A human-machine collaborative control method and system based on an intelligent steering wheel

Abstract

The application relates to a human-machine collaborative control method and system based on an intelligent steering wheel, wherein the method comprises the following steps: generating an electric signal corresponding to a driving operation based on the intelligent steering wheel; establishing a driver's internal intention prediction and psychological state monitoring model according to the electric signal by using a support vector machine algorithm; obtaining a driver's expectation based on the internal intention prediction result and the driver's psychological state; establishing a kinematics and dynamics model of a vehicle; obtaining road information, vehicle state and model constraints including road safety constraints and stability comfort constraints based on a vehicle-road perception module; constructing a human-machine collaborative controller according to the driver's expectation and the model constraints by using a model predictive control algorithm; solving an MPC controller optimization problem to obtain a current optimal front wheel steering angle and longitudinal acceleration, and outputting the current optimal front wheel steering angle and longitudinal acceleration to a vehicle bottom controller to realize human-machine collaborative control. Compared with the prior art, the application has the advantages of accurate driver intention prediction and strong human-machine collaboration capability.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of human-computer collaborative control, and in particular to a human-computer collaborative control method and system based on an intelligent steering wheel. BACKGROUND

[0002] In intelligent driving services, different drivers need to improve the driving experience in the form that best meets their psychological expectations. However, many existing automatic driving vehicles do not take into account the driving habits of the passengers, which can easily lead to inconsistent human-machine behavior during driving, and thus cause "machine does not trust people, and people does not trust machine". One of the current research hotspots is the individualized performance of intelligent driving in various functional scenarios.

[0003] In addition, in the field of assisted driving, existing driving assistance systems either directly intervene or provide warnings, which are typical local and one-dimensional support, and it is difficult to maintain the ability of the driver while utilizing the performance of sensors, actuators and other machines. Taking a front-wheel steering system as an example, early on, hydraulic power steering was developed on the basis of mechanical steering, and in recent years, variable transmission ratio and vehicle stability assistance functions have been developed. However, none of them can effectively control the human-machine collaboration. With the development of intelligent vehicles, shared steering that focuses on collaborative control has become a new control scheme.

[0004] In the field of human-machine co-driving, existing detection schemes based on drivers can be divided into two categories. One is to analyze the driver's steering behavior through physiological data collected by wearable devices; the other is to monitor and analyze the driving state of the driver through sensors of the vehicle system, such as cameras and controller area network (CAN). However, in the study of driver steering manipulation, the above methods may have some defects. For example, the devices used to obtain physiological information inevitably interfere with the driver; other precision sensors may have non-negligible cost and consumption; data from cameras can be affected by vibration, light and obstruction. Therefore, it is urgent to integrate small, low-cost and highly reliable sensors into the inherent hardware of the vehicle to accurately and effectively detect the driving state of the driver without affecting the normal operation of the driver, so as to accurately predict the steering intention in multiple dimensions.

[0005] For a parallel human-machine collaborative control system, there is often human-machine interaction during manipulation, so the driver is easy to grasp the state of the machine, but this also leads to conflicts between the driver and the machine. Therefore, the human-machine collaborative control system has high requirements for the human-like nature and safety of the machine. Existing collaborative controllers mostly achieve collaborative control through the upper steering mechanism to perceive the steering angle change, and the lower execution mechanism to output an additional compensation angle. This method has a large delay in acquiring signals, and the human-like nature and safety of the control process are also low.

[0006] In summary, an accurate, safe, human-like, efficient and direct personalized human-machine collaborative control system is urgently needed to be developed. SUMMARY

[0007] The purpose of the present application is to provide a human-machine collaborative control method and system based on intelligent steering wheel, realizing accurate, safe, human-like, efficient and direct human-machine collaborative control.

[0008] The purpose of the present application can be realized by the following technical solutions:

[0009] A human-machine collaborative control method based on intelligent steering wheel, comprising the following steps:

[0010] S1: Constructing an intelligent steering wheel based on a triboelectric nanosensor, when a driver holds the intelligent steering wheel to perform a driving operation, an electrical signal corresponding to the driving operation is generated;

[0011] S2: Using a support vector machine algorithm to establish a driver's internal intention prediction and psychological state monitoring model according to the obtained electrical signal;

[0012] S3: Based on the internal intention prediction result and the driver's psychological state, the driver's expectation is obtained, which includes the expected vehicle position, front wheel steering angle and longitudinal acceleration;

[0013] S4: Establishing a kinematics and dynamics model of the vehicle;

[0014] S5: Based on the vehicle-road perception module, the road information and the vehicle state are obtained, and a model constraint including road safety constraint and stability comfort constraint is constructed;

[0015] S6: Using a model predictive control algorithm to construct a human-machine collaborative controller according to the driver's expectation and the model constraint;

[0016] S7: Solving the MPC controller optimization problem to obtain the current optimal front wheel steering angle and longitudinal acceleration, and outputting it to the vehicle bottom controller to realize human-machine collaborative control.

