An automatic steering and driving humanoid trajectory planning method and system

Abstract

The present invention provides an automatic steering driving humanoid trajectory planning method and system. By collecting and processing human control data in a curved road scenario, the relative distance information between the vehicle and obstacles on both sides of the road and the vehicle kinematic information are obtained; based on the kernel principal component analysis (KPCA), non-linear features are extracted from the vehicle kinematic information; a multi-state loop prediction model is constructed based on the LSTM network, including weighted attention to key data of long-term and short-term information by cell state information; the non-linear features, the relative distance information between the vehicle and obstacles on both sides of the road, and the vehicle kinematic information are input into the multi-state loop prediction model to predict future driving trajectory points. Thereby improving the mapping accuracy of the network, better learning human control behaviors, and improving the stability of the planned trajectory.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving trajectory planning, and in particular, to a method and system for automatic steering driving humanoid trajectory planning. Background Art

[0002] In recent years, unattended technologies have been widely introduced in environments such as agriculture, environmental exploration, and disaster response. However, there are no obvious traffic markings on the roads in the above scenarios, and how the vehicle can turn safely and smoothly during automatic driving is an important issue that needs to be studied.

[0003] In the prior art, when applying, it is necessary to learn the human control characteristics based on learning algorithms, which can be roughly divided into indirect learning methods and direct learning methods. Under real driving conditions with complex uncertainty constraints, the optimization objectives and reward functions may not be comprehensive and accurate enough. And the trajectories generated by the indirect learning-based methods are usually selected from a candidate trajectory set generated from a fixed model or generated according to specific rules, and the selected trajectories may be different from the real trajectories of human drivers.

[0004] Therefore, how to plan a trajectory that conforms to human control behavior when the vehicle turns in a scenario without traffic markings is an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the purpose of the embodiments of the present invention is to use the vehicle speed, angular velocity, curvature of the curve change, non-linear characteristics among kinematic information, and the relative position relationship between obstacles on both sides of the road and the vehicle as inputs, and learn the human control behavior from the collected data through a deep learning model, so as to plan a trajectory route that conforms to human control habits in a curve scenario.

[0006] The first aspect of the present invention provides a method for automatic steering driving humanoid trajectory planning, and the method includes:

[0007] S101, collecting and processing human control data in a curve scenario to obtain the relative distance information between the vehicle and obstacles on both sides of the road and the vehicle kinematic information;

[0008] S102, extracting non-linear characteristics from the vehicle kinematic information based on the kernel principal component analysis method KPCA;

[0009] S103, establishing and training a multi-state recurrent prediction model; the multi-state recurrent prediction model is constructed based on the LSTM network, including key data of cell state information for weighted attention to long-term and short-term information; the model includes a state output gate;

[0010] S104. Input the non-linear feature, the relative distance information between the vehicle and the obstacles on both sides of the road, and the vehicle kinematic information into the multi-state cyclic prediction model to predict future driving trajectory points.

[0011] Further, the curve scenario includes an entering curve stage, a turning stage, and an exiting curve stage.

[0012] The vehicle kinematic information includes the vehicle driving speed and the vehicle driving angular velocity.

[0013] The S101 includes: calculating the distance positions between the vehicle and two obstacles on one side by using the relative position relationship between the current position of the vehicle and the obstacles on both sides.

[0014] Further, the S102 includes: mapping the input space data into the feature space, and then calculating the main components in the feature space.

[0015] The S103 includes: after extracting the main components by KPCA, taking the main components, the relative distance information between the vehicle and the obstacles on both sides of the road, and the vehicle kinematic information as the input of the deep network, and using the multi-state cyclic prediction model to predict future driving trajectory points, where a state gate structure is added to transmit the memory state.

[0016] Further, in the S103, the multi-state cyclic prediction model includes:

[0017] (10)

[0018] Where is the input gate, is the sigmoid activation function, is the weight matrix, is the short-term memory output, is the bias vector, is the input information;

[0019] (11)

[0020] Where is the sigmoid activation function, is the weight matrix, is the element-wise product, is the state cell, is the short-term memory output, is the bias vector, is the input information, is the forget gate;

[0021] (12)

[0022] Among them, is the forget gate, is the element-wise product, is the state cell, is the output of the forget gate;

[0023] (13)

[0024] Among them, is the non-linear transformation, is the state cell, is the weight matrix, is the activation function, is the element-wise product;

[0025] (14)

[0026] Among them, is the non-linear transformation, is the state cell, is the element-wise product, is the activation function, is the weight matrix, is the state transmission gate;

[0027] (15)

[0028] Among them, is the activation function, is the weight matrix, is the short-term memory output, is the bias vector, is the input information, is the output of the forget gate;

[0029] (16)

[0030] Among them, is the output of the forget gate, is the state transmission gate, is the input gate, is the element-wise product, is the upper layer output;

[0031] (17)

[0032] Among them, is the sigmoid activation function, is the weight matrix, is the short-term memory output, is the input information, is the output of the forget gate, is the bias vector, is the output gate;

[0033] (18)

[0034] wherein, is the activation function, is the state cell.

