CN109739218A - It is a kind of that outstanding driver's lane-change method for establishing model is imitated based on GRU network - Google Patents
It is a kind of that outstanding driver's lane-change method for establishing model is imitated based on GRU network Download PDFInfo
- Publication number
- CN109739218A CN109739218A CN201811579748.6A CN201811579748A CN109739218A CN 109739218 A CN109739218 A CN 109739218A CN 201811579748 A CN201811579748 A CN 201811579748A CN 109739218 A CN109739218 A CN 109739218A
- Authority
- CN
- China
- Prior art keywords
- lane
- change
- gru
- gru network
- driver
- 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.)
- Pending
Links
Abstract
The invention discloses a kind of to imitate outstanding driver's lane-change method for establishing model based on GRU network, belongs to intelligent automobile automatic Pilot field.This method comprises: the real train test under lane-change operating condition, acquires outstanding driver's steering characteristic parameter, Vehicle dynamic parameters and trajectory parameters, lane-change behavior data set is formed;Study is trained to lane-change behavioral data collection using GRU network, obtains imitating outstanding driver's lane-change model based on GRU network.The present invention nonlinear fitting ability powerful on long-term sequence using GRU network, realize one be simple and efficient imitate outstanding driver's lane-change model, it can guarantee on the basis of Fast Learning, further increase the accuracy of study, preferably accomplish to imitate the lane-change behavior of outstanding driver, there is certain reference in future for the field and other forecasting problems for being related to pilot model.
Description
Technical field
The invention belongs to intelligent automobile automatic Pilot fields, and in particular to a kind of to imitate outstanding driver based on GRU network
Lane-change method for establishing model.
Background technique
In recent years, with computer, internet, communication and navigation, automatic control, artificial intelligence, machine vision, accurate sensing
The rapid fusion of the high new technologies and advanced automobile technology such as device, high-precision map, intelligent automobile (or it is pilotless automobile, automatic
Driving) have become the research hotspot of world car engineering field and the new power of automobile industry growth.According to domestic and international power
The prediction of prestige media, the scientific and technical innovation in auto industry intelligence field will account for the 90% of entire auto industry.The intelligence of automobile
It is mainly reflected in and manual operation is substituted with automatic Pilot, the behavior of automobile and operating status controllably can be predicted, and can make up people
The deficiency of class sensory faculty mitigates driver behavior intensity, traffic accident caused by human factor is eliminated, according to real-time road condition information
It plans trip route, finally realizes road traffic " zero injures and deaths, zero congestion ".Therefore, intelligent automobile is safe and efficient, energy-efficient
Next-generation automobile, research intelligent automobile have particularly important meaning, it has also become the focus of Global Auto industrial circle.
On the other hand, pilot model research is the critical issue in intelligent automobile design process, more common at present
Model is the pilot model based on " preview follower " theory that Guo Konghui academician proposes, but its theory does not account for really driving
The steering habit and steering characteristic for the person of sailing have certain deficiency in terms of the manipulation level for simulating true driver.Based on mind
The more driving behavior of approaching to reality driver to a certain extent of pilot model through networked control theory, but its algorithm
Pace of learning is often relatively slow, while lacking good global optimizing ability, causes final output precision not high.Current rarer text
It offers from Human Simulating Intelligent Control field and intelligent vehicle course changing control is studied.
Summary of the invention
To solve the shortcomings of the prior art, the present invention provides a kind of, and the outstanding driver that imitates based on GRU network is changed
Road method for establishing model specifically develops a kind of intelligent automobile lane-change pilot model, from the lane-change behaviour for analyzing outstanding driver
Vertical dynamic behavior is started with, and is established intelligent automobile under lane-change operating condition and is turned to pilot model, really realizes apery lane-change, realize intelligence
The good apery lane-change handling safety of energy automobile and comfort, improve the intelligence degree of existing intelligent automobile.
The present invention is achieved through the following technical solutions above-mentioned technical purpose.