[0017] Preferably, the intelligent steering wheel is composed of a conventional steering wheel and a signal monitoring component, wherein the signal monitoring component includes a plurality of triboelectric nanosensors.

[0018] The triboelectric nanosensor is arranged on the rim of the steering wheel, and is one of a polyimide-polyurethane-copper sensor, a nylon-polytetrafluoroethylene-copper sensor and a nylon-polytetrafluoroethylene-aluminum sensor, which generates an electrical signal when the driver holds the steering wheel in a contact separation mode, wherein the contact separation mode is that when the two polymer films with different adsorption electron capabilities in the triboelectric nanosensor are in contact and separation, the potential difference caused by the triboelectric charge caused by the contact electrification in the interface region and the electrode generates an electrical signal.

[0019] The output waveform of the electrical signal corresponds to the driving operation.

[0020] The construction and training method of the driver's internal intention prediction and mental state monitoring model is:

[0021] Two SVM models are trained for internal intention prediction and mental state monitoring, respectively;

[0022] In the model training stage, a plurality of triboelectric nanosensor electrical signals are collected by a single-chip microcomputer during a plurality of driving operations, sent to a personal computer by a Bluetooth module, preprocessed by filtering and denoising, and labeled, and the label sequence L = {l1, l2, l3,..., ln} is recorded. n};

[0023] In the internal intention prediction model, label = -1 represents left turning, label = 0 represents centering, and label = 1 represents right turning; in the mental state monitoring model, label = -1 represents distraction, label = 0 represents normal state, and label = 1 represents nervous state;

[0024] After obtaining a sufficient data set, effective feature values are extracted to form a feature vector for training the classifier: a series of feature values are calculated from the time domain and the frequency domain, and the feature vector sequence X = {v1, v2, v3,..., vn} is obtained. n};

[0025] The classifier is trained according to the label sequence and the feature vector sequence.

[0026] For different categories of drivers, considering that different ages, genders, and personalities may lead to different driving operation forms of the same internal intention, data of different categories of drivers can be collected for training to obtain personalized prediction models.

[0027] In the application stage of the driver's internal intention prediction and mental state monitoring model, the triboelectric nanosensor electrical signals are collected by a single-chip microcomputer during driving operation, the signals are sent to the vehicle electronic control unit (ECU) through the user datagram protocol (UDP), the ECU calculates the feature values by sliding the time window, selects the trained model matched with the current driver category to process the data, thereby predicting the driver's internal intention, and monitoring the driver's mental state.

[0028] Preferably, the driver's internal intention refers to the driver's turning intention, and the prediction method can realize a prediction advance of about 0.5s compared with the actual steering wheel rotation.

[0029] Preferably, according to the monitored driver's psychological state, the system can timely remind and intervene when the driver may have dangerous driving behavior.

[0030] The driver's expectation includes the expected vehicle position [X ref ,Y ref ], front wheel steering angle δ h and longitudinal acceleration a xh , wherein the expected vehicle longitudinal position X ref is determined by the driver's steering intention and the safety distance from all preceding vehicles obtained by the vehicle-road perception module, and the safety distance from the preceding vehicle is determined by the current psychological state of the driver; the expected vehicle lateral position Y ref is determined by the driver's steering intention, that is, the lateral position of the target lane center line.

[0031] Preferably, the vehicle kinematics and dynamics model is a two-degree-of-freedom model of the vehicle, which aims to balance the steering characteristics of the vehicle and the convenience of the controller implementation.

[0032] Preferably, the vehicle-road perception module is composed of conventional sensing devices, including cameras, millimeter wave radar sensors, ultrasonic radar sensors, etc. for obtaining road information, GNSS positioning modules, etc. for obtaining vehicle pose information. According to the road information and vehicle pose information, the road safety constraints are obtained, and the stability safety constraints obtained by the vehicle kinematics and dynamics model are combined to form the constraints of the model predictive control.

[0033] The road safety constraints include road safety boundary constraints and traffic rule constraints (such as speed limit requirements).

[0034] The road safety boundary constraint can be expressed as:

[0035]

[0036] where y o is the lateral position of the vehicle's center of mass, y r is the right boundary information of the road obtained by the vehicle-road perception module, y l is the left boundary information of the road obtained by the vehicle-road perception module, w is the width of the vehicle, L f is the distance from the vehicle's center of mass to the front end point F of the vehicle, L r is the distance from the vehicle's center of mass to the rear end point R of the vehicle, and ψ is the vehicle yaw angle.

[0037] Taking the speed limit requirement as an example, the traffic rule constraint is:

[0038] v X,min ≤v X ≤v X,max

[0039] where v X,min and v X,max represent the minimum and maximum speed of the speed limit requirement, respectively, and v x represents the vehicle longitudinal speed.

[0040] The stability comfort constraints include vehicle physical constraints related to acceleration, front wheel steering angle, and side slip angle from the tire model, and task constraints related to the driving task.

