTM network, as shown in the following formula:

[0107]

[0108] in, It is a bidirectional long short-term memory network that can simultaneously capture past and future features of time-series data, outputting a 128-dimensional vehicle dynamic feature vector. Vehicle speed, unit: m / s Acceleration, unit: m / s² Slip ratio;

[0109] S25. Multimodal fusion: Multi-source features are fused using a Transformer encoder, as shown in the following formula:

[0110]

[0111] in, Radar characteristics, 256-dimensional reflection intensity characteristics. As environmental features, 64-dimensional temperature and humidity features are used to learn cross-modal associations through a self-attention mechanism, outputting a 1024-dimensional fused feature vector.

[0112] Step S3: The frictional thermal map generation module fuses features. The specific process of generating a friction coefficient heatmap using the friction force heatmap generation module is as follows:

[0113] S31. Construct a friction coefficient prediction network: using fused features As input, the friction coefficient is output through a 3-layer fully connected network, as shown in the following formula:

[0114]

[0115] in, These are fully connected layers, with dimensions of 1024→512→256→1 respectively. The activation function outputs the friction coefficient value of a single sampling point, ranging from 0 to 1.

[0116] S32. Grid map modeling: Divide the underground parking garage scene into 10cm×10cm grid units and associate them with static structure coordinates;

[0117] S33. Spatial Mapping of Friction Coefficient: The predicted values ​​are assigned to corresponding grid cells to generate a friction coefficient heatmap, as shown in the following formula:

[0118]

[0119] in, For grid The coordinates of the center point, in meters. The rasterized friction coefficient matrix has H×W dimensions;

[0120] S34. Heatmap Optimization: Combine historical data and environmental parameters to correct outliers and use Gaussian filtering to smooth the spatial distribution.

[0121] Step S4: The path planning module plans a safe path by improving the RRT algorithm. The specific process of the path planning module is as follows:

[0122] S41. Define the path cost function:

[0123]

[0124] in, Here, μ represents the average friction coefficient along the path, and μ is the mean value of the path-covered raster. This is the slope value, in degrees (°). Path length, in meters (m). The rate of change of curvature, in rad / m. These are weighting coefficients, with initial values ​​of 0.5, 0.2, 0.2, and 0.1 respectively.

[0125] S42. Improved RRT algorithm: Adopts a friction coefficient bias sampling strategy, prioritizing the selection of... The formula for the region is as follows:

[0126]

[0127] in, For the coordinates of the sampling point (two-dimensional coordinates in the underground parking garage scene) ), To retrieve the variable corresponding to the maximum value, For grid The coefficient of friction at that location This is a random perturbation term, ranging from 0 to 0.1, which is used to increase the sampling probability of high-friction areas.

[0128] S43. Uphill and downhill path optimization: The maximum safe vehicle speed is calculated based on the slope-friction coupling model, using the following formula:

[0129]

[0130] in, It is the acceleration due to gravity. For safety, a factor of 1.5 is used to output the maximum permissible speed for each road segment;

[0131] S44. Multi-objective optimization: The cost function weights are dynamically adjusted using the particle swarm optimization algorithm, as shown in the following formula:

[0132]

[0133] in, As a learning factor, It is a random number. For the individual's optimal, It is the global optimal solution.

[0134] Step S5: The adaptive control module implements trajectory tracking to achieve safe autonomous driving in underground parking garages during rain and snow. The specific process of the adaptive control module is as follows:

[0135] S1. Model-based predictive control of vehicle speed adjustment, with the following objective function:

[0136]

[0137] in, The safe speed curve output by the route planning system is used as a reference speed. To control the input, acceleration commands, For time-domain prediction, 5 steps are taken, corresponding to 1 second;

[0138] S2. Adaptive PID steering control, which adjusts the proportional coefficient based on the friction coefficient, as shown in the following formula:

[0139]

[0140] in, , The basic proportional coefficient, The trajectory tracking error is expressed in meters (m), and the steering angle command is output.

[0141] S3, Emergency Braking Control: When slip ratio When ABS is activated, the formula is as follows:

[0142]

[0143] in, Wheel speed (unit: m / s) The anti-lock braking system is activated at the safety threshold.

[0144] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. An autonomous driving safety control system for underground parking garages based on friction coefficient sensing, characterized in that: The system includes an environmental perception module, a multi-source data processing and fusion module, a friction heatmap generation module, a path planning module, and an adaptive control module. The environmental perception module collects multi-dimensional data from the underground parking garage during rain and snow based on polarization cameras, LiDAR, millimeter-wave radar, IMU, wheel speed sensors, and environmental sensors. The multi-source data processing and fusion module preprocesses and fuses the collected multi-dimensional data to output a fused feature vector. The frictional thermal map generation module uses fused features The system generates a friction coefficient heatmap as input, the path planning module plans a safe path by improving the RRT algorithm, and finally the adaptive control module realizes trajectory tracking to achieve safe autonomous driving in underground parking garages during rain and snow.

2. The underground parking garage automatic driving safety control system based on friction coefficient sensing according to claim 1, characterized in that: The specific process of the environmental perception module is as follows: S11. Acquire road surface image data through four channels of a polarization camera (0°, 45°, 90°, 135°) to obtain polarization characteristics of watermarks, snow accumulation, and ice areas; S12. Use a 128-line lidar to collect three-dimensional point cloud data and record information on road surface unevenness, water depth and snow thickness. S13. Collect road surface reflection signals using 77GHz millimeter-wave radar to distinguish the differences in reflection characteristics between dry and wet areas; S14. Use IMU and wheel speed sensors to collect vehicle dynamic data, including vehicle speed, acceleration, heading angle and wheel slip ratio; S15. Collect environmental data, including road surface temperature, air humidity and dew point temperature, through an infrared thermometer and humidity sensor; S16. Organize the collected data into standardized datasets, including image datasets, point cloud datasets, radar signal datasets, vehicle dynamic datasets, and environmental parameter datasets.

