CN113184040B - Unmanned vehicle line-controlled steering control method and system based on steering intention of driver - Google Patents

Unmanned vehicle line-controlled steering control method and system based on steering intention of driver Download PDF

Info

Publication number
CN113184040B
CN113184040B CN202110620427.1A CN202110620427A CN113184040B CN 113184040 B CN113184040 B CN 113184040B CN 202110620427 A CN202110620427 A CN 202110620427A CN 113184040 B CN113184040 B CN 113184040B
Authority
CN
China
Prior art keywords
steering
driver
unmanned vehicle
cnn
lstm
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110620427.1A
Other languages
Chinese (zh)
Other versions
CN113184040A (en
Inventor
杨炜
彭永康
蔡建沅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Dragon Totem Technology Hefei Co ltd
Original Assignee
Changan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Changan University filed Critical Changan University
Priority to CN202110620427.1A priority Critical patent/CN113184040B/en
Publication of CN113184040A publication Critical patent/CN113184040A/en
Application granted granted Critical
Publication of CN113184040B publication Critical patent/CN113184040B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B62—LAND VEHICLES FOR TRAVELLING OTHERWISE THAN ON RAILS
    • B62D—MOTOR VEHICLES; TRAILERS
    • B62D1/00—Steering controls, i.e. means for initiating a change of direction of the vehicle
    • B62D1/24—Steering controls, i.e. means for initiating a change of direction of the vehicle not vehicle-mounted
    • B62D1/28—Steering controls, i.e. means for initiating a change of direction of the vehicle not vehicle-mounted non-mechanical, e.g. following a line or other known markers
    • B62D1/283—Steering controls, i.e. means for initiating a change of direction of the vehicle not vehicle-mounted non-mechanical, e.g. following a line or other known markers for unmanned vehicles
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B62—LAND VEHICLES FOR TRAVELLING OTHERWISE THAN ON RAILS
    • B62D—MOTOR VEHICLES; TRAILERS
    • B62D5/00—Power-assisted or power-driven steering
    • B62D5/04—Power-assisted or power-driven steering electrical, e.g. using an electric servo-motor connected to, or forming part of, the steering gear
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Combustion & Propulsion (AREA)
  • Transportation (AREA)
  • Mechanical Engineering (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Control Of Position, Course, Altitude, Or Attitude Of Moving Bodies (AREA)
  • Steering Control In Accordance With Driving Conditions (AREA)

Abstract

The invention relates to an unmanned vehicle wire-controlled steering control method and system based on steering intention of a driver, wherein a Probabilistic Neural Network (PNN) is used for carrying out nonlinear fitting on the driving track characteristics of a skilled driver, acquiring the steering intention information of the driver and carrying out analysis and comparison with other track fitting methods; then, how to predict characteristic parameters such as steering wheel corners with the characteristics of skilled drivers and the like on the intelligent driving automobile is researched, a driver model and an intelligent steering controller imitating the characteristics of the skilled drivers are established, a steer-by-wire system realizes the control of the corners, the torque and the like, the unmanned automobile has the operating characteristics which are highly similar to those of the skilled drivers, and the steering stability of the unmanned automobile is improved.