[0041] The physical constraints are defined as follows:

[0042] a X,min ≤ a X ≤ a X,max

[0043] Δa X,min ≤ Δa x ≤ Δa x,max

[0044] δ min ≤ δ ≤ δ max

[0045] Δδ min ≤ Δδ ≤ Δδ max

[0046] α f,min ≤ α f ≤ α f,max

[0047] α r,min ≤ α r ≤ α r,max

[0048] where a X , Δa X represent the longitudinal acceleration and its rate, respectively, δ and Δδ represent the front wheel steering angle and its variation, respectively, and α f and α r represent the front and rear tire side slip angles, respectively, and the inequality constraints the relationship between each quantity and its corresponding physical constraint.

[0049] To simulate the comfortable acceleration and deceleration behavior of human drivers: during acceleration, the acceleration rate gradually decreases as the acceleration increases, while during deceleration, the acceleration rate gradually decreases as the acceleration decreases. This avoids the discomfort caused by continuously pressing the accelerator or brake pedal. Therefore, Δa X The constraint boundary is specifically defined as:

[0050] Δa X,min = w1(a x,min - a X )

[0051] Δa X,max =w2(a X,max -a X )

[0052] Among them, w1 and w2 represent appropriate weights.

[0053] The task constraints related to the driving task are defined as follows:

[0054] a y,min ≤a y ≤a y,max

[0055] where a y represents the lateral acceleration, a y,min and a y,max Indicates the minimum and maximum lateral acceleration values ​​set to meet the comfort requirements during lane change driving tasks.

[0056] Preferably, the model predictive control algorithm decouples the longitudinal and lateral motions of the vehicle and controls them separately using MPC, taking into account that the driver has different preferences for longitudinal and lateral motions.

[0057] The models used in the longitudinal and lateral MPC are both state space models. Within the prediction range, the state vector x, output y, and control variable u of the longitudinal and lateral motion are defined as:

[0058] x lon =[X,v X ] T ,y lon =X,u lon =a X

[0059] x lat =[v ψ r Y] T ,y lat =Y,u lat =δ

[0060] where x lon 、y lon 、u lon is the longitudinal model variable; x lat 、y lat and u lat are horizontal model variables, X, v X 、a X are the longitudinal position, longitudinal velocity and longitudinal acceleration in the global coordinate system, Y and ψ are the lateral position and yaw angle in the global coordinate system, r is the yaw rate, and δ is the front wheel turning angle.

[0061] Therefore, the longitudinal time step t can be defined separately p,lonand lateral time step t p,lat The discrete-time prediction model under longitudinal and lateral time step t

[0062]

[0063]

[0064]

[0065]

[0066] where N p,lon and N p,lat denote the longitudinal and lateral prediction horizon, respectively, and

[0067] C c,lon = [1, 0].

[0068] By linearizing the lateral nonlinear dynamics model of ego vehicle, the A lat , B lat , C lat and C c,lat matrices can be obtained.

[0069] In the prediction horizon, the desired longitudinal and lateral positions can be defined as:

[0070]

[0071]

[0072] where t is the current time. The desired longitudinal and lateral positions are from X ref and Y ref obtained in step S3.

[0073] The objective function of the MPC controller is:

[0074]

[0075] where N c is the control horizon, Q is the weight matrix for pursuing the compliance with the driving decision command; R u and R du are the weight matrices for minimizing the deviation of the control amount from the desired control amount and for minimizing the fluctuation of the control amount, respectively, which are related to the driving comfort; for the longitudinal control model, is X ref,t , u human|t is the desired longitudinal acceleration a Xh of the driver, for the lateral control model, is Y ref,t , u human|tThe front wheel turning angle δ is expected for the driver h .

[0076] The MPC optimization problem is:

[0077]

[0078] wherein,

[0079]

[0080]

[0081] The above model is then applied to longitudinal and lateral control respectively.

[0082] A human-machine collaborative control system based on an intelligent steering wheel, comprising a memory, a processor, and a program stored in the memory, wherein the processor implements the method as described above when executing the program.

[0083] Compared with the prior art, the present application has the following beneficial effects:

[0084] (1) The intelligent steering wheel with the installed triboelectric nanosensor used in the present application can effectively obtain the electrical signal corresponding to the driving operation of the driver without affecting the driver, and the triboelectric nanosensor has the characteristics of lightness, thinness, flexibility, and low cost, can adapt to various irregular surfaces, is easy to lay, has the self-power supply characteristic, does not need external power supply, and takes into account the manufacturing cost, installation convenience, and use comfort.

[0085] (2) The driver's internal intention prediction and psychological state monitoring model constructed in the present application can be personalized for analysis of different categories of drivers, improving the human-machine interaction experience. The internal intention prediction can realize a prediction lead time of about 0.5s, solving the single problem of the prediction signal and the delay problem of the traditional human-machine collaborative control system relying on the steering wheel angle sensor as the input, improving the control accuracy and timeliness of the human-machine collaborative control system, and playing an important role in emergency situations such as emergency obstacle avoidance.