3. The underground parking garage automatic driving safety control system based on friction coefficient sensing according to claim 2, characterized in that: The specific process of the multi-source data processing and fusion module is as follows: S21. Image Preprocessing: A U-Net-based rain and snow removal network is used to remove image noise. The formula is as follows: ； in, The original image containing rain and snow. To cover the code for rain and snow, For rain and snow kernel functions, This is a convolution operation; S22. Point cloud preprocessing: Noise points are removed using statistical filtering and voxel filtering, as shown in the following formula: ； in, The original point cloud, The mean of the neighborhood points. For Euclidean distance, The standard deviation of the coordinates of the neighboring points; S23. Radar signal processing: The constant false alarm rate (CFAR) algorithm is used to filter rain and snow clutter. The formula is as follows: ； in, For the detection threshold, For threshold coefficient, This represents the mean of the background noise. S24. Feature Extraction: Image features: Road surface state features are extracted using the MobileNetV3 network, as shown in the following formula: ； in, For the attention mechanism module; Point cloud features: 3D morphological features are extracted using the PointNet network, as shown in the following formula: ； Among them, SA is the sampling and grouping module, which aggregates local features based on ball query and outputs a 512-dimensional three-dimensional structural feature vector. Point cloud data with noise points removed; Dynamic characteristics: Time-series data is processed through a bidirectional LSTM network, as shown in the following formula: ； in, It is a bidirectional long short-term memory network that can simultaneously capture past and future features of time-series data, outputting a 128-dimensional vehicle dynamic feature vector. Vehicle speed, unit: m / s Acceleration, unit: m / s² Slip ratio; S25. Multimodal fusion: Multi-source features are fused using a Transformer encoder, as shown in the following formula: ； in, Radar characteristics, 256-dimensional reflection intensity characteristics. As environmental features, 64-dimensional temperature and humidity features are used to learn cross-modal associations through a self-attention mechanism, outputting a 1024-dimensional fused feature vector.

4. The underground parking garage automatic driving safety control system based on friction coefficient sensing according to claim 3, characterized in that: The specific process of the friction force heat map generation module is as follows: S31. Construct a friction coefficient prediction network: using fused features As input, the friction coefficient is output through a 3-layer fully connected network, as shown in the following formula: ； in, These are fully connected layers, with dimensions of 1024→512→256→1 respectively. The activation function outputs the friction coefficient value of a single sampling point, ranging from 0 to 1. S32. Grid map modeling: Divide the underground parking garage scene into 10cm×10cm grid units and associate them with static structure coordinates; S33. Spatial Mapping of Friction Coefficient: The predicted values ​​are assigned to corresponding grid cells to generate a friction coefficient heatmap, as shown in the following formula: ； in, For grid The coordinates of the center point, in meters. The rasterized friction coefficient matrix has dimensions of H×W; S34. Heatmap Optimization: Combine historical data and environmental parameters to correct outliers and use Gaussian filtering to smooth the spatial distribution.

5. The underground parking garage automatic driving safety control system based on friction coefficient sensing according to claim 4, characterized in that: The specific process of the path planning module is as follows: S41. Define the path cost function: ； in, Here, μ represents the average friction coefficient along the path, and μ is the mean value of the path-covered raster. This is the slope value, in degrees (°). Path length, in meters (m). The rate of change of curvature, in rad / m. These are weighting coefficients, with initial values ​​of 0.5, 0.2, 0.2, and 0.1 respectively. S42. Improved RRT algorithm: Adopts a friction coefficient bias sampling strategy, prioritizing the selection of... The formula for the region is as follows: ； in, For the coordinates of the sampling point (two-dimensional coordinates in the underground parking garage scene) ), To retrieve the variable corresponding to the maximum value, For grid The coefficient of friction at that location This is a random perturbation term, ranging from 0 to 0.1, which is used to increase the sampling probability of high-friction areas. S43. Uphill and downhill path optimization: The maximum safe vehicle speed is calculated based on the slope-friction coupling model, using the following formula: ； in, It is the acceleration due to gravity. For safety, a factor of 1.5 is used to output the maximum permissible speed for each road segment; S44. Multi-objective optimization: The cost function weights are dynamically adjusted using the particle swarm optimization algorithm, as shown in the following formula: ； in, As a learning factor, It is a random number. For the individual's optimal, It is the global optimal solution.

6. The underground parking garage automatic driving safety control system based on friction coefficient sensing according to claim 5, characterized in that: The specific process of the adaptive control module is as follows: S1. Model-based predictive control of vehicle speed adjustment, with the following objective function: ； in, The safe speed curve output by the route planning system is used as a reference speed. To control the input, acceleration commands, For time-domain prediction, 5 steps are taken, corresponding to 1 second; S2. Adaptive PID steering control, which adjusts the proportional coefficient based on the friction coefficient, as shown in the following formula: ； in, , The basic proportional coefficient, The trajectory tracking error is expressed in meters (m), and the steering angle command is output. S3, Emergency Braking Control: When slip ratio When ABS is activated, the formula is as follows: ； in, Wheel speed (unit: m / s) The anti-lock braking system is activated at the safety threshold.

Images