[0035] Furthermore, in S103, when training the multi-state recurrent prediction model, the mean square error (MSE) is used as the loss function to train the network, and the formula is:

[0036] ;

[0037] where N is the number of samples in the training dataset, is the actual trajectory coordinate value of the operator, is the predicted value of the trajectory coordinate, and the Adamw optimizer is used to adaptively optimize the network parameters.

[0038] The second aspect of the present invention provides an automatic steering driving humanoid trajectory planning system, which includes an acquisition and processing module, an analysis module, a model establishment module, and a prediction module.

[0039] The acquisition and processing module is used to acquire and process human operation data in a curved road scene, and obtain the relative distance information between the vehicle and the obstacles on both sides of the road and the vehicle kinematic information;

[0040] The analysis module is used to extract non-linear features from the vehicle kinematic information based on the kernel principal component analysis (KPCA);

[0041] The model establishment module is used to establish and train a multi-state recurrent prediction model; the multi-state recurrent prediction model is constructed based on the LSTM network, and includes key data of cell state information for weighting and attention to long-term and short-term information; the model includes a state output gate.

[0042] The prediction module is used to input the non-linear features, the relative distance information between the vehicle and the obstacles on both sides of the road, and the vehicle kinematic information into the multi-state recurrent prediction model to predict future driving trajectory points.

[0043] Furthermore, the curved road scene includes an entry phase, a turning phase, and an exit phase;

[0044] The vehicle kinematic information includes the vehicle driving speed and the vehicle driving angular velocity;

[0045] The acquisition and processing module further includes calculating the distance positions between the vehicle and two obstacles on one side by using the relative position relationship between the current position of the vehicle and the obstacles on both sides.

[0046] Further, the analysis module further includes a component for mapping the input spatial data into a feature space and then calculating the principal components in the feature space.

[0047] The model establishment module further includes a component for extracting the principal components through KPCA, using the principal components, the relative distance information between the vehicle and the obstacles on both sides of the road, and the vehicle kinematic information as the input of the deep network, and using the multi-state recurrent prediction model to predict the future driving trajectory points, wherein a state gate structure is added to transmit the memory state.

[0048] In addition, a third aspect of the present invention provides an electronic device, which includes: one or more processors and a memory. The memory is used to store one or more computer programs, and the computer programs are configured to be executed by the one or more processors. The programs include steps for executing the automatic steering driving humanoid trajectory planning method described in the first aspect above.

[0049] In addition, a fourth aspect of the present invention provides a storage medium that stores a computer program; the program is loaded and executed by a processor to implement the steps of the automatic steering driving humanoid trajectory planning method described in the first aspect above.

[0050] In the solution of the present invention, by collecting and processing human control data in a curve scenario, the relative distance information between the vehicle and the obstacles on both sides of the road and the vehicle kinematic information are obtained; based on the kernel principal component analysis method KPCA, non-linear features are extracted from the vehicle kinematic information; a multi-state recurrent prediction model is established and trained; the multi-state recurrent prediction model is constructed based on the LSTM network and includes a cell state information for weighting and focusing on key data of long-term and short-term information; the model includes a state output gate; the non-linear features, the relative distance information between the vehicle and the obstacles on both sides of the road, and the vehicle kinematic information are input into the multi-state recurrent prediction model to predict the future driving trajectory points. A human control data set is constructed for training and testing the deep network. Then, cell state information is introduced into the LSTM network to weight and focus on key data of long-term and short-term information, and the state output gate is used to alleviate the gradient disappearance phenomenon during the training process, thereby improving the mapping accuracy of the network, better learning human control behaviors, and improving the stability of the planned trajectory. Description of the Drawings

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as a limitation of the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0052] Figure 1 is the step flowchart of the automatic steering driving humanoid trajectory planning method disclosed in the embodiments of the present invention;

[0053] Figure 2 is the data acquisition device of the multi-modal sensing system disclosed in the embodiments of the present invention;

[0054] Figure 3 (a) is the road curvature of the curved road scene disclosed in the embodiments of the present invention;

