Self-adaptive track control method for unmanned mine car

Abstract

The invention relates to the field of automatic driving, in particular to a self-adaptive track control method for an unmanned mine car. Comprising the following steps: acquiring motion track parameters of the unmanned mine car: judging whether a road is curved or straight, deviating a safe space, judging whether a target vehicle exists in a target area or not, following a target number of the unmanned mine car, a longitudinal distance between the target vehicle and the unmanned mine car, a transverse distance between the target vehicle and the unmanned mine car, and a speed difference between the unmanned mine car and the target vehicle; preprocessing the motion trail parameters to obtain input features; inputting the input features into a preset neural network model to obtain the driving track offset of the unmanned mine car; the driving track offset is compared with two preset early warning threshold values, the unmanned mine car is controlled to turn back to the preset driving path based on the comparison result, and the second early warning threshold value is larger than the first early warning threshold value. According to the invention, through accurate feature extraction and screening and a two-stage early warning offset control strategy, early prediction and correction of trajectory offset are realized, and rear-end collision danger is greatly reduced.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving, and in particular to an adaptive trajectory control method for an unmanned mining vehicle. Background Art

[0002] Unmanned mining vehicles are automated equipment in open-pit mines. They use unmanned driving technology to automatically load, unload and transport ore, significantly improving the production efficiency of mines. However, during the unmanned driving process, due to the complexity and unpredictability of road conditions, the mining vehicle may experience trajectory deviation, that is, the deviation between the vehicle's driving trajectory and the predetermined trajectory. This deviation may cause the vehicle to deviate from the predetermined route or collide with other vehicles or obstacles. Therefore, it is crucial to warn and correct the trajectory deviation.

[0003] The existing unmanned mine car trajectory deviation warning and correction methods are mainly achieved through unmanned driving systems. The deviation control processing methods include speed reduction, emergency braking, and reminding the driver to take over. However, this processing method will result in a poor driving experience, panic among drivers and passengers, and even cause collisions. Especially in the complex environment of open-pit mines, due to the large size and heavy load of mine cars, irregular roads and poor road conditions, emergency braking increases the risk of rear-end collisions, and after the risk is eliminated, it needs to accelerate again, resulting in increased vehicle energy consumption. Summary of the invention

[0004] In order to solve the above problems, the present invention provides an adaptive trajectory control method for an unmanned mining vehicle.

[0005] The method includes:

[0006] Step 1: Obtain the motion trajectory parameters of the unmanned mining vehicle, which include road curvature judgment, offset safety space, whether the target vehicle exists in the target area, the number of unmanned mining vehicles following the target vehicle, the longitudinal distance between the target vehicle and the unmanned mining vehicle, the lateral distance between the target vehicle and the unmanned mining vehicle, and the speed difference between the unmanned mining vehicle and the target vehicle;

[0007] Step 2: preprocessing the motion trajectory parameters to obtain input features;

[0008] Step 3: Input the input features into the preset neural network model to obtain the driving trajectory offset of the unmanned mining vehicle;

[0009] Step 4: compare the driving trajectory offset with the preset second warning threshold and the preset first warning threshold, and based on the comparison result, control the unmanned mining vehicle to turn back to the preset driving path, and the second warning threshold is greater than the first warning threshold.

[0010] Furthermore, the steps of step 2 include:

[0011] De-noising the motion trajectory parameters by wavelet denoising;

[0012] Extract features from the denoised motion trajectory parameters;

[0013] The extracted features are screened using the Pearson analysis method to obtain the input features.

[0014] Furthermore, the step of extracting features from the denoised motion trajectory parameters includes:

[0015] Based on the preset signal processing method, the features of the denoised motion trajectory parameters are extracted, and the time domain, frequency domain and time-frequency domain feature features are retained according to the preset proportions.

[0016] Furthermore, the network structure of the preset neural network model is an n1-n2-1 structure, wherein n2=2*n1+1, n1 is the number of neurons in the input layer, n2 is the number of neurons in the hidden layer, and 1 is the output layer.