It is a kind of that outstanding driver's lane-change method for establishing model is imitated based on GRU network, comprising the following steps:
S1 selects the driving training school of different driving ages, different sexes to train as driver, and driver is with stabilization
Speed carries out the real train test under lane-change operating condition, acquires outstanding driver's steering characteristic parameter, Vehicle dynamic parameters and track
Parameter after track of vehicle parameter is carried out geodetic coordinates conversion respectively, and road boundary and makes the difference, calculates the laterally offset of vehicle
Amount, i.e. vehicle and road boundary distance Dy;The disposal of gentle filter is carried out to steering characteristic parameter, Vehicle dynamic parameters, is obtained
To raw data set, laterally offset amount and raw data set constitute outstanding driver's lane-change behavior data set;The driver turns
It include angular signal, dtc signal and tarnsition velocity signal to characteristic parameter, the Vehicle dynamic parameters include yaw angle speed
It spends signal, roll angle signal and lateral acceleration signal, the track of vehicle parameter is differential GPS signal;
S2, using outstanding driver's lane-change behavior data set as sample, using the angular signal of subsequent time as label, structure
Build training sample data and test sample data;Data set xt={ Dy, vel, Ay, Steer Ang, Steer Vel, Msteer },
The label is yt={ SteerAng }, wherein DyFor the lateral distance of vehicle distances road boundary, vel is vehicle current vehicle speed,
AyFor vehicle lateral acceleration, SteerAng is steering wheel angle, and SteerVel is steering wheel angle speed, and Msteer is direction
Disk torque;
S3 constructs GRU network model;
The GRU network model refers to the lane-change behavior ginseng according to n period before t moment (t-n+1, t-n+2 ..., t)
The lane-change behaviors at number prediction t+1 moment, the specific structure of the GRU network model are input layer, hidden layer and output layer, wherein
Input layer is made of 6 units, and GRU hidden layer is made of 10 GRU units, and output layer is made of a unit, and input layer with
It is connected entirely between GRU hidden layer, GRU hidden layer and output layer, the activation loss function of the GRU network model selects gradient descent method
In adaptive moments estimation optimization algorithm, loss function be square difference function;The GRU network model includes input layer, GRU hidden
Layer and output layer, connect entirely between input layer and GRU hidden layer, GRU hidden layer and output layer;It is wrapped in the structure of the GRU hidden layer
It includes and updates door zt, resetting door rt, hidden layer current state ht, htBy previous moment state ht-1With the candidate for being added to current state
Value htCollective effect simultaneously updates;
S4 obtains imitating outstanding driver's lane-change based on GRU network using training sample data training GRU network model
Model;Trained step includes: that the data in data set are normalized, and initializes each layer weight of GRU network, takes certain
One lot data is as this training set, by this input data xtIt is input to network, is calculated by the update of GRU network
Corresponding output yt, calculate the output y of GRU networktWith predicted value ytError, utilize the adaptive moments estimation in gradient descent method
Optimization algorithm updates GRU network weight, completes model training;
Outstanding driver's steering characteristic parameter, Vehicle dynamic parameters and trajectory parameters are input to lane-change model, obtained by S5
It obtains outstanding driver's lane-change behavior of subsequent time: steering characteristic parameter, Vehicle dynamic parameters and the track before t moment is joined
Number is input to lane-change model, predicts the angular signal at t+1 moment, the i.e. lane-change behavior of t+1 moment outstanding driver, then rolls
It is dynamic to update.
Compared with prior art, the beneficial effects of the present invention are:
1. outstanding driver's lane-change behavior data used in the present invention ensure that imitate outstanding driver's lane-change model pre-
The driver's lane-change behavior measured is safe, comfortable and reliable.
Outstanding driver's lane-change method for establishing model, this method benefit are imitated based on GRU network 2. the present invention provides a kind of
With the update door and resetting door in GRU, the information in long-term sequence can be saved, and will not at any time and remove or because with
Predict uncorrelated and remove, the powerful nonlinear fitting ability on long-term sequence, realize one be simple and efficient imitate it is excellent
Elegant driver's lane-change model.