[0055] Figure 3 (b) is the vehicle driving speed information disclosed in the embodiments of the present invention;

[0056] Figure 3 (c) is the vehicle driving angular velocity information disclosed in the embodiments of the present invention;

[0057] Figure 3 (d) is the relative position information between the vehicle and the left obstacle disclosed in the embodiments of the present invention;

[0058] Figure 3 (e) is the relative position information between the vehicle and the right obstacle disclosed in the embodiments of the present invention;

[0059] Figure 3 (f) is the data acquisition site disclosed in the embodiments of the present invention;

[0060] Figure 4 is the schematic diagram of the multi-state cyclic prediction model network structure disclosed in the embodiments of the present invention;

[0061] Figure 5 is the comparison of the parameter errors of each method disclosed in the embodiments of the present invention;

[0062] Figure 6 is the fitting situation of the predicted value and the true value of the proposed method disclosed in the embodiments of the present invention;

[0063] Figure 7 is the variable curvature curved road experiment disclosed in the embodiments of the present invention;

[0064] Figure 8 is the straight road - curved road scene experiment disclosed in the embodiments of the present invention;

[0065] Figure 9is an experiment on the curved - straight road scenario disclosed in the embodiments of the present invention;

[0066] Figure 10 is the prediction error of each parameter after migrating the weights in the embodiments of the present invention;

[0067] Figure 11 is the prediction error of each parameter without migrating the weights in the embodiments of the present invention. Detailed implementation manners

[0068] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more complete and comprehensive, and will fully convey the concept of the example embodiments to those skilled in the art.

[0069] In addition, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of the embodiments of this application. However, those skilled in the art will realize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. can be used. In other cases, well - known methods, devices, implementations, or operations are not shown or described in detail to avoid obscuring aspects of this application.

[0070] The block diagrams shown in the drawings are only functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0071] The flowcharts shown in the drawings are only illustrative and do not necessarily include all the content and operations / steps, nor do they necessarily have to be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to the actual situation.

[0072] It should be noted that: "a plurality" mentioned in this embodiment refers to two or more.

[0073] The implementation details of the technical solutions of the embodiments of this application are elaborated in detail below:

[0074] In this embodiment, a multi-modal sensing system is used to collect human-operated vehicle driving data, and non-linear features are extracted from the vehicle kinematic information by the KPCA method. Then, the relative position relationship between the current position of the vehicle and the obstacles on both sides of the road is calculated, and the relative position distance is restricted within the range of the front and rear two obstacles. On this basis, after combining with the vehicle kinematic state information in time series to screen the data with poor quality, a human-operated data set is constructed for the training and testing of the deep network. Then, the cell state information is introduced into the LSTM network to weight and focus on the key data of the long-term and short-term information, and the gradient disappearance phenomenon in the training process is alleviated through the state output gate, thereby improving the mapping accuracy of the network. After comparison with various methods, it can be seen from the Error, RMSE, and MAE indexes of the planned trajectory and the true value that the method in this embodiment can better learn human-operated behaviors, and the SD indexes of the speed and angular velocity can show that the planned trajectory of the method in this embodiment has better stability.

[0075] As Figure 1 shown is the step flowchart of the humanoid trajectory planning method for automatic steering driving in this embodiment. The method includes:

[0076] S101, collect and process the human-operated data in the curved road scenario, and obtain the relative distance information between the vehicle and the obstacles on both sides of the road and the vehicle kinematic information.

[0077] Specifically, in this embodiment, a multi-modal sensing system is used to collect the human-operated vehicle driving information and process it into a human-operated behavior data set, and then the future trajectory points are mapped through an improved deep learning network. Specifically as follows:

[0078] A. Collection and processing of human-operated data

[0079] Figure 2 is the data collection device of the multi-modal sensing system in this embodiment, which collects the human-operated data in the curved road scenario.

[0080] Furthermore, the curved road scenario includes an entry phase, a turning phase, and an exit phase; the vehicle kinematic information includes the vehicle driving speed and the vehicle driving angular velocity; the S101 includes: using the relative position relationship between the current position of the vehicle and the obstacles on both sides, calculating the distance positions between the vehicle and two obstacles on one side.