[0086] (3) The method of the present application controls in the longitudinal and lateral directions respectively, constructs the distribution problem of the driving right as a model predictive control problem, and converts the road safety and stability comfort requirements into the constraints of model predictive control, thereby guaranteeing the safety of collaborative control under the premise of meeting the most sufficient driving right of the driver. BRIEF DESCRIPTION OF DRAWINGS

[0087] Figure 1 The flowchart of the method of the present application;

[0088] Figure 2It is a structural schematic diagram of a triboelectric nanosensor, wherein 1 is a conductive copper foil, 2 is a polytetrafluoroethylene thin layer, and 3 is a nylon thin layer.

[0089] Figure 3 It is a schematic diagram of a vehicle lateral dynamics model used in the application.

[0090] Figure 4 It is a schematic diagram of a vehicle-road perception module constructed for obtaining road safety constraints.

[0091] Figure 5 It is a schematic diagram of a road safety boundary constraint established in the application. DETAILED DESCRIPTION

[0092] The application will be described in detail below with reference to the accompanying drawings and specific embodiments. The embodiments are implemented on the premise of the technical solution of the application, and detailed implementation modes and specific operation processes are given, but the protection scope of the application is not limited to the following embodiments.

[0093] The embodiment provides a human-machine collaborative control method based on an intelligent steering wheel, which can be applied to various vehicles equipped with steering wheels, such as cars, buses, trucks, and special-purpose vehicles. Figure 1 As shown in the figure, the method comprises the following steps:

[0094] S1: Constructing an intelligent steering wheel based on a triboelectric nanosensor, and generating an electric signal corresponding to a driving operation when a driver holds the intelligent steering wheel to perform the driving operation.

[0095] The intelligent steering wheel in the embodiment is composed of a traditional steering wheel and a signal monitoring component, wherein the signal monitoring component comprises a plurality of triboelectric nanosensors. Specifically, the signal monitoring component comprises a plurality of sensor groups, the plurality of sensor groups are uniformly arranged around the rim of the steering wheel, each sensor group comprises three triboelectric nanosensors, and the triboelectric nanosensors of each sensor group are respectively arranged on the front side, the outer peripheral side, and the rear side of the steering wheel at the same angle, so as to completely monitor the driving operation of the driver and generate a corresponding electric signal.

[0096] The triboelectric nanosensor arranged on the rim of the steering wheel is one of a polyimide-polyurethane-copper sensor, a nylon-polytetrafluoroethylene-copper sensor, and a nylon-polytetrafluoroethylene-aluminum sensor, and generates an electric signal when the driver holds the steering wheel in a contact separation mode, wherein the contact separation mode is that when two active triboelectric materials with different adsorption electron capabilities are in contact under the action of extrusion or bending, friction charges will be generated on the surfaces of the two active triboelectric materials (for example Figure 2When the polytetrafluoroethylene layer 2 and the nylon thin layer 3 are in contact (the polytetrafluoroethylene thin layer 2 is positively charged, and the nylon thin layer 3 is negatively charged), the triboelectric charge caused by the contact electrification will cause a potential difference in the interface area and the electrode. If an external load is connected, an electric current will flow, generating an electrical signal corresponding to the driving operation. Based on this principle, a pressure sensor is made to detect the driving operation of the driver. Specifically, when the palm is in contact with the steering wheel and pressure is applied, the triboelectric charge caused by the contact electrification will generate a potential difference in the interface area, which is manifested as a high potential signal in the triboelectric nanosensor, and the potential is positively correlated with the pressure.

[0097] As shown in Figure 2 In this embodiment, a nylon-polytetrafluoroethylene-copper triboelectric nanosensor is used, which includes a conductive copper foil 1, a polytetrafluoroethylene thin layer 2, a nylon thin layer 3, and a heat shrinkable package as the outermost layer to cover the sensor. The top surface of the nylon thin layer 3 and the polytetrafluoroethylene thin layer 2 both have copper as the electrode. The prepared sensor has a sandwich structure, in which the free surface of the polytetrafluoroethylene thin layer 2 is located on the free surface of the nylon thin layer 3. Then, a layer of heat shrinkable film is covered on the two thin films. The heat shrinkable package is heated using a handheld heat gun to keep the polymer layer attached and completely cover and tighten the core part of the sensor. Using the heat shrinkable package can manufacture a sandwich structure-based sensor, which is encapsulated as a whole, light and thin, flexible, highly flexible and heat-resistant.

[0098] In this embodiment, the signal monitoring assembly includes eight sensor groups, each of which is arranged at an interval of 45° around the steering wheel rim to achieve the arrangement of the signal monitoring assembly. Specifically, the wires of all triboelectric nanosensors are hidden under the film and are collected to the signal collection and processing assembly through the internal cavity of the steering wheel. The greater the holding force, the higher the voltage of the generated electrical signal. A total of 24 triboelectric nanosensors monitor the holding information in real time. The film of the sensor is easy to maintain, disassemble and replace. Moreover, the sensor does not require additional power supply and is a self-powered device. The steering wheel can be equipped with a personalized outer steering wheel cover without affecting the reaction sensitivity of the triboelectric nanofilm.