3. update door and resetting door in the GRU network that the present invention designs, solve the problems, such as the gradient disappearance of standard RNN,
The two gate vectors determine last output, it is ensured that on the basis of Fast Learning, further increase the standard of study
True property preferably accomplishes the lane-change behavior for imitating outstanding driver, following to be related to pilot model with other for the field
Forecasting problem has certain reference.
Detailed description of the invention
Fig. 1 is lane-change schematic diagram;
Fig. 2 is a kind of flow chart for imitating outstanding driver's lane-change model based on GRU network;
Fig. 3 is the structure chart of GRU module;
Fig. 4 is the modeling process schematic diagram of GRU network;
Fig. 5 is a kind of prediction effect figure for imitating outstanding driver's lane-change model based on GRU network;
Fig. 6 is a kind of prediction deviation figure for imitating outstanding driver's lane-change model based on GRU network.
Specific embodiment
Present invention will be further explained with reference to the attached drawings and specific examples, but protection scope of the present invention is simultaneously
It is without being limited thereto.It should be noted that the combination of technical characteristic described in following embodiments or technical characteristic should not be recognized
For be it is isolated, they can be combined with each other to reaching superior technique effect.
It is a kind of that outstanding driver's lane-change method for establishing model is imitated based on GRU network, comprising the following steps:
S1, data acquisition and preparation: it as shown in Figure 1, carrying out the real train test under lane-change operating condition, acquires outstanding driver and turns
To characteristic parameter, Vehicle dynamic parameters and trajectory parameters;Driver's steering characteristic parameter include angular signal, dtc signal and
Tarnsition velocity signal, Vehicle dynamic parameters include yaw rate signal, roll angle signal and lateral acceleration signal, vehicle
Trajectory parameters include differential GPS signal;Geodetic coordinates conversion is carried out to trajectory parameters, calculates the laterally offset amount of vehicle;It is right
Collected steering characteristic parameter, Vehicle dynamic parameters carry out the disposal of gentle filter, obtain raw data set, laterally offset amount
Outstanding driver's lane-change behavior data set is collectively formed with raw data set;Specifically:
S1.1 selects outstanding driver: coach's long campaigns driving instruction and driving practice due to driving training school
Work, therefore its driving ability is relatively good, the different driving trainings for driving ages, different sexes of the present embodiment selection 5
It instructs school coach and is used as outstanding driver;
S1.2 devises the typical condition of lane-change, the speed of vehicle is determining during test in actual vehicle trials
, including 30KM/h and 40km/h, i.e., lane-change test is carried out in the case where driver maintains the two stabilizing speeds;
S1.3 after track of vehicle parameter is carried out geodetic coordinates conversion respectively, and road boundary and makes the difference, calculates vehicle
Laterally offset amount, i.e. vehicle and road boundary distance Dy;
The present embodiment selects the Bath filter of different rank to handle experimental data: wherein yaw velocity, angle of heel
Signal and lateral acceleration signal select second order Butterworth filter, angular signal, dtc signal and the choosing of tarnsition velocity signal
Select single order Butterworth filter.
S2, data set obtain: using outstanding driver's lane-change behavior data set as sample, by the angular signal of subsequent time
As label, the data set of building includes sample data 3400 altogether, and therein 90% is used as training sample, and remaining 10% makees
For test sample, i.e., include 3060 lane-change data in training sample, includes 340 lane-change data in test sample, respectively will
These data are as the training sample data and test sample data for imitating outstanding driver's lane-change model based on GRU network;Institute
Stating data set includes xt={ Dy, vel, Ay, SteerAng, SteerVel, Msteer }, the label is yt={ SteerAng },
Wherein DyFor the lateral distance of vehicle distances road boundary, vel is vehicle current vehicle speed, AyFor vehicle lateral acceleration,
SteerAng is steering wheel angle, and SteerVel is steering wheel angle speed, and Msteer is steering wheel torque;It is adopted in the present embodiment
The lane-change execution duration is 9s, time interval 0.05s.