[0017] Furthermore, the specific steps of step 4 include:

[0018] Determining whether the driving trajectory deviation exceeds a preset second warning threshold;

[0019] If the second warning threshold is exceeded, active steering control and independent differential braking control are used to control the unmanned mining vehicle to turn back to the preset driving path;

[0020] If the second warning threshold is not exceeded, determining whether the driving trajectory deviation exceeds a preset first warning threshold;

[0021] If the first warning threshold is exceeded, independent differential braking control is used to control the unmanned mining vehicle to turn back to the preset driving path;

[0022] If it does not exceed the first warning threshold, no action will be taken.

[0023] Furthermore, the step of controlling the unmanned mining vehicle to turn back to the preset driving path in step 4 includes:

[0024] Add safety constraints to the motion parameters of the unmanned mining vehicle;

[0025] According to the safety constraints, the unmanned mining vehicle is controlled to turn back to a preset driving path.

[0026] Furthermore, the step of adding safety constraints to the preset motion parameters of the unmanned mining vehicle includes:

[0027] A maximum center of mass sideslip angle threshold is set for the center of mass sideslip angle of the unmanned mining vehicle, and / or a maximum yaw angular velocity threshold is set for the yaw angular velocity of the unmanned mining vehicle.

[0028] Furthermore, according to the safety constraint, the step of controlling the unmanned mining vehicle to turn back to the preset driving path includes:

[0029] Monitor the real-time motion parameters of the unmanned mining vehicle and determine whether the real-time motion parameters meet the safety constraints;

[0030] If it does not meet the requirements, active steering control and / or independent differential braking control are used to correct the motion parameters and control the unmanned mining vehicle to turn back to the preset driving path.

[0031] The system in the present invention corresponds to the method, and the specific preferred embodiments of the method are also applicable to the system.

[0032] One or more technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages:

[0033] The present invention achieves early prediction and correction of trajectory deviation through precise feature extraction and screening and dual-level early warning deviation control strategy, greatly reducing the risk of rear-end collision. At the same time, by avoiding emergency braking and re-acceleration, energy consumption is effectively reduced. In addition, this smooth and timely correction control improves the driving stability of the unmanned mining vehicle, thereby enhancing the driving experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 The embodiment of the present invention provides DETAILED DESCRIPTION

[0035] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. Before describing in detail the technical solutions of each embodiment of the present invention, the nouns and terms involved are explained. In this specification, components with the same name or the same number represent similar or identical structures and are only for illustrative purposes.

[0036] The method of the present invention is as follows Figure 1 As shown, the specific steps are as follows:

[0037] 1. Data Acquisition

[0038] 1.1 Determine the straightness of the road

[0039] The camera captures the edge of the road or the marking line to determine whether the vehicle deviates from the driving track. For road recognition in different lighting environments in open-pit mines. Extract the local gradient features, local grayscale features and direction features of the road boundary that are less affected by the global grayscale and image noise; at the same time, the information of multiple features of the road boundary and the proportion of effective statistical units is fused to accurately fit the road curve of the unmanned mining vehicle.

[0040] The camera first acquires the image in front of the vehicle, then optimizes the image through preprocessing steps (such as noise reduction, contrast enhancement, color space conversion, etc.), then performs image segmentation to distinguish between road and non-road areas, and then obtains the position and shape information of the road edge through feature extraction algorithms such as edge detection and Hough transform. Finally, based on the shape of the road edge, if it is a straight line or an approximate straight line, the road is judged to be straight, and if it is a curve, the road is judged to be curved.

[0041] 1.2 Offset Safe Space

[0042] LiDAR acquires high-precision three-dimensional information about the vehicle's surroundings by emitting laser pulses and receiving reflected light waves. It then calculates the relative position and distance between the vehicle's current position and surrounding obstacles (such as other vehicles, rocks, walls, etc.) to obtain a safe space for the vehicle to deviate from the center line of the road.

[0043] 1.3 Whether the target vehicle exists in the target area

[0044] In this embodiment, the target vehicle refers to other vehicles that the unmanned mine car needs to pay attention to and avoid. The target area is defined as the area within a preset angle range in front of and on the sides of the unmanned mine car. Then, a camera is used to capture images in this area, and a target detection algorithm is used to identify other mine cars or obstacles in the image. If other mine cars are detected, it is considered that the target vehicle exists in the target area.