Description

Unmanned vehicle line-controlled steering control method and system based on steering intention of driver
Technical Field
The invention relates to the field of unmanned driving, in particular to an unmanned vehicle linear control steering control method and system based on steering intention of a driver.
Background
The steer-by-wire control eliminates a mechanical connecting device between a steering wheel and a steering wheel, thoroughly gets rid of the inherent defects of the traditional steering system, has the characteristics of safety, comfort, economy, good operation stability and the like, and is convenient to integrate with other systems and coordinate and control uniformly;
the existing unmanned vehicle mainly depends on an intelligent driving instrument which is mainly a computer system in the vehicle to realize unmanned driving. The vehicle-mounted sensor is used for sensing the surrounding environment of the vehicle, and the steering and the speed of the vehicle are controlled according to the road, the vehicle position and the obstacle information obtained by sensing, so that the vehicle can safely and reliably run on the road. And the traditional 'man-vehicle-road' closed-loop control mode is fundamentally changed, and an uncontrollable driver is requested from the closed-loop system, so that the efficiency and the safety of the traffic system are greatly improved.
The problems of the prior art are that the unmanned vehicle is unstable when steering, and the control level of a real skilled driver is not enough.
Disclosure of Invention
The invention aims to provide a method and a system for controlling steering by wire control of an unmanned vehicle based on steering intention of a driver, which are stable in steering and can simulate the control level of a real skilled driver;
an unmanned vehicle wire-controlled steering control method based on steering intention of a driver;
collecting steering test parameters under different steering conditions in a virtual driving simulation environment, and performing nonlinear fitting by using a Probabilistic Neural Network (PNN) according to the collected driving track characteristics of a driver to obtain steering intention information of the driver;
analyzing the steering operation characteristic memorability of the driver, and establishing a driver model based on a CNN-LSTM hybrid algorithm;
establishing a humanoid intelligent steering controller according to the acquired steering intention information of the driver;
acquiring the turning angle of the unmanned vehicle and the output of each power parameter based on a driver model and an intelligent steering controller;
further, the probabilistic neural network PNN performs nonlinear fitting according to the following method:
classifying the questions, wherein the classification questions are as follows: c = c 1 Or c = c 2 ；
Calculating prior probability:
h 1 ＝p(c 1 )，
h 2 ＝p(c 2 )
h 1 +h 2 ＝1
given an input vector x = [ x ] 1 ，x 2 ，…，x N ]Obtaining a set of observations;
the basis for classification is:
Figure BDA0003099308120000021
p(c 1 i x) is the occurrence of x, class c 1 The posterior probability of (d);
according to Bayes' formula, the posterior probability is equal to
Figure BDA0003099308120000022
Further, for classification problem c = c 1 Or c = c 2 The classification rule of (2) is adjusted, and the adjustment method is as follows:
defining an action alpha i To assign an input vector to c i Action of (a) ij For input vectors belonging to c j Take action of i The resulting loss is taken as action alpha i The expected risks of (c) are:
Figure BDA0003099308120000023
assuming that the loss for correct classification is zero, the input is assigned to c 1 The expected risk for a class is:
R(c 1 |x)＝λ 12 p(c 2 |x)
the bayesian decision rule becomes:
Figure BDA0003099308120000024
the probability density function is of the form:
Figure BDA0003099308120000031
c＝c i ，i＝arg min(R(c i |x))
f i is of class c i Is determined.
Further, the method for establishing the driver model by the CNN-LSTM hybrid algorithm is as follows:
the CNN-LSTM network model comprises two parts: a CNN structure in the Time Distributed wrapper and an LSTM structure outside the wrapper;
the CNN in the Time Distributed wrapper consists of two convolution layers, a pooling layer and a flattening layer, and the number of convolution kernels is changed to 16 i;
wherein the input format of the data of the input layer is (15 x 1);
adopting a ReLU function as an activation function, adopting maximum pooling in a pooling layer, wherein an external LSTM part consists of a hidden layer containing 100 neurons and a fully-connected output layer, and selecting the ReLU function as an activation function of the hidden layer, wherein the input of the ReLU function is the output of the last layer of the wrapper;
setting a validation _ data parameter in a fit () function, recording the loss of each training iteration in a training set and a testing set, and drawing a loss graph after the training and the testing are finished;
time Distributed is a wrapper for Kems1 that can apply a separate layer to each Time step of the input;