S3, as shown in figure 4, GRU network model constructs: the GRU network model includes three layers, respectively input layer, GRU
Hidden layer and output layer;
S3.1, network inputs and output:
GRU network model refers to pre- according to the lane-change behavior parameter of n period before t moment (t-n+1, t-n+2 ..., t)
The lane-change behavior at t+1 moment is surveyed, the historical data of n period may make up following matrix before n times lane-change t moment in test:
In above formula, xtThe lane-change behavior parameter for indicating t moment, in the present embodiment, it is assumed that n=10;
The lane-change behavior data for needing to predict, which can be used to lower matrix, to be indicated:
S3.2, network structure determine:
The specific structure of the GRU network model is input layer, GRU hidden layer and output layer, and wherein input layer is by 6 units
Composition, GRU hidden layer are made of 10 GRU units, and output layer is made of a unit;Input layer and GRU hidden layer, GRU hidden layer with
Connection type between output layer is full connection.
The structure of GRU hidden layer is as shown in Figure 3:
Z in the structuretIt indicates to update door, rtReset door, htIndicate hidden layer current state, xtIndicate input, htIt indicates to be added
To the candidate value of current state;Update door ztCalculation formula it is as follows:
zt=σ (Uzxt+Wzht-1) (1)
In formula, σ is to indicate sigmoid function, is the activation primitive of gate cell, UzFor GRU network inputs and htBetween power
Weight matrix, WzFor the GRU network output of last moment and htBetween weight matrix, ht-1For historic state information;
Calculated ztValue indicates that the status information at previous moment is transmitted to the scale of current state between 0 to 1,
Ratio in other words, this rate value are codetermined by the state and current input value X of previous moment t-1.
The activation primitive of GRU network selection in the present embodiment is tanh, and input value X and the state value at previous moment are total
New state can be generated using activation primitive with by linear transformation, only this state value and the shape without containing history
State information, the state value that only this is codetermined by input and preceding state in t, this new status information are known as waiting
Select state value ht, htCalculating it is as follows:
In formula, UhFor GRU network inputs and update the weight matrix between door, WhFor last moment GRU network output with
The weight matrix between door is updated, ο indicates that corresponding element is multiplied;
Resetting door r in above formulatCalculating it is as follows:
rt=σ (Urxt+Wrht-1) (3)
In formula, UrFor the weight matrix between GRU network inputs and reset gate, WrFor last moment GRU network output with
Weight matrix between reset gate;
Resetting door meaning indicate the previous moment status information how many participate in the candidate for being added to current state
Value htIf r=0, it is added to the candidate value h of current statetIt is obtained by input data X, if r=1, is indicated previous completely
The status information at a moment completely participates in the generation of new state-h;
It finally include historic state information ht-1With the current state h of new state-h (candidate state)tCalculating it is as follows:
Current state htBy previous moment state ht-1With the candidate value h for being added to current statetCollective effect simultaneously updates,
The ratio of update is by update door ztIt determines, rtIt determines in the candidate value h for being added to current statetWhen previous moment state ht-1's
Degree of participation.
S3.3, training parameter:
Adaptive moments estimation optimization algorithm in the activation loss function selection gradient descent method of the GRU network model,
Loss function is square difference function: learning rate 0.006, time step 20, batch size 60.
S4 obtains imitating outstanding pilot model: using training sample training GRU network model, obtaining based on GRU network
Imitate outstanding driver's lane-change model;
As shown in Fig. 2, the GRU training step specifically:
Data in training sample are normalized S4.1, and are divided into n batch, and each batch size is 60;
The input data x of t moment GRU networkt, label is the angular signal y of subsequent timet;
S4.2 initializes each layer weight of GRU network, the weight W=0 of GRU hidden layer, the weight U=0, i=0 of input layer;
S4.3 takes a certain lot data as this training set, t=0;
S4.4, by this input data xtIt is input to GRU network, corresponding output y is calculated by the update of GRU networkt;
S4.5 calculates the output y of GRU networktWith predicted value ytError;
S4.6, judges whether the training of this batch is completed, if so, otherwise t=t+1, and return to S4.1.4;
S4.7 updates GRU network weight using the adaptive moments estimation optimization algorithm in gradient descent method;
S4.8, judges whether all data are completed to train, if so, executing S4.1.9, otherwise returns to S4.1.3;
Whether S4.9, i=i+1, training of judgement number reach requirement, if returning to S4.1.2 without if, otherwise complete model
Training, obtains imitating outstanding driver's lane-change model based on GRU network.