[0081] The specific curved road scenario is as Figure 3 shown. Figure 3 (a) is the road curvature of the curved road scenario; Figure 3 (b) is the vehicle driving speed information; Figure 3 (c) is the vehicle driving angular velocity information; Figure 3(d) is the relative position information between the vehicle and the obstacle on the left; Figure 3 (e) is the relative position information between the vehicle and the obstacle on the right; Figure 3 (f) is the data collection scene. This scene is roughly divided into the cornering stage - Turning phase - Corner exit , each stage presents a variable curvature shape, stage It is an approximate straight line section with a small change in curvature. The rear curvature increases significantly, and considering the stability and safety of the vehicle, the vehicle's driving radius increases first and then decreases, leaving Enter Afterwards, the operator controls the vehicle to slowly return to the virtual center line of the road (the test road in the figure has no center line or marking line) according to his own habits and environmental changes. The obstacles on both sides of the road are asymmetrically distributed. This embodiment uses the relative position relationship between the current position of the vehicle and the obstacles on both sides to calculate the distance between the vehicle and the two obstacles on one side. It can be seen from the figure that the distance between the obstacles on both sides changes regularly during the driving process of the vehicle. Therefore, this embodiment uses a deep network to learn human control habits from the processed state information.

[0082] S102, extracting nonlinear features from the vehicle kinematics information based on Kernel Principal Component Analysis (KPCA).

[0083] Furthermore, the S102 includes: mapping the input space data into the feature space, and then calculating the main components in the feature space.

[0084] Specifically, in this embodiment, since the collected human manipulation data presents a nonlinear relationship, in order to better learn human manipulation habits, this embodiment adopts the kernel principal component analysis (KPCA) method to map the input space data to the feature space while retaining the nonlinear characteristics of the original data as much as possible, and then calculate the main components in the feature space. , defined as,

[0085] , (1)

[0086] Here we assume and is a nonlinear mapping. In order to find the principal components, it is necessary to solve the eigenvalue problem in the feature space, as follows

[0087] , (2)

[0088] in , is an eigenvector and the coefficients , thus the following formula is obtained

[0089] , (3)

[0090] Formula 2 is expanded into

[0091] , (4)

[0092] After combining formulas (1), (3) and (4), the following formula is obtained

[0093] , (5)

[0094] After that, the inner product is replaced by and the multiplication is abbreviated. Then formula (5) can be equivalent to:[[]]

[0095] , (6)

[0096] Formula (6) can be further rewritten as:[[]]

[0097] (7)

[0098] where , is an (n×n) matrix defined as .

[0099] After that, this embodiment runs the PCA algorithm. The eigenvector is obtained, where the eigenvalue is . The first principal component of x times is mapped into , and the formula is as follows

[0100] (8)

[0101] The kernel function used in this embodiment is

[0102] (9)

[0103] S103. Establish and train a multi-state cyclic prediction model; the multi-state cyclic prediction model is constructed based on the LSTM network and includes key data of cell state information for weighted attention to long-term and short-term information; the model includes a state output gate.

[0104] Furthermore, in S103, when training the multi-state cyclic prediction model, the mean square error (MSE) is used as the loss function to train the network.

[0105] S103 includes: after extracting the principal components through KPCA, using the principal components, the relative distance information between the vehicle and the obstacles on both sides of the road, and the vehicle kinematic information as the input of the deep network, and using the multi-state recurrent prediction model to predict future driving trajectory points, where a state gate structure is added to transmit the memory state.

[0106] S104, input the non-linear features, the relative distance information between the vehicle and the obstacles on both sides of the road, and the vehicle kinematic information into the multi-state recurrent prediction model to predict future driving trajectory points.

[0107] Specifically, in this embodiment, the cell state information of the previous time step and the cell state information output by the forget gate in the current time step are referenced, the long-term and short-term key information is weighted and concerned, and the state transmission gate is introduced to alleviate the gradient disappearance phenomenon during the training process, thereby improving the mapping ability of the network. Specifically, after extracting the principal components through KPCA, together with the relative distance information between the vehicle and the obstacles on both sides of the road and the vehicle kinematic information processed in the previous section, they are used as the input of the deep network, and an improved multi-state recurrent prediction model is used to predict future driving trajectory points, where a state gate structure is added to transmit the memory state, thereby alleviating the gradient disappearance phenomenon during the training process and improving the mapping ability of the network.