and then, dropout is adopted to perform fitting processing, the formula of which is transformed as follows,
dropout uses a pre-neural network:
Figure BDA0003099308120000032
Figure BDA0003099308120000033
dropout used neural network:
Figure BDA0003099308120000034
Figure BDA0003099308120000035
/>
Figure BDA0003099308120000036
Figure BDA0003099308120000037
w is the weight and b is the bias. After multiple times of experimental training, setting P in the CNN model and the LSTM model to be 0.1, and setting P in the CNN-LSTM model to be 0.2;
and finally, optimizing the learning rate by adopting an Adam optimization algorithm, wherein a derivation formula is as follows:
Figure BDA0003099308120000041
m t ＝μ*m t-1 +(1+μ)*g t
n t ＝v*n t-1 +(1+v)*g t 2
Figure BDA0003099308120000042
Figure BDA0003099308120000043
Figure BDA0003099308120000044
wherein g is t Is the gradient of the time step, m t ，n t For the first order estimate and the second order estimate of the gradient,
Figure BDA0003099308120000045
can be considered as to the desired g t ，g t 2 (ii) an estimate of (d); />
Figure BDA0003099308120000046
Then is to m t ，n t By a correction of Δ θ t The formula shows that the Adam algorithm has constraint capacity on the learning rate.
Further, a humanoid intelligent steering controller is established through a humanoid intelligent control HSIC;
the HSIC algorithm of the humanoid intelligent control is as follows:
Figure BDA0003099308120000047
wherein u is the control output, K p Is a scale factor, k is a suppression factor, e is an error,
Figure BDA0003099308120000048
as rate of change of error, e m，i Is the ith peak of the error;
a complex system, a complex process or a complex control task is decomposed into a plurality of simple subsystems, sub-processes or sub-tasks which can be executed independently according to the following formula:
G(E，T)＝F((g 1 (x 1 x 2 ，…，x n ，t)，g 2 (x 1 x 2 ，…，x n ，t)，…，g N (x 1 x 2 ，…，x n ，t)，E N ，T N ))
in the formula: g ∈ Σ N denotes the total task, x j J-th variable, g, representing the system 1 (x 1 x 2 ，…，x n T) denotes the ith subtask, E N E is Esper is N multiplied by N and is a planned subtask space characteristic set, T N E sigma N is a planned subtask time feature set; f (-) identifies the hierarchical structure of the high-order associated schema, and is a feature model based on time and space.
Furthermore, the steering test parameters under different steering conditions are collected, wherein the steering test parameters comprise vehicle speed, roll angle, lateral acceleration and parameter data of a steering wheel.
Unmanned vehicle drive-by-wire steering control system based on driver's intention of turning includes:
a data acquisition platform: collecting vehicle state information and steering test parameters under different steering working conditions;
a neural network module: receiving data acquired by a data acquisition platform, and performing nonlinear fitting on the driving track characteristics of a skilled driver by using a Probabilistic Neural Network (PNN) to acquire steering intention information of the driver; meanwhile, a CNN-LSTM hybrid algorithm is applied to establish a driver model;
the vehicle control unit is used for receiving the steering intention information of the driver sent by the neural network module, establishing an intelligent steering controller, obtaining the turning angle and each power output parameter of the unmanned vehicle by combining with the driver model, and outputting the turning angle and each power output parameter of the unmanned vehicle;
a human-simulated operation platform: and receiving the turning angle and each power output parameter of the unmanned vehicle sent by the vehicle control unit, sending the output parameters to a steering control system, and controlling the steering of the unmanned vehicle through the steering control system.
The invention has the beneficial effects that: nonlinear fitting is carried out on the driving track characteristics of a skilled driver by utilizing a probabilistic neural network PNN, steering intention information of the driver is obtained, and the driving track characteristics are analyzed and compared with other track fitting methods; then, how to predict characteristic parameters such as steering wheel corners with the characteristics of skilled drivers and the like on the intelligent driving automobile is researched, a driver model and an intelligent steering controller imitating the characteristics of the skilled drivers are established, a steer-by-wire system realizes the control of the corners, the torque and the like, the unmanned automobile has the operation characteristics which are highly similar to those of the skilled drivers, and the steering stability of the unmanned automobile is improved.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention;
FIG. 2 is a control block diagram of an embodiment of the present invention;
the invention is described in detail below with reference to the figures and examples.
Examples
1, shown in figure 1, a method for controlling the steering of an unmanned vehicle by wire based on the steering intention of a driver