[0045] 1.4 Number of unmanned mining vehicles to follow

[0046] In this embodiment, the following vehicle target refers to other vehicles in front of the unmanned mining vehicle that need to be paid attention to during driving and may need to adjust the driving direction or speed according to its driving status. The camera captures the image in front of the unmanned mining vehicle, identifies and tracks other vehicles in the image through the target detection and tracking algorithm, and counts the number of other vehicles in front of the unmanned mining vehicle. This number is the number of following vehicle targets of the unmanned mining vehicle.

[0047] 1.5 Longitudinal distance between target vehicle and unmanned mining vehicle

[0048] In this embodiment, the longitudinal distance refers to the distance between the unmanned mining vehicle and the target vehicle in the direction of travel. The radar measures the distance between the vehicle and the obstacle or other vehicle in front by transmitting a signal and receiving the reflected signal. If the obstacle is another vehicle, then the measured distance is the longitudinal distance between the target vehicle and the unmanned mining vehicle. Road boundary enhancement detection method based on lidar. Background separation is achieved based on the distribution characteristics of the scanned point cloud. This process removes the road surface while retaining obstacles and part of the road boundary, and combines the foreground and near-field separation results to perform long-distance expansion of the road boundary.

[0049] 1.6 Lateral distance between target vehicle and unmanned mining vehicle

[0050] In this embodiment, the lateral distance refers to the distance between the unmanned mining vehicle and the target vehicle in a direction perpendicular to the travel direction. The radar calculates the lateral position of the target vehicle relative to the unmanned mining vehicle by measuring the angle difference between the transmitted signal and the reflected signal, thereby obtaining the lateral distance between the target vehicle and the vehicle.

[0051] 1.7 Speed ​​difference between the unmanned mining vehicle and the target vehicle

[0052] In the application scenario of unmanned mining trucks, the on-board GPS can obtain the speed of the unmanned mining truck, compare the GPS speed information of the target vehicle and the unmanned mining truck, and calculate the difference between the two to obtain the speed difference between the unmanned mining truck and the target vehicle.

[0053] 2. Data processing and feature extraction

[0054] 2.1 Wavelet denoising

[0055] Wavelet denoising is an effective denoising method in signal processing. It uses the multi-scale decomposition capability of wavelet transform to convert the original signal into the wavelet domain, where it is easier to distinguish between signal and noise. It then processes the wavelet coefficients by setting a threshold to suppress noise, and finally restores the processed signal to the time domain through inverse wavelet transform.

[0056] Due to the entry and exit of external obstacles during the driving of the unmanned mine car, the signal type collected by the unmanned mine car is periodic. Through the analysis of historical data, the driving deviation contained in the collected signal is mainly concentrated in the low-frequency part, while the interference noise is concentrated in the high-frequency part. Therefore, a one-layer wavelet decomposition is used, and the wavelet reconstruction signal function is used to reconstruct the low-frequency signal, so as to achieve the purpose of controlling the useless part of the signal and enhancing the useful part of the signal.

[0057] 2.2 Feature Extraction

[0058] For the seven types of data extracted in step 1, each is regarded as a signal type, and after wavelet denoising, the features are extracted using the preset signal processing method. These features include time domain, frequency domain and time-frequency domain features, and their corresponding feature numbers are 14, 4 and 32 respectively, for a total of 50 features.

[0059] 2.3 Feature Screening

[0060] In the present invention, 50 features are extracted from seven signal types, and the feature space dimension obtained is as high as 350 dimensions. However, not all of these features have a strong correlation with the offset of the unmanned mining vehicle's driving trajectory. Some features may have a weak correlation with the offset, or even no correlation. Such features are not helpful for model training and may even have a negative impact on model performance.

[0061] Therefore, it is necessary to screen these features to find out the truly meaningful features that are strongly correlated with the deviation of the unmanned mining vehicle's travel trajectory. In the present invention, the Pearson correlation analysis method is used for preliminary screening. Pearson correlation analysis is a commonly used feature screening method, which measures the linear correlation between each feature and the target variable by calculating the correlation coefficient between them. The present invention sets a fixed parameter as a threshold value. Only when the absolute value of the correlation coefficient between the feature and the trajectory deviation is greater than this threshold value, the feature is considered to be meaningful and retained.