[0108] As Figure 4 shown is the schematic diagram of the network structure of the multi-state recurrent prediction model of this embodiment. Among them, the state cell stores long-term memory information, which plays an important role in sequence prediction, while the hidden state stores key short-term memory information. Therefore, in this embodiment, the information in the state cell is weighted and concerned during the state update of the LSTM, and a state transmission gate is introduced to reduce the omission during the state information transmission process. Figure 4 The orange part in

[0109] (10)

[0110] Among them, is the input gate, is the sigmoid activation function, is the weight matrix, is the short-term memory output, is the bias vector, is the input information;

[0111] (11)

[0112] Among them, is the sigmoid activation function, is the weight matrix, is the element-wise product, is the state cell, is the short-term memory output, is the bias vector, is the input information, is the forget gate;

[0113] (12)

[0114] Among them, is the forget gate, is the element-wise product, is the state cell, is the output of the forget gate;

[0115] (13)

[0116] Among them, is the non-linear transformation, is the state cell, is the weight matrix, is the activation function, is the element-wise product;

[0117] (14)

[0118] Among them, is the non-linear transformation, is the state cell, is the element-wise product, is the activation function, is the weight matrix, is the state transmission gate;

[0119] (15)

[0120] Among them, is the activation function, is the weight matrix, is the short-term memory output, is the bias vector, is the input information, is the output of the forget gate;

[0121] (16)

[0122] Among them, is the output of the forget gate, is the state transmission gate, is the input gate, is the element-wise product, is the upper layer output;

[0123] (17)

[0124] Among them, is the sigmoid activation function, is the weight matrix, is the short-term memory output, is the input information, is the output of the forgetting gate, is the bias vector, is the output gate;

[0125] (18)

[0126] Among them, is the activation function, is the state cell.

[0127] After the model is established, in this embodiment, the processed human manipulation behavior data is used to train the improved network, and the input data is scaled to the range of [0, 1] using the Min-Max normalization method. The mean square error (MSE) is used as the loss function to train the network, and the formula is:

[0128] (19)

[0129] Among them, N is the number of samples in the training dataset, is the actual trajectory coordinate value of the operator, is the predicted trajectory coordinate value. The Adamw optimizer is used to adaptively optimize the network parameters. When making predictions after training, the output data is rescaled back to the normal value.

[0130] To verify the effectiveness of the method proposed in this embodiment, ablation experiment comparison, different curve scene experiment comparison, mainstream methods, and the latest time series prediction method experiment comparison are carried out in this embodiment. The anthropomorphic degree of the planned trajectories of each method is evaluated using Error, RMSE, and MAE between the predicted value and the true value, and the trajectory smoothness is evaluated by the SD (standard deviation) of the predicted speed and angular velocity.

[0131] In the experiment, in this embodiment, a series of experiments are carried out to analyze the influence of each part of the method proposed in this embodiment, and Error, RMSE, and MAE are used for evaluation. For the convenience of description, hereinafter, this embodiment uses Proposed to represent the method in this embodiment, L to represent the LSTM network, LC to represent the LSTM + state information method, LCR to represent the LSTM + state information + relative position information, and LCRG to represent the LSTM + state information + relative position information + state gate structure.

[0132] As Figure 5 shown in this embodiment is the comparison of the error of each method parameter; Figure 6 shown in this embodiment is the fitting situation of the predicted value and the true value of the proposed method. Table 1 shows the comparison of RMSE and MAE of the predicted parameters of different ablation experiment methods in this embodiment.

[0133] Table 1 Comparison of RMSE and MAE of predicted parameters of different ablation experiment methods

[0134] ;

[0135] From Figure 5 、 6 and Table 1, it can be seen that the method (Proposed) proposed in this embodiment is the best in terms of the prediction accuracy of the trajectory value, speed and angular velocity. When using the LSTM model (L) for prediction, the output error is relatively large. After cyclically applying the output cell state of the forget gate in the LSTM model (LC), the overall output accuracy is improved by 4.2%, indicating that the improved LSTM cell state cycle model has a promoting effect on the prediction accuracy; after adding the relative distance constraint of the obstacles on both sides of the road, the output accuracy is significantly improved by 41.5%, so it can be seen that the data processing method in this embodiment has high effectiveness; after adding the state transfer gate to the LSTM cell state cycle model (LCR), the output accuracy is improved by 47.7%, and it can be seen that the cell state information can provide more historical information in the long-term and short-term prediction process, thus enhancing the prediction ability of the model; after introducing the non-linear feature constraint, the prediction accuracy is improved by 75.6%, and it can be seen that the strong regularity of the processed non-linear features can better assist the model to learn the human control trajectory, so as to predict the trajectory information that conforms to the human control habit.

[0136] To verify the generalization of the method proposed in this embodiment, we conducted tests in three different curve scenarios: variable curvature curve, straight curve, and curve straight.