S5, output imitate outstanding driver and turn to behavior: by outstanding driver's steering characteristic parameter, Vehicle dynamic parameters and
Trajectory parameters, which are input to, imitates outstanding driver's lane-change model based on GRU network, obtains outstanding driver's lane-change of subsequent time
Behavior: by the data input lane-change model before t moment, the angular signal at t+1 moment, i.e. t+1 moment outstanding driving are predicted
Then the lane-change behavior of member is rolled and is updated;Prediction result is as it can be seen in figures 5 and 6, the model loss function after training finally reaches
7.2330775e-05, root-mean-square error 0.12750373, index of conformity 0.99989593, prediction deviation are 2.65 ° maximum ', knot
Fruit shows that prediction effect is preferable.
Although the present invention has been presented for some embodiments, it will be appreciated by those of skill in the art that not departing from
In the case where spirit of that invention, the embodiment of the present invention can be changed.Above-described embodiment is only exemplary, should not be with
Restriction of the embodiment of the present invention as interest field of the present invention.
Claims (9)
1. a kind of imitate outstanding driver's lane-change method for establishing model based on GRU network, which comprises the following steps:
S1 carries out the real train test under lane-change operating condition, acquires outstanding driver's steering characteristic parameter, Vehicle dynamic parameters and rail
Mark parameter is calculated the laterally offset amount of vehicle by trajectory parameters;Steering characteristic parameter, Vehicle dynamic parameters are carried out smooth
Filtering processing, obtains raw data set, and laterally offset amount and raw data set constitute outstanding driver's lane-change behavior data set;
S2, using outstanding driver's lane-change behavior data set as sample, using the angular signal of subsequent time as label, building instruction
Practice sample data and test sample data;
S3 constructs GRU network model;
S4 obtains imitating outstanding driver's lane-change model based on GRU network using training sample data training GRU network model;
Outstanding driver's steering characteristic parameter, Vehicle dynamic parameters and trajectory parameters are input to lane-change model, under acquisition by S5
Outstanding driver's lane-change behavior at one moment.
2. according to claim 1 imitate outstanding driver's lane-change method for establishing model based on GRU network, feature exists
In driver's steering characteristic parameter includes angular signal, dtc signal and tarnsition velocity signal, the dynamics of vehicle ginseng
Number includes yaw rate signal, rolls angle signal and lateral acceleration signal, and the track of vehicle parameter is differential GPS letter
Number.
3. according to claim 1 imitate outstanding driver's lane-change method for establishing model based on GRU network, feature exists
In, the GRU network model include input layer, GRU hidden layer and output layer, input layer and GRU hidden layer, GRU hidden layer and output
It is connected entirely between layer.
4. according to claim 1 or 2 imitate outstanding driver's lane-change method for establishing model, feature based on GRU network
It is, the S1 specifically: select the different driving ages, the driving training school of different sexes trains as driver, driving
Member carries out lane-change test with stabilizing speed, after track of vehicle parameter is carried out geodetic coordinates conversion respectively, and road boundary and does
Difference calculates the laterally offset amount of vehicle, i.e. vehicle and road boundary distance Dy。
5. according to claim 4 imitate outstanding driver's lane-change method for establishing model based on GRU network, feature exists
In data set x in the S2t={ Dy, vel, Ay, SteerAng, SteerVel, Msteer }, the label is yt=
{ SteerAng }, wherein DyFor the lateral distance of vehicle distances road boundary, vel is vehicle current vehicle speed, AyFor lateral direction of car plus
Speed, SteerAng are steering wheel angle, and SteerVel is steering wheel angle speed, and Msteer is steering wheel torque.