[0062] However, Pearson correlation analysis can only capture the linear relationship between features and target variables, but not the nonlinear relationship. In addition, the features selected by Pearson correlation analysis may be redundant, that is, some features may contain information that highly overlaps with other features, and such features are not helpful for model training.

[0063] Therefore, the present invention adopts a method based on feature importance for secondary screening.

[0064] 3. Migration result prediction

[0065] The filtered feature values ​​are input into the pre-built neural network model. A neural network is a model composed of multiple layers of neurons, which can perform nonlinear mapping on the input data to capture the complex relationship between the data.

[0066] The present invention uses a specific neural network structure, namely, n1-n2-1 structure, where n2=2*n1+1, n1 is the number of neurons in the input layer, n2 is the number of neurons in the hidden layer, and 1 is the output layer. This structure has good performance and can effectively handle the trajectory deviation prediction task of the present invention. The filtered eigenvalues ​​are combined into a eigenvector as the input of the neural network, and the unmanned mining vehicle driving trajectory deviation is used as the output of the network. The network learns how to predict the trajectory deviation based on the input eigenvector through training.

[0067] The present invention adopts the cross-validation method to perform model validation. This method divides the data set into multiple different training sets and test sets, and then repeatedly trains and tests, so as to obtain the performance of the model on different data and provide a robust estimate of the model performance.

[0068] The optimization goal of the model is to identify the square value of the error, which emphasizes the penalty for large errors and helps to train a more accurate model.

[0069] 4. Dual-level warning deviation control

[0070] The present invention selects to use active steering control of the front wheels or independent differential braking control of the rear wheels according to the deviation of the driving track of the unmanned mining vehicle. The purpose of this strategy is to enable the unmanned mining vehicle to effectively track the track while maintaining driving stability.

[0071] Determine the deviation of the driving track of the unmanned mine car predicted by the neural network model. If the deviation of the driving track reaches the preset first warning threshold, use independent differential braking control to change the braking force of the rear wheels to make the unmanned mine car return to the center line of the lane. Independent differential braking control refers to a method of changing the driving direction of the unmanned mine car by controlling the braking force of the rear wheels of the unmanned mine car. In this method, the braking force of each rear wheel can be controlled independently, so the steering of the unmanned mine car can be achieved by changing the difference in the braking force of the two rear wheels.

[0072] If the deviation of the driving track reaches the preset second warning threshold, the coordinated control of active steering control and independent differential braking control is used to quickly return the unmanned mining vehicle to the lane line by changing the steering angle of the front wheels and the braking force of the rear wheels. The second warning threshold is greater than the first warning threshold. Active steering control refers to a method of controlling the driving direction of the unmanned mining vehicle by changing the steering angle of the front wheels of the unmanned mining vehicle. In this method, the steering angle is usually controlled by the steering system of the unmanned mining vehicle (such as a steering wheel or a steering motor).

[0073] In order to effectively reduce the variables related to the stability of the unmanned mine car when the dual-level warning offset control takes effect and ensure the driving stability of the unmanned driving, safety constraints are added to the motion parameters of the unmanned mine car, and differential braking is used to suppress the sideslip angle of the center of mass, thereby improving the trajectory tracking ability and driving stability of the unmanned mine car. The motion parameters include the sideslip angle of the center of mass of the mine car and the yaw angular velocity.

[0074] Adding safety constraints to motion parameters means constraining motion parameters not to exceed the preset safety range. For example, a maximum center of mass side slip angle threshold is set for the center of mass side slip angle of the unmanned mining vehicle, and / or a maximum yaw angular velocity threshold is set for the yaw angular velocity of the unmanned mining vehicle. Too large a center of mass side slip angle may cause the vehicle to roll over, and too large a yaw angular velocity may cause the vehicle to lose control. By setting reasonable thresholds for these parameters, the stable operation of the vehicle can be guaranteed to a certain extent.