[0137] As Figure 7 shown in this embodiment is the variable curvature curve experiment; as Figure 8 shown in this embodiment is the straight curve scenario experiment; as Figure 9 shown in this embodiment is the curve straight scenario experiment. Specifically as Figures 7 - 9 shown, we migrated the weights trained in the previous section and performed training, and the data volume was reduced to 1 / 3 of the original, greatly reducing the data collection and processing costs. Comparing Figure 10 、 11As can be seen from Table 2, the prediction accuracy after migrating the pre-trained weights is 84.4% higher than that without migrating the pre-trained weights, verifying that the method in this embodiment has good generalization, indicating that the method in this embodiment can accurately predict trajectory information in the case of small batch data. Among them, Figure 10 are the prediction errors of each parameter after migrating the weights in this embodiment; Figure 11 are the prediction errors of each parameter without migrating the weights in this embodiment.

[0138] In Table 2, Scene represents the RMSE and MAE for prediction when migrating the pre-trained model in different scenarios, and NScene represents the RMSE and MAE for prediction when not migrating the pre-trained model in different scenarios. Figures 7 - 9 In (a), it is the driving route of the vehicle controlled by different operators in the variable curvature bend scenario; (b) is the comparison chart of the bend curvature and the collected data curvature, and the yellow box is the position where the road curvature changes greatly; (c) is the schematic situation of the abnormal scenario where danger is likely to occur; (d)-(k) are the normal driving processes of the vehicle from the starting point to the end point.

[0139] Table 2 Comparison of RMSE and MAE of predicted parameters in different bend scenarios

[0140] ;

[0141] Since there is little research on the turning of vehicles in an environment without traffic markings, and the method in this paper belongs to the time series prediction problem, we compared it with the mainstream methods and the latest time prediction methods, as follows:

[0142] (a) GRU: This model solves the problem of gradient disappearance or gradient explosion that occurs in the RNN network when processing long sequence data by introducing an update gate and a reset gate, thereby capturing long-term dependencies and reducing the dependence on unimportant information.

[0143] (b) Bilstm: This model consists of two LSTM layers, one for processing the forward sequence and the other for processing the reverse sequence, and predicts the future trajectory by considering the information of the previous and subsequent time points.

[0144] (c) Netraj: This model adds a defined sliding time context window to the seq2seq network with an encoder-decoder structure to capture the long-term and short-term temporal dependencies between trajectories, and finds the spatial correlation of features through a spatial attention layer, thereby predicting the trajectory.

[0145] (d) CAMD BiGRU: This model combines a novel restricted attention mechanism layer in the Bi-GRU module to extract the temporal attributes between data, and then uses Bi-GRU again to capture the periodic features in the data, thereby predicting future trajectory points.

[0146] It can be seen from Table 3 that the trajectory accuracy obtained by the method in this paper is the best. The error between the trajectory information planned by the GRU method and the true value is relatively large. This is because the network structure of this method is relatively simple, and its ability to extract the dependence relationship between sequences is limited, resulting in poor quality of the planned trajectory. The Bilstm method can better understand the time series attributes between trajectory points, and the prediction error is reduced by 11.5%. After the Netraj method pays attention to the important features in the time dimension and space dimension, the prediction accuracy of the model is significantly improved, and the accuracy is increased by 32.8%. After the CAMD BiGRU method captures the time attributes and periodicity between data, the prediction accuracy is increased by 27.6%. The method in this paper can strongly extract the correlation from the processed data through the improved model structure, and the output accuracy is increased by 72.3%. Thus, it shows that the method in this paper can well predict the trajectory data that conforms to human control habits.

[0147] Table 3 Comparison of RMSE and MAE of prediction parameters of comparison methods

[0148] ;

[0149] Since the changes in the vehicle's speed and angular velocity can intuitively reflect the smoothness of the planned trajectory, the standard deviations of the predicted vehicle speed and angular velocity in this paper are compared. It can be seen from Table 4 that the method in this paper has obvious advantages in terms of the stability of the planned trajectory, and compared with the Figure 6 fitting effect, it can be seen that the method in this paper can well learn human control behavior habits, and the planned trajectory has good smoothness, which can ensure the safety and smoothness of the vehicle.