S1, collecting steering test parameters under different steering conditions in a virtual driving simulation environment, and performing nonlinear fitting by using a Probabilistic Neural Network (PNN) according to the collected driving track characteristics of a driver to obtain steering intention information of the driver;
steering test parameters under different steering conditions are collected in a virtual driving simulation environment, and nonlinear fitting is carried out by using a Probabilistic Neural Network (PNN) according to the collected driving track characteristics of a driver to obtain steering intention information of the driver. The algorithm for applying the probabilistic neural network PNN to perform the nonlinear fitting is as follows:
classifying the questions, wherein the classification questions are as follows: c = c 1 Or c = c 2 ；
Calculating prior probability:
h 1 ＝p(c 1 )，
h 2 ＝p(c 2 )
h 1 +h 2 ＝1
given an input vector x = [ x ] 1 ，x 2 ，…，x N ]Obtaining a set of observations;
the basis for classification is:
Figure BDA0003099308120000061
p(c 1 i x) is the occurrence of x, class c 1 The posterior probability of (d);
according to Bayes' formula, the posterior probability is equal to
Figure BDA0003099308120000062
When a skilled driver drives a vehicle, an optimal driving path is planned according to the vehicle and environment information, and then a steering wheel, an accelerator pedal or a brake pedal is operated, so that the vehicle can stably drive along the planned path. Although the unmanned vehicle can perform path planning using advanced algorithms (such as bezier curves, spline curves, etc.) according to the relevant information, the generated path is very different from the path traveled by the actual driver, which deteriorates the comfort of the autonomous vehicle. By learning the driving path of a skilled driver under a specific steering working condition, the unmanned vehicle can control the vehicle to steer smoothly like the skilled driver.
The neural network can realize better fitting of a nonlinear curve of the track change according to the historical track data of the vehicle. The PNN can be regarded as a radial basis function neural network, and integrates density function estimation and Bayesian decision theory on the basis of the RBF network. Under certain conditions which are easy to meet, the discrimination boundary realized by PNN gradually approaches the Bayesian optimal discrimination surface. For simplicity of the analysis process.
When classifying and deciding, the input vector should be classified into the class with higher posterior probability. In practical application, the loss and the risk are also required to be considered, and c is 1 Class sample classification as c 2 Class (c) and 2 class sample subdivision into c 1 The losses caused by classes often vary widely, so that the classification rules need to be adjusted.
For classification problem c = c 1 Or c = c 2 The classification rule of (2) is adjusted, and the adjustment method is as follows:
defining an action alpha i To assign an input vector to c i Action of (a) ij For input vectors belonging to c j Take action of i The resulting loss is taken action a i The expected risks of (c) are:
Figure BDA0003099308120000071
assuming that the loss for correct classification is zero, the input is assigned to c 1 The expected risks for a class are:
R(c 1 |x)＝λ 12 p(c 2 |x)
the bayesian decision rule becomes:
Figure BDA0003099308120000072
the probability density function is of the form:
Figure BDA0003099308120000073
c＝c i ，i＝arg min(R(c i |x))
f i is of class c i Is determined.
S2, analyzing steering operation characteristics of a driver, and establishing a driver model based on a CNN-LSTM hybrid algorithm;
the method comprises the steps of analyzing steering characteristic parameters of a driver, such as steering wheel angle, torque, steering wheel speed signals and other steering operation characteristics. The adopted characteristic analysis specifically comprises the following steps: firstly, analyzing factors influencing the driving track under the steering working condition; the second is to analyze the interrelationship between steering wheel angle, torque and angular velocity. And establishing a driver model based on a CNN-LSTM hybrid algorithm. The algorithm is described in detail below: the CNN (convolutional neural network) algorithm can modularize data in the aspect of prediction, but because the correlation of space and time related to driving data is not considered, the data processed by the CNN is combined with an LSTM (long-short-time memory neural network) algorithm to be connected in a correlation manner, and then prediction is carried out in the LSTM, so that the correlation between the space positions when the vehicle drives is ensured to be considered, and the combination of the two algorithms not only improves the real-time performance, but also improves the accuracy of the spatial position relationship because the CNN prediction cannot be correlated.