[0099] During driving, the driver holds the steering wheel to generate a driving operation; the triboelectric nanosensors distributed around the outer rim of the steering wheel generate an electrical signal under pressure; and the electrical signal is collected by wires to the signal collection and processing assembly located in the central cavity of the steering wheel.

[0100] S2: An SVM algorithm is used to establish a driver's intrinsic intention prediction and psychological state monitoring model according to the obtained electrical signal.

[0101] Two SVM models are trained for intrinsic intention prediction and psychological state monitoring, respectively;

[0102] In the model training stage, a plurality of groups of triboelectric nanometer sensor electric signals are collected by the single-chip microcomputer during a plurality of driving operations, and are sent to the personal computer by the Bluetooth module, data preprocessing is performed on the signals after filtering and denoising, and the signals are labeled, and a label sequence L={l1, l2, l3,...,l n};

[0103] In the internal intention prediction model, label=-1 represents left turning, label=0 represents centering, and label=1 represents right turning; in the psychological state monitoring model, label=-1 represents distraction, label=0 represents normal state, and label=1 represents nervous state.

[0104] After obtaining a sufficient data set, effective feature values are extracted to form a feature vector for training of the classifier: a series of feature values are calculated from the time domain and the frequency domain, and a feature vector sequence X={v1, v2, v3,...,v n};

[0105] The classifier is trained according to the label sequence and the feature vector sequence.

[0106] For different categories of drivers, considering that different ages, genders, and personalities may lead to different driving operation forms of the same internal intention, data of different categories of drivers can be collected for training to obtain personalized prediction models.

[0107] In the application stage of the model, the triboelectric nanometer sensor electric signals are collected by the single-chip microcomputer during driving operation, the signals are sent to the vehicle electronic control unit (ECU) by the user datagram protocol (UDP), the ECU calculates the feature values by sliding a time window, selects the trained model matched with the current driver category to process the data, thereby predicting the internal intention of the driver, and monitoring the psychological state of the driver.

[0108] The internal intention of the driver refers to the turning intention of the driver, and the prediction method can realize a prediction advance of about 0.5s compared with the actual rotation of the steering wheel.

[0109] According to the monitored psychological state of the driver, the system can timely remind and intervene when the driver may have dangerous driving behavior, specifically, the steering wheel vibration or buzzer warning function can be set when the driver is distracted.

[0110] Specifically, the signal collection and processing assembly in the embodiment is composed of an embedded single-chip microcomputer, which includes a signal acquisition module, a communication module, and a storage module. Correspondingly, the signal processing module is a single-chip microcomputer processor, and the storage module is a single-chip microcomputer TF card. The communication module in the embodiment is a Bluetooth module. The signal acquisition module is based on embedded technology and realizes functions by a miniature single-chip microcomputer assisted by a simple filter circuit. The signal acquisition module mainly uses an ADC analog-digital conversion module to realize analog-digital conversion of sensor electrical signals to digital signals. The signals of various sensors are stored in arrays after being filtered, and the corresponding time sequence is transmitted to a personal computer by the Bluetooth module for storage and analysis, or the data is directly stored to the TF card by the single-chip microcomputer.

[0111] The signal conversion and transmission part of the signal collection and processing assembly is small in size and simple in structure, and can be built-in in the central cavity of the steering wheel. The signals of various sensors are collected by wires to the receiving end of the single-chip microcomputer.

[0112] In the embodiment, the single-chip microcomputer is an Infineon TC264 single-chip microcomputer, which has multiple ADC analog-digital conversion modules and UART communication modules, meeting the signal acquisition requirements of the arranged 24 sensors. The baud rate of the Bluetooth communication is set to 115200, which can quickly transmit a large amount of data, including time sequences and sensor voltage signals at corresponding time points. The master and slave of the Bluetooth can transmit instructions to each other.

[0113] Further, Kalman filtering and median filtering are used in the embodiment to filter and denoise the collected data set. The maximum value, minimum value, average value, variance, kurtosis, and skewness of each sensor data are extracted to form a feature sequence X, which is used for model training.

[0114] To meet the individual needs, we recruited volunteers from twelve categories of drivers, including young, middle-aged, and old drivers, male and female drivers with skilled driving skills and those who have just obtained a driver's license, to collect driving operation data on a driving simulator. We trained models matched with different categories of drivers to predict the intrinsic intention of the drivers and monitor their psychological state.

[0115] S3: Based on the intrinsic intention prediction result and the psychological state of the driver, the driver's expectation is obtained.