[0075] Independent differential brake control can change the direction and stability of the vehicle by changing the difference in braking force between the rear wheels. For example, if the center of mass slip angle is detected to be too large, the braking force on the side with the larger slip angle can be increased and the braking force on the side with the smaller slip angle can be reduced to adjust the vehicle back to a stable state. Similarly, if the yaw rate is detected to be too large, the increase in yaw rate can be suppressed by adjusting the braking force accordingly.

[0076] The above-described embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. An adaptive trajectory control method for an unmanned mining vehicle, characterized in that: The following steps are involved: Step 1: Obtain the motion trajectory parameters of the unmanned mining vehicle, which include road curvature judgment, offset safety space, whether the target vehicle exists in the target area, the number of unmanned mining vehicles following the target vehicle, the longitudinal distance between the target vehicle and the unmanned mining vehicle, the lateral distance between the target vehicle and the unmanned mining vehicle, and the speed difference between the unmanned mining vehicle and the target vehicle; Step 2: preprocessing the motion trajectory parameters to obtain input features; Step 3: Input the input features into the preset neural network model to obtain the driving trajectory offset of the unmanned mining vehicle; Step 4: compare the driving trajectory offset with the preset second warning threshold and the preset first warning threshold, and based on the comparison result, control the unmanned mining vehicle to turn back to the preset driving path, and the second warning threshold is greater than the first warning threshold.

2. The method for adaptive trajectory control of an unmanned mining vehicle according to claim 1, characterized in that: Step 2 includes: De-noising the motion trajectory parameters by wavelet denoising; Extract features from the denoised motion trajectory parameters; The extracted features are screened using the Pearson analysis method to obtain the input features.

3. The method for adaptive trajectory control of an unmanned mining vehicle according to claim 1, characterized in that: The step of extracting features from the denoised motion trajectory parameters comprises: Based on the preset signal processing method, the features of the denoised motion trajectory parameters are extracted, and the time domain, frequency domain and time-frequency domain feature features are retained according to the preset proportions.

4. The method for adaptive trajectory control of an unmanned mining vehicle according to claim 1, characterized in that: The network structure of the preset neural network model is an n1-n2-1 structure, wherein n2=2*n1+1, n1 is the number of input layer neurons, n2 is the number of hidden layer neurons, and 1 is the output layer.

5. The method for adaptive trajectory control of an unmanned mining vehicle according to claim 1, characterized in that: The specific steps of step 4 include: Determining whether the driving trajectory deviation exceeds a preset second warning threshold; If the second warning threshold is exceeded, active steering control and independent differential braking control are used to control the unmanned mining vehicle to turn back to the preset driving path; If the second warning threshold is not exceeded, determining whether the driving trajectory deviation exceeds a preset first warning threshold; If the first warning threshold is exceeded, independent differential braking control is used to control the unmanned mining vehicle to turn back to the preset driving path; If it does not exceed the first warning threshold, no action will be taken.

6. The method for adaptive trajectory control of an unmanned mining vehicle according to claim 1, characterized in that: The steps of controlling the unmanned mining vehicle to turn back to the preset driving path in step 4 include: Add safety constraints to the motion parameters of the unmanned mining vehicle; According to the safety constraints, the unmanned mining vehicle is controlled to turn back to a preset driving path.

7. The method for adaptive trajectory control of an unmanned mining vehicle according to claim 6, characterized in that: The step of adding safety constraints to the preset motion parameters of the unmanned mining vehicle includes: A maximum center of mass sideslip angle threshold is set for the center of mass sideslip angle of the unmanned mining vehicle, and / or a maximum yaw angular velocity threshold is set for the yaw angular velocity of the unmanned mining vehicle.

8. The method for adaptive trajectory control of an unmanned mining vehicle according to claim 6, characterized in that: According to the safety constraints, the steps of controlling the unmanned mining vehicle to turn back to the preset driving path include: Monitor the real-time motion parameters of the unmanned mining vehicle and determine whether the real-time motion parameters meet the safety constraints; If not, active steering control and / or independent differential braking control are used to correct the motion parameters and control the unmanned mining vehicle to turn back to the preset driving path.

Images