A CNN-LSTM-based trajectory prediction model that has the benefit of supporting very long input sequences that can be read as block or subsequence information by the CNN model and then combined together by the LSTM model, requiring further splitting of each sample into more subsequences when using the CNN-LSTM combination model; the CNN model will read the information for each subsequence, while the LSTM aggregates the information from these subsequences and outputs a predicted value.
The method for establishing the driver model by the CNN-LSTM hybrid algorithm is as follows:
the CNN-LSTM network model comprises two parts: a CNN structure in the Time Distributed wrapper and an LSTM structure outside the wrapper;
the CNN in the Time Distributed wrapper consists of two convolution layers, a pooling layer and a flattening layer, and the number of convolution kernels is changed to 16 i;
wherein the input format of the data of the input layer is (15 x 1);
adopting a ReLU function as an activation function, adopting maximum pooling in a pooling layer, wherein an external LSTM part consists of a hidden layer containing 100 neurons and a fully-connected output layer, and selecting the ReLU function as an activation function of the hidden layer, wherein the input of the ReLU function is the output of the last layer of the wrapper;
setting a validation _ data parameter in a fit () function, recording the loss of each training iteration in a training set and a testing set, and drawing a loss graph after the training and the testing are finished;
time Distributed is a wrapper of Kems1 that can apply one, one independent layer to each Time step of the input;
dropout is then used for the fitting process, the formula of which is transformed as follows, dropout using the previous neural network:
Figure BDA0003099308120000081
Figure BDA0003099308120000082
dropout used neural network:
Figure BDA0003099308120000091
Figure BDA0003099308120000092
Figure BDA0003099308120000093
Figure BDA0003099308120000094
w is the weight and b is the bias. After a plurality of times of experimental training, setting P in the CNN model and the LSTM model to be 0.1, and setting P in the base CNN-LSTM model to be 0.2;
and finally, optimizing the learning rate by adopting an Adam optimization algorithm, wherein a derivation formula is as follows:
Figure BDA0003099308120000095
m t ＝μ*m t-1 +(1+μ)*g t
n t ＝v*n t-1 +(1+v)*g t 2
Figure BDA0003099308120000096
Figure BDA0003099308120000097
Figure BDA0003099308120000098
wherein g is t Is the gradient of the time step, m t ，n t For the first order estimate and the second order estimate of the gradient,
Figure BDA0003099308120000099
can be considered as to the desired g t ，g t 2 (ii) an estimate of (d); />
Figure BDA00030993081200000910
Then is to m t ，n t By correction of Δ θ t The formula shows that the Adam algorithm has constraint capacity on the learning rate.
S3, establishing a humanoid intelligent steering controller according to the acquired steering intention information of the driver; and acquiring the turning angle of the unmanned vehicle and the output of each power parameter based on the driver model and the intelligent steering controller.
Establishing a humanoid intelligent steering controller according to the acquired steering intention information of the driver, wherein the specific process of establishing the humanoid intelligent steering controller is as follows: the method takes 'humanoid' as a guiding idea, applies a research method of HSIC to decompose the complex task of humanoid steering control, deeply researches the technologies of characteristic parameter selection, control mode classification, control parameter setting and the like in the steering process of the HSIC theory, and further designs the multi-mode coordinated HSIC controller. Corresponding to the steering system control, it can be considered that: the perception input information processing is the measurement of the motion state of the automobile and the extraction of the characteristics. The internal model is a process of analyzing the working characteristics of the steering system, combining prior knowledge to determine a control mode and designing a human-simulated intelligent controller.
The HSIC algorithm of the humanoid intelligent control is as follows:
Figure BDA0003099308120000101
wherein u is the control output, K p Is a scale factor, k is a suppression factor, e is an error,
Figure BDA0003099308120000102
as rate of change of error, e m，i Is the ith peak of the error;
a complex system, a complex process or a complex control task is decomposed into a number of simple subsystems, sub-processes or sub-tasks that can be executed independently as follows:
G(E，T)＝F((g 1 (x 1 x 2 ，…，x n ，t)，g 2 (x 1 x 2 ，…，x n ，t)，…，g N (x 1 x 2 ，…，x n ，t)，E N ，T N ))
in the formula: g ∈ Σ N denotes the total task, x j J-th variable, g, representing the system 1 (x 1 x 2 ，…，x n T) denotes the ith subtask, E N E is Esper is N multiplied by N and is a planned subtask space characteristic set, T N E sigma N is a planned subtask time characteristic set; f (-) identifies the hierarchical structure of the high-order associated schema, and is a feature model based on time and space.