[0116] The driver's expectation includes the expected vehicle position [X ref ,Y ref ], the front wheel steering angle δ h , and the longitudinal acceleration a Xh . The expected vehicle longitudinal position X ref is determined by the driver's steering intention and the safety distance from all preceding vehicles obtained by the vehicle-road perception module, and the safety distance from the preceding vehicle is determined by the current psychological state of the driver; the expected vehicle lateral position Yref is determined by the driver's steering intention, i.e., the lateral position of the target lane center line.

[0117] Specifically, in this embodiment, the safe distance d s,normal between the vehicle and the preceding vehicle under normal psychological state of the driver is defined as s The relationship between the safe distance d

[0118]

[0119] This is because the nervous state is often in a complex or even dangerous traffic scene, and the weight of the safe distance on the driving decision needs to be emphasized.

[0120] S4: Establish the kinematic and dynamic model of the vehicle.

[0121] In this embodiment, after comparing various degrees of freedom models, a two-degree-of-freedom model of the vehicle is selected. This model has the ability to fully display the steering characteristics of the vehicle, and is also conducive to the implementation of the controller. When establishing the kinematic and dynamic model of the vehicle, the research scenario of this embodiment is that the vehicle is driving on a flat road, and the motion of the vehicle is regarded as a planar motion.

[0122] The longitudinal kinematic model of the vehicle is defined as:

[0123]

[0124]

[0125] where X is the longitudinal position, and v X represents the longitudinal velocity, and a X represents the longitudinal acceleration, all in the global coordinate system.

[0126] In order to more accurately consider the lateral dynamics of the vehicle and improve the stability of the lateral motion, a linear tire model is adopted, as shown in Figure 3 wherein:

[0127]

[0128]

[0129]

[0130]

[0131] where u and v represent the longitudinal and lateral velocities in the vehicle coordinate system, respectively, r represents the yaw rate, Y and ψ represent the lateral position and yaw angle of the vehicle in the global coordinate system, respectively, m and Iz respectively represent the vehicle mass and the moment of inertia, l f and l r respectively represent the distance from the vehicle center of mass to the front and rear axles, F yf and F yr respectively represent the lateral forces of the front and rear tires, for a linear tire model, the lateral force calculation formula is:

[0132]

[0133]

[0134] where C αf and C αr respectively represent the lateral stiffness of the front and rear tires.

[0135] It should be noted that the kinematic and dynamic models of the vehicle in this embodiment can be directly transferred to the Frenet coordinate system through coordinate transformation.

[0136] S5: Obtain road information, vehicle state based on the vehicle-road perception module, and construct model constraints including road safety constraints and stability comfort constraints.

[0137] In this embodiment, the vehicle-road perception module constructed by traditional sensing devices, such as Figure 4 As shown, including camera, millimeter wave radar sensor, ultrasonic wave radar sensor, etc. for obtaining road information, GNSS positioning module, etc. for obtaining vehicle pose information. According to the road information and the vehicle pose information, the road safety constraints are obtained, and the stability safety constraints obtained from the vehicle kinematic and dynamic model are combined to constitute the constraints of the model predictive control.

[0138] The road safety constraints include road safety boundary constraints, traffic rule constraints (such as speed limit requirements).

[0139] The road safety boundary constraint can be expressed as:

[0140]

[0141] where y o is the lateral position of the vehicle center of mass, y r is the right boundary information of the road obtained by the vehicle-road perception module, y l is the left boundary information of the road obtained by the vehicle-road perception module, w is the vehicle width, L f is the distance from the vehicle center of mass to the front end point F of the vehicle, L r is the distance from the vehicle center of mass to the rear end point R of the vehicle, and ψ is the vehicle yaw angle.

[0142] Taking the speed limit requirement as an example, the traffic rule constraint is:

[0143] v X,min ≤v X ≤v X,max

[0144] where v X,min and v X,max denote the minimum and maximum speed of the speed limit requirement, respectively, and v X denotes the vehicle longitudinal speed.

[0145] The stability comfort constraints include vehicle physical constraints related to acceleration, front wheel steering angle, and side slip angle from the tire model, and task constraints related to the driving task.

[0146] The physical constraints are defined as follows:

[0147] a X,min ≤a X ≤a X,max

[0148] Δa X,min ≤Δa X ≤Δa X,max

[0149] δ min ≤δ≤δ max

[0150] Δδ min ≤Δδ≤Δδ max

[0151] α f,min ≤α f ≤α f,max

[0152] α r,min ≤α r ≤α r,max

[0153] where a X , Δa X denote the longitudinal acceleration and its rate, respectively, δ and Δδ denote the front wheel steering angle and its variation, respectively, and α f and α r denote the front and rear tire side slip angles, respectively, and the inequality constraints the relationship of each quantity with its corresponding physical constraint.

[0154] To simulate the comfortable acceleration and deceleration behavior of human drivers: during acceleration, the acceleration rate gradually decreases as the acceleration increases, while during deceleration, the acceleration rate gradually decreases as the acceleration decreases. This avoids the discomfort caused by continuously pressing the accelerator or brake pedal. Therefore, Δa X The constraint boundary is specifically defined as:

[0155] ΔaX,min = w1(a X,min -a X )

[0156] Δa X,max = w2(a X,max -a X )

[0157] where w1 and w2 represent suitable weights.