[0150] Table 4 Comparison of SD of predicted speed and angular velocity

[0151] ;

[0152] In summary, in this embodiment, in order to plan a trajectory that conforms to human driving behavior when the vehicle turns in a traffic-line-free scenario, a human-like driving trajectory planning method is proposed. After collecting data on human driving of the vehicle through a self-designed multi-modal sensing system, the KPCA method is used to extract non-linear features from the vehicle kinematic information. Then, the relative distance between the current position of the vehicle and the obstacles on both sides of the road is calculated, and the relative distance is restricted within the range of the front and rear obstacles. On this basis, after combining with the vehicle kinematic state information in time series to screen out the data with poor quality, a human driving data set is constructed for the training and testing of the deep network. Then, cell state information is introduced into the LSTM network to weight and focus on the key data of long-term and short-term information, and the state output gate is used to alleviate the phenomenon of gradient disappearance during the training process, thereby improving the mapping accuracy of the network. After comparing with multiple methods, it can be seen from the Error, RMSE, and MAE indexes of the planned trajectory and the true value that the method in this embodiment can better learn human driving behavior, and the SD indexes of speed and angular velocity can show that the planned trajectory of the method in this embodiment has better smoothness. Therefore, the method in this embodiment can enable the vehicle to plan a safe and smooth trajectory route when turning in a traffic-line-free environment.

[0153] In the second aspect of this embodiment, a human-like trajectory planning system for automatic steering driving is provided. The system includes a collection and processing module, an analysis module, a model establishment module, and a prediction module.

[0154] The collection and processing module is used to collect and process human driving data in a curve scenario, and obtain the relative distance information between the vehicle and the obstacles on both sides of the road and the vehicle kinematic information.

[0155] The analysis module is used to extract non-linear features from the vehicle kinematic information based on the kernel principal component analysis method (KPCA).

[0156] The model establishment module is used to establish and train a multi-state cyclic prediction model; the multi-state cyclic prediction model is constructed based on the LSTM network, and includes cell state information to weight and focus on the key data of long-term and short-term information; the model includes a state output gate.

[0157] The prediction module is used to input the non-linear features, the relative distance information between the vehicle and the obstacles on both sides of the road, and the vehicle kinematic information into the multi-state cyclic prediction model to predict future driving trajectory points.

[0158] Furthermore, the curve scenario includes an entry phase, a turning phase, and an exit phase.

[0159] The vehicle kinematic information includes the vehicle driving speed and the vehicle driving angular velocity.

[0160] The acquisition and processing module further includes a component for calculating the distance positions between the vehicle and two obstacles on one side by using the relative position relationship between the current position of the vehicle and the obstacles on both sides.

[0161] Furthermore, the analysis module further includes a component for mapping the input spatial data into a feature space and then calculating the principal components in the feature space.

[0162] The model establishment module further includes a component for extracting the principal components through KPCA, using the principal components, the relative distance information between the vehicle and the obstacles on both sides of the road, and the vehicle kinematic information as the input of a deep network, and predicting future driving trajectory points by using the multi-state recurrent prediction model, wherein a state gate structure is added to transmit the memory state.

[0163] In addition, an embodiment of the present application also discloses an electronic device, which includes one or more processors and a memory. The memory is used to store one or more computer programs, and the computer programs are configured to be executed by the one or more processors. The programs include steps for executing the automatic steering driving humanoid trajectory planning method as described above.

[0164] In addition, an embodiment of the present application also provides a storage medium, which stores a computer program; the program is loaded and executed by a processor to implement the steps of the automatic steering driving humanoid trajectory planning method as described above.

[0165] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this embodiment can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0166] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed couplings or direct couplings or communication connections to each other can be indirect couplings or communication connections through some interfaces, devices or units, and can also be electrical, mechanical or other forms of connection.

[0167] The units described as separate components may or may not be physically separated. As a unit, those of ordinary skill in the art can realize that the units and algorithm steps described in combination with the embodiments disclosed in this embodiment can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0168] In addition, in each embodiment of the present invention, the functional units can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0169] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a grid device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0170] The specific embodiments described above further elaborate on the objective, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only for the specific embodiments of the present invention and is not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc., made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An automatic steering and driving humanoid trajectory planning method, characterized in that, The method includes: S101, collecting and processing human control data in a curved road scenario to obtain the relative distance information between the vehicle and the obstacles on both sides of the road and the vehicle kinematic information; S102, extracting non-linear features from the vehicle kinematic information based on the kernel principal component analysis (KPCA); S103, establishing and training a multi-state recurrent prediction model; the multi-state recurrent prediction model is constructed based on the LSTM network and includes cell state information to weight and focus on the key data of long-term and short-term information; the model includes a state output gate, and the multi-state recurrent prediction model includes: （10） Among them, is the input gate, is the sigmoid activation function, is the weight matrix, is the short-term memory output, is the bias vector, is the input information （11） Among them, is the sigmoid activation function, is the weight matrix, is the element-wise product, is the state cell, is the short-term memory output, is the bias vector, is the input information, is the forget gate; （12） Among them, is the forgetting gate, is the element-wise product, is the state cell, is the output of the forgetting gate; （13） Among them, is a non-linear transformation, is a state cell, is a weight matrix, is an activation function, is an element-wise product; （14） Among them, is a non-linear transformation, is a state cell, is an element-wise product, is an activation function, is a weight matrix, is a state transmission gate; （15） Among them, is the activation function, is the weight matrix, is the short-term memory output, is the bias vector, is the input information, is the output of the forgetting gate; （16） Among them, is the output of the forget gate, is the state transfer gate, is the input gate, is the element-wise product, is the output of the upper layer; （17） Among them, is the sigmoid activation function, is the weight matrix, is the short-term memory output, is the input information, is the output of the forgetting gate, is the bias vector, is the output gate; （18） Among them, is the activation function, is the state cell; S104, inputting the non-linear features, the relative distance information between the vehicle and the obstacles on both sides of the road, and the vehicle kinematic information into the multi-state recurrent prediction model to predict future driving trajectory points.