As shown in fig. 2, the wire-controlled steering control system for the unmanned vehicle based on the steering intention of the driver comprises:
the data acquisition platform can acquire the state information of the vehicle and required parameter data, such as vehicle speed, roll angle, lateral acceleration, parameter data of a steering wheel and the like;
the neural network module is used for carrying out nonlinear fitting on the driving track characteristics of a skilled driver by using a probabilistic neural network PNN, processing various data, and establishing a following driver model by using a CNN-LSTM hybrid algorithm;
the intelligent vehicle tracking system comprises a driver model, a vehicle speed control module and a vehicle speed control module, wherein the driver model is established on the basis that the ambient environment and the internal state parameters of the vehicle are known, and the control input of a system is designed based on a certain control theory, so that the intelligent vehicle can reach and finally track an expected track at an expected speed;
and the vehicle control unit is used for controlling the vehicle to run according to the state information of the vehicle during steering and the parameter data of the vehicle. And is shared with the humanoid operating platform and the steering control system through a vehicle-mounted local area network (CAN bus). Controlling the turning angle and the power parameter output of the unmanned vehicle according to the driver model and the steering controller;
the human-simulated operation platform is characterized in that kinematics researches the change rule of an object position along with time from the perspective of geometry; the human-simulated characteristic of the track under the action of human-simulated steering control can be truly reflected according to the track planned by the vehicle kinematic model, and the curve human-simulated control system model is practically applied to an intelligent driving automobile;
the steering control system and the unmanned vehicle linear control steering are a nonlinear time-varying complex system, have excellent control characteristics based on the steering intention of a driver, and are successfully applied to certain objects which are difficult to control in the industrial field by a humanoid intelligent theory. According to the parameter data of the driver during steering obtained by experiments, the steering controller is used for controlling the parameters of the unmanned vehicle during steering so as to achieve the unmanned vehicle linear control steering based on the steering intention of the driver. And signals are sent to the speed changer and the brake system, so that the unmanned vehicle can meet the steering requirement based on a skilled driver, and the stable change of the speed is realized.
A wire-controlled steering control system of an unmanned vehicle based on steering intention of a driver is characterized in that the working process of a steering strategy is as follows:
the vehicle control unit judges parameter values such as a tire corner, a vehicle speed and the like when the unmanned vehicle turns according to a turning position signal and a driving state signal when turning which are acquired by the data acquisition platform, the neural network module and the driver model;
the vehicle-mounted local area network of the humanoid operating platform acquires the vehicle speed, the tire rotation angle and the like of the vehicle controller when the vehicle controller processes shared steering;
the steering control system defines the maximum steering angle of the front wheel in a corresponding vehicle speed range according to the vehicle speed to limit the steering angle of the wheel, and controls the corresponding wheel steering angle when the wheel reaches a specific position in the steering process;
the human-simulated operation platform sends a signal to the transmission, so that the unmanned vehicle can meet the up-down shifting operation based on the change of the driver in the corresponding vehicle speed range;
the humanoid operation platform sends signals to the braking systems of all wheels, so that the braking of the wheels when the driver steers is achieved as much as possible, and the braking when the driver steers is carried out at a certain time point or position during steering.
When the unmanned vehicle without the steering controller is in a steering state, the steering and aligning speed of tires can be too slow or too fast, the too slow causes insensitive steering, and understeer can occur; too fast may cause larger yaw velocity of the vehicle body, deteriorate the steering quality of the vehicle, even cause instability during high-speed driving, and endanger the driving safety. The system can well solve the problems through the steering control system, so that the steering control quality of the unmanned vehicle is as close as possible to the steering intention and control of a driver.
The above embodiments are merely illustrative and should not be construed as limiting the scope of the invention, which is intended to be covered by the claims.