[0158] The task constraints related to driving tasks are defined as follows:

[0159] a y,min ≤ a y ≤ a y,max

[0160] where a y represents lateral acceleration, a y,min and a y,max represent the minimum and maximum values of lateral acceleration set to meet the comfort requirements during the execution of the lane change driving task.

[0161] S6: A model predictive control algorithm is used to construct a human-machine collaborative controller according to the driver's expectations and model constraints.

[0162] In this embodiment, the model predictive control algorithm takes into account the different preferences of the driver for longitudinal and lateral motion, decouples the longitudinal and lateral motion of the vehicle, and uses MPC for control respectively.

[0163] The models used by the longitudinal and lateral MPCs are both state space models. Within the prediction range, the state vectors x, output quantities y, and control quantities u of the longitudinal and lateral motion are defined as follows:

[0164] x lon = [X, v X ] T y lon = X, u lon = a X

[0165] x lat = [v, ψ, Y] T y lat = Y, u lat = δ

[0166] where x lon , y lon , and u lon are longitudinal model variables; x lat , y lat , and u lat are lateral model variables, X, v X , aX respectively, are the longitudinal position, velocity and acceleration in global coordinate system, Y and ψ are the lateral position and yaw angle in global coordinate system, r is the yaw rate, and δ is the front wheel steering angle.

[0167] Thus, the discrete-time longitudinal and lateral prediction models can be defined as: p,n p,

[0168]

[0169]

[0170]

[0171]

[0172] where N p,lon and N p,lat are the longitudinal and lateral prediction horizon, respectively, and

[0173] C c,lon = [1, 0].

[0174] The matrices A lat , B lat , C lat and C c,lat can be obtained by linearizing the lateral nonlinear dynamics model of ego vehicle.

[0175] The desired longitudinal and lateral positions in the prediction horizon can be defined as:

[0176]

[0177]

[0178] where t is the current time. The desired longitudinal and lateral positions are X ref and Y ref from step S3.

[0179] The objective function of the MPC controller is:

[0180]

[0181] where N c is the control horizon, Q is the weight matrix for pursuing the compliance with the driving decision command; R u and R du are the weight matrices for minimizing the deviation of the control amount from the desired control amount and for minimizing the fluctuation of the control amount, respectively, which are related to the driving comfort; for the longitudinal control model,​​ is X ref,t is u human|t is the driver's desired longitudinal acceleration a Xh , for the lateral control model, is Y ref,t is u human|t is the driver's desired front wheel steering angle δ h .

[0182] The MPC optimization problem is:

[0183]

[0184] where,

[0185]

[0186]

[0187] The above models are then applied to longitudinal and lateral control respectively.

[0188] S7: solve the MPC controller optimization problem to obtain the current optimal front wheel steering angle and longitudinal acceleration, and output them to the vehicle bottom controller to realize human-machine collaborative control.

[0189] The simulation results show that under safe conditions, the driver's operation is almost not intervened, and under dangerous conditions, the machine can intervene in time to avoid the danger of collision. At the same time, compared with the use of steering wheel angle sensor, the prediction value with advance based on the frictional electrostatic nanosensor and SVM model described in S2 and S3 effectively improves the performance of collaborative control under various conditions.

[0190] The experimental data show that the signal change of the vehicle-mounted steering wheel angle sensor is obviously later than that of the frictional electrostatic nanosensor, with an average delay of 0.5s. The research shows that inputting this prediction value with advance into the model predictive controller will help to optimize the human-machine collaborative control task. Under dangerous conditions, since the machine can intervene in control earlier, the stability and safety of operation are improved; under safe conditions, this advance helps to reduce the difference between the prediction value and the true value, and better realize the following of the driver's intention.

[0191] The above detailed the preferred embodiments of the present application. It should be understood that those skilled in the art can make many modifications and changes without creative labor according to the concept of the present application. Therefore, any technical solution obtained by logical analysis, reasoning or limited experiment by those skilled in the art on the basis of the prior art according to the concept of the present application shall be within the protection scope determined by the claims.