6. according to claim 1 or 3 imitate outstanding driver's lane-change method for establishing model, feature based on GRU network
It is, the GRU network model refers to the lane-change behavior parameter according to n period (t-n+1, t-n+2 ..., t) before t moment
Predict the lane-change behavior at t+1 moment, the specific structure of the GRU network model is input layer, hidden layer and output layer, wherein defeated
Enter layer to be made of 6 units, GRU hidden layer is made of 10 GRU units, and output layer is made of a unit, and input layer and GRU
It is connected entirely between hidden layer, GRU hidden layer and output layer, in the activation loss function selection gradient descent method of the GRU network model
Adaptive moments estimation optimization algorithm, loss function be square difference function.
7. according to claim 6 imitate outstanding driver's lane-change method for establishing model based on GRU network, feature exists
In including updating door z in the structure of the GRU hidden layert, resetting door rt, hidden layer current state ht, htBy previous moment state
ht-1With the candidate value h for being added to current statetCollective effect simultaneously updates.
8. according to claim 1 imitate outstanding driver's lane-change method for establishing model based on GRU network, feature exists
In the step of training includes: that the data in data set are normalized in the S4, each layer power of initialization GRU network
Weight, takes a certain lot data as this training set, by this input data xtIt is input to network, is updated by GRU network
Corresponding output y is calculatedt, calculate the output y of GRU networktWith predicted value ytError, using adaptive in gradient descent method
It answers moments estimation optimization algorithm to update GRU network weight, completes model training.
9. according to claim 1 or claim 7 imitate outstanding driver's lane-change method for establishing model, feature based on GRU network
It is, the S5 specifically: steering characteristic parameter, Vehicle dynamic parameters and the trajectory parameters before t moment are input to lane-change
Model predicts the angular signal at t+1 moment, the i.e. lane-change behavior of t+1 moment outstanding driver, then rolls and updates.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811579748.6A CN109739218A (en) | 2018-12-24 | 2018-12-24 | It is a kind of that outstanding driver's lane-change method for establishing model is imitated based on GRU network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811579748.6A CN109739218A (en) | 2018-12-24 | 2018-12-24 | It is a kind of that outstanding driver's lane-change method for establishing model is imitated based on GRU network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109739218A true CN109739218A (en) | 2019-05-10 |
Family
ID=66359626
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811579748.6A Pending CN109739218A (en) | 2018-12-24 | 2018-12-24 | It is a kind of that outstanding driver's lane-change method for establishing model is imitated based on GRU network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109739218A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110346517A (en) * | 2019-07-25 | 2019-10-18 | 中南大学 | A kind of smart city industrial air pollution visualization method for early warning and its system |
CN110597245A (en) * | 2019-08-12 | 2019-12-20 | 北京交通大学 | Automatic driving track-changing planning method based on quadratic planning and neural network |
CN110826695A (en) * | 2019-10-30 | 2020-02-21 | 京东数字城市(成都)科技有限公司 | Data processing method, device and computer readable storage medium |
CN111145552A (en) * | 2020-01-06 | 2020-05-12 | 重庆大学 | Planning method for vehicle dynamic lane changing track based on 5G network |
CN111746544A (en) * | 2020-07-13 | 2020-10-09 | 吉林大学 | Lane changing method for embodying individual behavior of driver |
CN111897353A (en) * | 2020-07-08 | 2020-11-06 | 西北工业大学 | Aircraft maneuvering trajectory prediction method based on GRU |