Claims (5)

1. A wire-controlled steering control method of an unmanned vehicle based on steering intention of a driver is characterized in that:
collecting steering test parameters under different steering conditions in a virtual driving simulation environment, and performing nonlinear fitting by using a Probabilistic Neural Network (PNN) according to the collected characteristics of the driving track of the driver to obtain steering intention information of the driver, wherein the nonlinear fitting of the Probabilistic Neural Network (PNN) is performed according to the following method:
classifying the questions, wherein the classification questions are as follows: c = c 1 Or c = c 2 ；
Calculating prior probability:
h 1 ＝p(c 1 )
h 2 ＝p(c 2 )
h 1 +h 2 ＝1
given an input vector x = [ x ] 1 ,x 2 ,...,x N ]Obtaining a set of observations;
the basis for classification is:
Figure FDA0004056750740000011
p(c 1 i x) is the occurrence of x, class c 1 A posterior probability of (d);
according to Bayes' formula, the posterior probability is equal to
Figure FDA0004056750740000012
；
Analyzing the steering operation characteristics of a driver, and establishing a driver model based on a CNN-LSTM hybrid algorithm; the method for establishing the driver model by the CNN-LSTM hybrid algorithm is as follows:
the CNN-LSTM network model comprises two parts: a CNN structure in the Time Distributed wrapper and an LSTM structure outside the wrapper;
the CNN in the Time Distributed wrapper consists of two convolution layers, a pooling layer and a flattening layer, and the number of convolution kernels is changed to 16 i;
wherein the input format of the data of the input layer is (15 x 1);
adopting a ReLU function as an activation function, adopting maximum pooling in a pooling layer, wherein an external LSTM part consists of a hidden layer containing 100 neurons and a fully-connected output layer, and selecting the ReLU function as an activation function of the hidden layer, wherein the input of the ReLU function is the output of the last layer of the wrapper;
setting a validation _ data parameter in a fit () function, recording the loss of each training iteration in a training set and a testing set, and drawing a loss graph after the training and the testing are finished;
time Distributed is a wrapper of Kems1 that is able to apply-a separate layer to each Time step of the input;
and then, dropout is adopted to perform overfitting processing, the formula of which is transformed as follows,
dropout uses a pre-neural network:
Figure FDA0004056750740000021
Figure FDA0004056750740000022
dropout used neural network:
Figure FDA0004056750740000023
/>
Figure FDA0004056750740000024
Figure FDA0004056750740000025
Figure FDA0004056750740000026
w is the weight and b is the bias; after a plurality of times of experimental training, setting P in the CNN model and the LSTM model to be 0.1, and setting P in the CNN-LSTM model to be 0.2;
and finally, optimizing the learning rate by adopting an Adam optimization algorithm, wherein a derivation formula is as follows:
g t ＝▽ θ J(θ t-1 )
m t ＝μ*m t-1 +(1+μ)*g t
n t ＝ν*n t-1 +(1+ν)*g t 2
Figure FDA0004056750740000027
Figure FDA0004056750740000028
Figure FDA0004056750740000029
wherein g is t Is the gradient of the time step, m t ，n t For the first and second order estimates of the gradient,
Figure FDA0004056750740000031
can be considered as corresponding to the desired g t ，g t 2 (ii) an estimate of (d); />
Figure FDA0004056750740000032
Then is to m t ，n t By correction of Δ θ t The formula can show that the Adam algorithm has constraint capacity on the learning rate;
establishing a humanoid intelligent steering controller according to the acquired steering intention information of the driver; and acquiring the turning angle of the unmanned vehicle and the output of each power parameter based on the driver model and the intelligent steering controller.
2. The method for controlling steer-by-wire of an unmanned vehicle based on driver's steering intention according to claim 1, wherein c = c is a classification problem 1 Or c = c 2 The classification rule of (2) is adjusted, and the adjustment method is as follows:
defining an action alpha i To assign an input vector to c i Action of (a) ij For input vectors belonging to c j Take action of i The resulting loss is taken as action alpha i The expected risks of (c) are:
Figure FDA0004056750740000033
assuming that the loss for correct classification is zero, the input is assigned to c 1 The expected risks for a class are:
R(c 1 |x)＝λ 12 p(c 2 |x)
the bayesian decision rule becomes:
Figure FDA0004056750740000034
the probability density function is of the form:
Figure FDA0004056750740000035
c＝c i ,i＝argmin(R(c i |x))
f i is of class c i Is determined.
3. The method for controlling the wire-controlled steering of the unmanned vehicle based on the steering intention of the driver according to claim 1, wherein the humanoid intelligent steering controller is established by a humanoid intelligent control HSIC;
the HSIC algorithm of the humanoid intelligent control is as follows:
Figure FDA0004056750740000041
wherein u is the control output, K p Is a scale factor, k is a suppression factor, e is an error,
Figure FDA0004056750740000042
as rate of change of error, e m,i Is the ith peak of the error;
a complex system, a complex process or a complex control task is decomposed into a number of simple subsystems, sub-processes or sub-tasks that can be executed independently as follows:
G(E,T)＝F((g 1 (x 1 x 2 ,…,x n ,t),g 2 (x 1 x 2 ,…,x n ,t),…,g N (x 1 x 2 ,…,x n ,t),E N ,T N ))
in the formula: g ∈ Σ N denotes the total task, x j J-th variable, g, representing the system 1 (x 1 x 2 ,…,x n T) denotes the ith subtask, E N E is Esper is N multiplied by N and is a planned subtask space characteristic set, T N E sigma N is a planned subtask time characteristic set; f (-) identifies the hierarchical structure of the high-order associated schema, and is a characteristic model based on time and space.
4. The method for controlling the steering of the unmanned vehicle by wire based on the steering intention of the driver as claimed in claim 1, wherein the steering test parameters under different steering conditions are collected, and the steering test parameters comprise vehicle speed, roll angle, lateral acceleration and steering wheel parameter data.
5. A control system to which the method for controlling steer-by-wire of an unmanned vehicle based on driver's steering intention according to claim 1 is applied, characterized by comprising:
a data acquisition platform: collecting vehicle state information and steering test parameters under different steering working conditions;
a neural network module: receiving data acquired by a data acquisition platform, and performing nonlinear fitting on the driving track characteristics of a skilled driver by using a Probabilistic Neural Network (PNN) to acquire steering intention information of the driver; meanwhile, a CNN-LSTM hybrid algorithm is applied to establish a driver model;
the vehicle control unit is used for receiving the steering intention information of the driver sent by the neural network module, establishing an intelligent steering controller, obtaining power output parameters of the corner of the unmanned vehicle by combining with the driver model, and outputting the corner and the power output parameters of the unmanned vehicle;
a human-simulated operation platform: and receiving power output parameters of each corner of the unmanned vehicle sent by the vehicle control unit, sending the output parameters to a steering control system, and controlling the steering of the unmanned vehicle through the steering control system.
CN202110620427.1A 2021-06-03 2021-06-03 Unmanned vehicle line-controlled steering control method and system based on steering intention of driver Active CN113184040B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110620427.1A CN113184040B (en) 2021-06-03 2021-06-03 Unmanned vehicle line-controlled steering control method and system based on steering intention of driver