Claims

1. A human-machine collaborative control method based on an intelligent steering wheel, characterized in that, The method comprises the following steps: S1: constructing an intelligent steering wheel based on triboelectric nanosensors, and generating an electrical signal corresponding to a driving operation when a driver holds the intelligent steering wheel to perform the driving operation; S2: establishing a driver's intrinsic intention prediction and mental state monitoring model according to the obtained electrical signal by using a support vector machine algorithm; The construction and training method of the driver's intrinsic intention prediction and mental state monitoring model is as follows: Two SVM models are trained for intrinsic intention prediction and mental state monitoring, respectively; In the model training stage, a plurality of groups of triboelectric nanosensor electric signals are collected by the single-chip microcomputer during a plurality of driving operations, are sent to the personal computer by the Bluetooth module, are subjected to filtering and denoising data preprocessing, are marked, and a label sequence is recorded ; intrinsic intention prediction model, represents left turn, represents centering, represents right turn; mental state monitoring model, represents distraction, represents normal state, represents tense state; After obtaining enough data sets, the effective feature values are extracted to form the feature vectors for the classifier to train: a series of feature values are calculated from the time domain and the frequency domain to obtain the feature vector sequence ; The classifier is trained according to the label sequence and the feature vector sequence; S3: obtaining a driver's expectation based on the intrinsic intention prediction result and the driver's mental state, wherein the driver's expectation includes an expected vehicle position, a front wheel steering angle and a longitudinal acceleration; S4: establishing a kinematics and dynamics model of the vehicle; S5: obtaining road information and vehicle state based on a vehicle-road perception module, and constructing a model constraint including road safety constraints and stability comfort constraints; S6: constructing a man-machine collaborative controller according to the driver's expectation and the model constraint by using a model predictive control algorithm; S7: solving an MPC controller optimization problem to obtain a current optimal front wheel steering angle and longitudinal acceleration, and outputting the same to a vehicle bottom controller to realize man-machine collaborative control. 2.The human-machine collaborative control method based on the intelligent steering wheel according to claim 1, wherein, The intelligent steering wheel is composed of a conventional steering wheel and a signal monitoring component, wherein the signal monitoring component comprises a plurality of triboelectric nanosensors. 3.The human-machine collaborative control method based on the intelligent steering wheel according to claim 2, characterized in that, The triboelectric nanosensor is arranged on the rim of the steering wheel, is one of a polyimide-polyurethane-copper sensor, a nylon-polytetrafluoroethylene-copper sensor and a nylon-polytetrafluoroethylene-aluminum sensor, and generates an electrical signal when the driver holds the steering wheel in a contact separation mode, wherein the contact separation mode refers to that when two polymer films with different adsorption electron capabilities in the triboelectric nanosensor are in contact and separation, a potential difference caused by triboelectric charges caused by the contact electrification in the interface region and the electrode generates an electrical signal. 4.The human-machine collaborative control method based on the intelligent steering wheel according to claim 1, wherein, In the application stage, the driver's intrinsic intention prediction and mental state monitoring model collects the electrical signal of the triboelectric nanosensor through a single-chip microcomputer during the driving operation, sends the signal to a vehicle electronic control unit (ECU) through a user datagram protocol (UDP), the ECU calculates feature values by sliding a sensor electrical signal sequence through a sliding time window, selects a trained model matched with the current driver category to process the data, thereby predicting the driver's intrinsic intention and monitoring the driver's mental state.

5. The human-machine collaborative control method based on the intelligent steering wheel according to claim 1, characterized in that, the driver expectation includes a desired vehicle position , a front wheel steering angle and a longitudinal acceleration , wherein the desired vehicle longitudinal position is determined by the driver steering intention and the safety distance to all preceding vehicles obtained by the car-road perception module, and the safety distance to preceding vehicles is determined by the current mental state of the driver; and the desired vehicle lateral position is determined by the driver steering intention, i.e. the lateral position of the target lane center line. 6.The human-machine collaborative control method based on the intelligent steering wheel according to claim 1, wherein, The model predictive control algorithm considers that the driver has different preferences for longitudinal and lateral motion, decouples the longitudinal and lateral motion of the vehicle, and controls them by using MPC respectively: The models employed by the longitudinal and lateral MPCs are state space models, in which the state vectors of longitudinal and lateral motion are defined as the output quantities and the control quantities respectively within the prediction horizon wherein , , is a longitudinal model variable; , and are lateral model variables, are longitudinal position, longitudinal velocity and longitudinal acceleration in the global coordinate system, respectively, and are lateral position and yaw angle in the global coordinate system, respectively, is the yaw rate, is the front wheel steering angle; The discrete-time prediction model under longitudinal time steps and lateral time steps is defined as wherein and respectively denote the longitudinal and lateral prediction range, and By linearizing the lateral nonlinear dynamics model of the ego vehicle, we get , , and matrix; In the prediction range, the expected longitudinal and lateral positions are defined as: wherein is the current time, the desired longitudinal position and the lateral position from step S3.

7. The human-machine collaborative control method based on the intelligent steering wheel according to claim 1, characterized in that, The objective function of the MPC controller is: wherein is a control range, is a weight matrix pursuing the fitting of the driving decision command; and are weight matrices for minimizing the deviation of the control amount from the desired control amount and for minimizing the fluctuation of the control amount, respectively, and are related to driving comfort; for the longitudinal control model, , , for the lateral control model, , . 8.The human-machine collaborative control method based on the intelligent steering wheel according to claim 7, wherein, The MPC controller optimization problem is: Wherein, 。 9.A human-machine collaborative control system based on an intelligent steering wheel, comprising a memory, a processor, and a program stored in the memory, characterized in that, The processor implements the method of any one of claims 1-8 when executing the program.

Images