CN112373483A (en) * | 2020-11-23 | 2021-02-19 | 浙江天行健智能科技有限公司 | Vehicle speed and steering prediction method based on forward neural network |
CN112578419A (en) * | 2020-11-24 | 2021-03-30 | 南京邮电大学 | GPS data reconstruction method based on GRU network and Kalman filtering |
CN112937564A (en) * | 2019-11-27 | 2021-06-11 | 初速度(苏州)科技有限公司 | Lane change decision model generation method and unmanned vehicle lane change decision method and device |
CN113359764A (en) * | 2021-07-02 | 2021-09-07 | 合肥工业大学 | GRU-based driver model optimization method |
CN113665574A (en) * | 2021-10-22 | 2021-11-19 | 清华大学 | Intelligent automobile lane change duration prediction and anthropomorphic track planning method |
CN114677837A (en) * | 2022-03-16 | 2022-06-28 | 杭州海康威视数字技术股份有限公司 | Traffic simulation method and device based on radar vision data and electronic equipment |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105620480A (en) * | 2015-12-29 | 2016-06-01 | 东南大学 | Decision making method for intelligent vehicle autonomic lane change opportunity |
CN106873584A (en) * | 2017-01-11 | 2017-06-20 | 江苏大学 | Pilotless automobile apery turns to the method for building up of rule base |
CN107766615A (en) * | 2017-09-18 | 2018-03-06 | 东南大学 | Mainline toll station ETC Warning Marks based on dynamics simulation are set apart from method for evaluating safety |
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 |
CN108320051A (en) * | 2018-01-17 | 2018-07-24 | 哈尔滨工程大学 | A kind of mobile robot dynamic collision-free planning method based on GRU network models |
-
2018
- 2018-12-24 CN CN201811579748.6A patent/CN109739218A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105620480A (en) * | 2015-12-29 | 2016-06-01 | 东南大学 | Decision making method for intelligent vehicle autonomic lane change opportunity |
CN106873584A (en) * | 2017-01-11 | 2017-06-20 | 江苏大学 | Pilotless automobile apery turns to the method for building up of rule base |
CN107766615A (en) * | 2017-09-18 | 2018-03-06 | 东南大学 | Mainline toll station ETC Warning Marks based on dynamics simulation are set apart from method for evaluating safety |
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 |
CN108320051A (en) * | 2018-01-17 | 2018-07-24 | 哈尔滨工程大学 | A kind of mobile robot dynamic collision-free planning method based on GRU network models |
Non-Patent Citations (2)
Title |
---|
LING HUANG等: ""Capturing Drivers" Lane Changing Behaviors on Operational Level by Data Driven Methods"", 《SPECIAL SECTION ON ADVANCED BIG DATA ANALYSIS FOR VEHICULAR SOCIAL NETWORKS》 * |
吴昊等: ""基于改进LRCN 模型的驾驶行为图像序列识别方法"", 《江苏大学学报》 * |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110346517A (en) * | 2019-07-25 | 2019-10-18 | 中南大学 | A kind of smart city industrial air pollution visualization method for early warning and its system |
CN110597245A (en) * | 2019-08-12 | 2019-12-20 | 北京交通大学 | Automatic driving track-changing planning method based on quadratic planning and neural network |
CN110826695A (en) * | 2019-10-30 | 2020-02-21 | 京东数字城市(成都)科技有限公司 | Data processing method, device and computer readable storage medium |
CN110826695B (en) * | 2019-10-30 | 2021-05-25 | 京东数字城市(成都)科技有限公司 | Data processing method, device and computer readable storage medium |
CN112937564A (en) * | 2019-11-27 | 2021-06-11 | 初速度(苏州)科技有限公司 | Lane change decision model generation method and unmanned vehicle lane change decision method and device |
CN111145552A (en) * | 2020-01-06 | 2020-05-12 | 重庆大学 | Planning method for vehicle dynamic lane changing track based on 5G network |
CN111145552B (en) * | 2020-01-06 | 2022-04-29 | 重庆大学 | Planning method for vehicle dynamic lane changing track based on 5G network |
CN111897353A (en) * | 2020-07-08 | 2020-11-06 | 西北工业大学 | Aircraft maneuvering trajectory prediction method based on GRU |
CN111897353B (en) * | 2020-07-08 | 2022-08-02 | 西北工业大学 | Aircraft maneuvering trajectory prediction method based on GRU |
CN111746544A (en) * | 2020-07-13 | 2020-10-09 | 吉林大学 | Lane changing method for embodying individual behavior of driver |