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110620427.1A CN113184040B (en) 2021-06-03 2021-06-03 Unmanned vehicle line-controlled steering control method and system based on steering intention of driver

Publications (2)

Publication Number Publication Date
CN113184040A CN113184040A (en) 2021-07-30
CN113184040B true CN113184040B (en) 2023-04-07

Family

ID=76975873

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110620427.1A Active CN113184040B (en) 2021-06-03 2021-06-03 Unmanned vehicle line-controlled steering control method and system based on steering intention of driver

Country Status (1)

Country Link
CN (1) CN113184040B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114454951B (en) * 2021-12-30 2022-12-06 南京航空航天大学 Dual-motor steer-by-wire system and convolutional neural network fault-tolerant control method thereof
CN116776204B (en) * 2023-06-26 2024-02-23 清华大学 Driver risk sensitivity differentiation characterization method, device, equipment and medium

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112737930A (en) * 2020-12-18 2021-04-30 国网辽宁省电力有限公司 Intelligent data communication network shutdown delay processing system and delay processing method

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107813820A (en) * 2017-10-13 2018-03-20 江苏大学 A kind of unmanned vehicle lane-change paths planning method for imitating outstanding driver
US11467579B2 (en) * 2019-02-06 2022-10-11 Perceptive Automata, Inc. Probabilistic neural network for predicting hidden context of traffic entities for autonomous vehicles
CN111332362B (en) * 2020-03-10 2021-06-25 吉林大学 Intelligent steer-by-wire control method integrating individual character of driver
CN112622886B (en) * 2020-12-20 2022-02-15 东南大学 Anti-collision early warning method for heavy operation vehicle comprehensively considering front and rear obstacles
CN112677977B (en) * 2020-12-28 2022-08-05 科大讯飞股份有限公司 Driving state identification method and device, electronic equipment and steering lamp control method

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112737930A (en) * 2020-12-18 2021-04-30 国网辽宁省电力有限公司 Intelligent data communication network shutdown delay processing system and delay processing method

Also Published As

Publication number Publication date
CN113184040A (en) 2021-07-30

Similar Documents

Publication Publication Date Title
CN110568760B (en) Parameterized learning decision control system and method suitable for lane changing and lane keeping
CN109885883B (en) Unmanned vehicle transverse motion control method based on GK clustering algorithm model prediction
CN111775949B (en) Personalized driver steering behavior auxiliary method of man-machine co-driving control system
Plöchl et al. Driver models in automobile dynamics application
CN112622903B (en) Longitudinal and transverse control method for autonomous vehicle in vehicle following driving environment
CN111332362B (en) Intelligent steer-by-wire control method integrating individual character of driver
CN113184040B (en) Unmanned vehicle line-controlled steering control method and system based on steering intention of driver
CN104462716B (en) A kind of the brain-computer interface parameter and kinetic parameter design method of the brain control vehicle based on people's bus or train route model
CN112141101A (en) Method and system for pre-aiming safety path based on CNN and LSTM
CN114379583A (en) Automatic driving vehicle trajectory tracking system and method based on neural network dynamics model
CN115303289A (en) Vehicle dynamics model based on depth Gaussian, training method, intelligent vehicle trajectory tracking control method and terminal equipment
CN114942642A (en) Unmanned automobile track planning method
Hermansdorfer et al. A concept for estimation and prediction of the tire-road friction potential for an autonomous racecar
Chen et al. An improved IOHMM-based stochastic driver lane-changing model
Tian et al. Personalized lane change planning and control by imitation learning from drivers
Sousa et al. Nonlinear tire model approximation using machine learning for efficient model predictive control
CN114906128A (en) Automatic parking motion planning method based on MCTS algorithm
CN115525054B (en) Method and system for controlling tracking of edge path of unmanned sweeper in large industrial park
CN114323698B (en) Real vehicle experiment platform testing method for man-machine co-driving intelligent vehicle
CN114995426A (en) Unmanned vehicle trajectory tracking control method and system based on neural network dynamic model and vehicle-mounted control equipment
CN111857112B (en) Automobile local path planning method and electronic equipment
Chang et al. A quantum PSO algorithm for feedback control of semi-autonomous driver assistance systems
CN110977965A (en) Robot, method of controlling the same, and computer storage medium
Fényes Application of data-driven methods for improving the peformances of lateral vehicle control systems
Wang et al. Design and evaluation of a driver intent based mobile control interface for ground vehicles

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20240116

Address after: 230000 floor 1, building 2, phase I, e-commerce Park, Jinggang Road, Shushan Economic Development Zone, Hefei City, Anhui Province

Patentee after: Dragon totem Technology (Hefei) Co.,Ltd.

Address before: 710000 central section of south two ring road, Yanta District, Xi'an, Shaanxi

Patentee before: CHANG'AN University