2. The automatic steering and driving humanoid trajectory planning method according to claim 1, wherein, The curved road scenario includes an entering curve stage, a turning stage, and an exiting curve stage; The vehicle kinematic information includes the vehicle driving speed and the vehicle driving angular velocity; The S101 includes: calculating the distance positions between the vehicle and two obstacles on one side by using the relative position relationship between the current position of the vehicle and the obstacles on both sides.

3. The automatic steering and driving humanoid trajectory planning method according to claim 2, characterized in that The S102 includes: mapping the input space data into the feature space and then calculating the main components in the feature space; The S103 includes: after extracting the main components through KPCA, using the main components, the relative distance information between the vehicle and the obstacles on both sides of the road, and the vehicle kinematic information as the input of the deep network, and using the multi-state recurrent prediction model to predict future driving trajectory points, where a state gate structure is added to transmit the memory state.

4. The automatic steering and driving humanoid trajectory planning method according to claim 1, wherein In the S103, when training the multi-state recurrent prediction model, the mean square error (MSE) is used as the loss function to train the network, and the formula is: ； Among them, N is the number of samples in the training dataset, is the actual trajectory coordinate value of the operator, is the predicted trajectory coordinate value, and the Adamw optimizer is used to adaptively optimize the network parameters.

5. An automatic steering and driving humanoid trajectory planning system, characterized in that, The system includes a collection and processing module, an analysis module, a model establishment module, and a prediction module, and is characterized in that: The collection and processing module is used for collecting and processing human control data in a curved road scenario to obtain the relative distance information between the vehicle and the obstacles on both sides of the road and the vehicle kinematic information; The analysis module is used for extracting non-linear features from the vehicle kinematic information based on the kernel principal component analysis (KPCA); The model establishment module is used for establishing and training a multi-state recurrent prediction model; the multi-state recurrent prediction model is constructed based on the LSTM network and includes cell state information to weight and focus on the key data of long-term and short-term information; the model includes a state output gate; The prediction module is used for inputting the non-linear features, the relative distance information between the vehicle and the obstacles on both sides of the road, and the vehicle kinematic information into the multi-state recurrent prediction model to predict future driving trajectory points.

6. The automatic steering and driving humanoid trajectory planning system according to claim 5, characterized in that, The curved road scenario includes an entering curve stage, a turning stage, and an exiting curve stage; The vehicle kinematic information includes the vehicle driving speed and the vehicle driving angular velocity; The collection and processing module further includes calculating the distance positions between the vehicle and two obstacles on one side by using the relative position relationship between the current position of the vehicle and the obstacles on both sides.

7. The automatic steering and driving humanoid trajectory planning system according to claim 6, wherein The analysis module further includes a component for mapping the input spatial data into a feature space and then calculating the principal components in the feature space; The model establishment module further includes a component for, after extracting the principal components through KPCA, using the principal components, the relative distance information between the vehicle and the obstacles on both sides of the road, and the vehicle kinematic information as the input of a deep network, and using the multi-state recurrent prediction model to predict future driving trajectory points, wherein a state gate structure is added to transmit the memory state.

8. An electronic device, the electronic device comprising: One or more processors, a memory for storing one or more computer programs; characterized in that the computer programs are configured to be executed by the one or more processors, and the programs include steps for executing the automatic steering and driving humanoid trajectory planning method according to any one of claims 1-5.

9. A storage medium, the storage medium storing a computer program; characterized in that, The program is loaded and executed by the processor to implement the steps of the automatic steering and driving humanoid trajectory planning method according to any one of claims 1-5.

Images