CN111746544B (en) * | 2020-07-13 | 2021-05-25 | 吉林大学 | Lane changing method for embodying individual behavior of driver |
CN112373483A (en) * | 2020-11-23 | 2021-02-19 | 浙江天行健智能科技有限公司 | Vehicle speed and steering prediction method based on forward neural network |
CN112578419A (en) * | 2020-11-24 | 2021-03-30 | 南京邮电大学 | GPS data reconstruction method based on GRU network and Kalman filtering |
CN112578419B (en) * | 2020-11-24 | 2023-12-12 | 南京邮电大学 | GPS data reconstruction method based on GRU network and Kalman filtering |
CN113359764A (en) * | 2021-07-02 | 2021-09-07 | 合肥工业大学 | GRU-based driver model optimization method |
CN113665574A (en) * | 2021-10-22 | 2021-11-19 | 清华大学 | Intelligent automobile lane change duration prediction and anthropomorphic track planning method |
CN114677837A (en) * | 2022-03-16 | 2022-06-28 | 杭州海康威视数字技术股份有限公司 | Traffic simulation method and device based on radar vision data and electronic equipment |
CN114677837B (en) * | 2022-03-16 | 2023-08-25 | 杭州海康威视数字技术股份有限公司 | Traffic simulation method and device based on radar data and electronic equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109739218A (en) | It is a kind of that outstanding driver's lane-change method for establishing model is imitated based on GRU network | |
CN108407805B (en) | A kind of vehicle automatic parking method based on DQN | |
CN110060475A (en) | A kind of multi-intersection signal lamp cooperative control method based on deeply study | |
CN110136481A (en) | A kind of parking strategy based on deeply study | |
CN109726804B (en) | Intelligent vehicle driving behavior personification decision-making method based on driving prediction field and BP neural network | |
CN110969848A (en) | Automatic driving overtaking decision method based on reinforcement learning under opposite double lanes | |
CN109733415A (en) | A kind of automatic Pilot following-speed model that personalizes based on deeply study | |
CN109242251A (en) | Vehicular behavior safety detecting method, device, equipment and storage medium | |
CN110389585A (en) | The speed planning device based on study for automatic driving vehicle | |
CN105139072A (en) | Reinforcement learning algorithm applied to non-tracking intelligent trolley barrier-avoiding system | |
CN110304074A (en) | A kind of hybrid type driving method based on stratification state machine | |
CN109976340A (en) | Man-machine cooperation dynamic obstacle avoidance method and system based on deep reinforcement learning | |
CN107229973A (en) | The generation method and device of a kind of tactful network model for Vehicular automatic driving | |
CN109727490A (en) | A kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field | |
Hu et al. | Learning a deep cascaded neural network for multiple motion commands prediction in autonomous driving | |
CN112508164B (en) | End-to-end automatic driving model pre-training method based on asynchronous supervised learning | |
CN109466552A (en) | Intelligent driving lane keeping method and system | |
CN114379583A (en) | Automatic driving vehicle trajectory tracking system and method based on neural network dynamics model | |
CN116134292A (en) | Tool for performance testing and/or training an autonomous vehicle planner | |
CN114282433A (en) | Automatic driving training method and system based on combination of simulation learning and reinforcement learning | |
CN113715842A (en) | High-speed moving vehicle control method based on simulation learning and reinforcement learning | |
CN109240279A (en) | A kind of robot navigation method of view-based access control model perception and spatial cognition neuromechanism | |
CN113552883A (en) | Ground unmanned vehicle autonomous driving method and system based on deep reinforcement learning | |
CN113759901A (en) | Mobile robot autonomous obstacle avoidance method based on deep reinforcement learning | |
Huynh et al. | A Method of Deep Reinforcement Learning for Simulation of Autonomous Vehicle Control. |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190510 |
|
RJ01 | Rejection of invention patent application after publication |