Process of steering a vehicle in a turn

Abstract

The invention relates to a method for controlling a motor vehicle (10) traveling on a traffic lane (31) of a road (30), comprising: ¤ when the motor vehicle is controlled by a driver (40), a step of acquiring data relating to the trajectory of the motor vehicle when turning, ¤ when said automatic centering function is activated, steps of: - determining a center line (T0) of said traffic lane, - calculating a control instruction for the motor vehicle as a function of said center line and determined parameters, and - controlling a steering actuator according to said control instruction.According to the invention, provision is made to select a cornering driving mode based on said data acquired when the motor vehicle is driven by the driver, and, when the automatic centering function of the motor vehicle is activated, at least part of the parameters are determined based on the selected cornering driving mode. Figure for abstract: Fig.1.

Description

Title of the invention: Method for steering a vehicle in a bend Technical field of the invention

[0001] The present invention relates generally to driving aids for motor vehicles.

[0002] The invention relates more specifically to a method for controlling a motor vehicle traveling on a traffic lane of a road, comprising: H when the motor vehicle is driven by a driver (and therefore an automatic centering function of the motor vehicle in said traffic lane is not activated), a step of acquiring data relating to the trajectory of the motor vehicle when turning relative to edge lines of said traffic lane and / or relative to the edge lines, H when said automatic centering function is activated, steps of: - determination of a center line of said traffic lane, - calculation, by a computer on board said motor vehicle, of a steering instruction for the motor vehicle as a function of said center line and determined parameters, and - control by the computer of a steering actuator of said motor vehicle according to said control instruction.

[0003] The invention also relates to a motor vehicle suitable for implementing such a method. It applies more particularly to cars and other motorized vehicles traveling on roads. State of the art

[0004] In an effort to make motor vehicles safer, they are currently being equipped with driving assistance functions and even highly automated driving functions.

[0005] These are typically centering or lane centering functions (better known respectively by the English acronym LCA for “Lane Centering Assist” or LKA for “Lane Keeping Assist”).

[0006] A centering function (hereinafter referred to as the "LCA function") needs, in order to operate, to know the position of the edges of the traffic lane taken by the vehicle. Currently, it is known to use a sensor, such as a camera, which incorporates image processing means in order to determine the position of each of the lane edge marking lines.

[0007] A computer on board the vehicle can then deduce the position of the center line of the traffic lane taken, which then allows it to control automatically the vehicle so that the latter follows this center line.

[0008] This solution is however not entirely satisfactory for the driver and the other passengers of the vehicle since, when activating the LCA function, this function will have a way of driving the vehicle which will change from that of the driver, which could be a source of discomfort or even irritation for the driver. Presentation of the invention

[0009] In order to overcome the aforementioned drawback of the state of the art, the present invention proposes to determine, when the LCA function is not activated, the manner in which the driver takes turns and then, when this function is activated, to steer the vehicle by copying the driver's way of driving when turning.

[0010] More particularly, the invention proposes a method as defined in the introduction, in which provision is made to select a cornering driving mode as a function of said data acquired when the motor vehicle is driven by the driver, and in which, when the LCA function for automatic centering of the motor vehicle is activated, at least part of the parameters (of the LCA function) are determined as a function of the cornering driving mode selected.

[0011] Thus, thanks to the invention, different ways of taking turns are modeled. It is then planned, when the driver is driving, to determine which of these ways is used by the driver. When activating the LCA function, the steering parameters of the vehicle when cornering are adjusted so that the vehicle follows trajectories of the same type as those which were taken when the driver was driving.

[0012] It is noted here that for this, the trajectory that the vehicle must follow is not modified (the center line of the traffic lane remains the reference trajectory), only the parameters used by the LCA function are adjusted so that the vehicle follows the desired trajectories when turning. In this way, it is not necessary to reprogram either the algorithms for implementing the LCA function, or the algorithms for calculating the reference trajectory.

[0013] Other advantageous and non-limiting characteristics of the method according to the invention, taken individually or in all technically possible combinations, are the following: - the said cornering driving mode is selected by means of a neural network - said neural network is of the recurrent type, preferably of the GRU type; - a preliminary operation of parameterization of the calculator is planned, including a step of developing a database associating with test trajectories of cornering driving modes, a step of learning the neural network using said database, and saving the neural network in the computer; - said data are chosen from at least: a time indicator, a distance between the motor vehicle and at least one of the edge lines, a curvature data item of at least one of the edge lines, a data item relating to the dynamics of the motor vehicle, a distance between the motor vehicle and a point of at least one of the edge lines which is the detected point furthest from the motor vehicle; - the cornering driving mode is selected based at least on the data acquired in the last corner taken by the motor vehicle before activation of the automatic centering function of the motor vehicle; - the driving mode is selected from a list comprising at least five predetermined and distinct cornering driving modes, and preferably at least nine cornering driving modes; - if at least in one bend, the trajectory of the motor vehicle did not correspond to any of the bend driving modes in the said list, it is planned in after-sales service to add to the said list at least one new bend driving mode predetermined and distinct from the other bend driving modes; - said parameter is a reference value to be achieved and / or a gain associated with: a lateral deviation between the motor vehicle and the center line, and / or a heading angle between the motor vehicle and the center line, and / or a position error integral between the motor vehicle and the center line.

[0014] The invention also relates to a motor vehicle comprising: - acquisition means adapted to acquire, when the motor vehicle is moving on a traffic lane, data relating to the edge lines of said traffic lane, - a steering actuator adapted to control the motor vehicle, and - a computer programmed to implement a control method as mentioned above.

[0015] Of course, the various features, variants and embodiments of the invention may be combined with each other in various combinations to the extent that they are not incompatible or mutually exclusive. Detailed description of the invention

[0016] The description which follows with reference to the appended drawings, given as non-limiting examples, will make it clear what the invention consists of and how it can be implemented.

[0017] In the attached drawings:

[0018] [Fig.l] is a schematic perspective view of a motor vehicle suitable for implementing a method according to the invention;

[0019] [Fig.2] is a schematic top view of the motor vehicle of [Fig.l] driving on a road;

[0020] [Fig.3] represents a functional diagram of the calculations implementing an automatic centering function of the motor vehicle of [Fig.l] in the center of its traffic lane;

[0021] [Fig.4] illustrates turning trajectories;

[0022] [Fig.5] is a diagram illustrating steps of implementing the control method according to the invention.

[0023] In [Fig.l], a motor vehicle 10 is shown which is suitable for implementing the invention.

[0024] This is a car. Alternatively, it could be another type of vehicle (truck, motorcycle, etc.).

[0025] In this figure, the motor vehicle 10 is traveling on a traffic lane 31 of a road 30. It can be seen that the road 30 has two lateral marking lines 34, 35 (which delimit it) and a central marking line 33 delimiting two traffic lanes 31, 32.

[0026] A traffic lane is herein defined as the part of a road on which only one vehicle is permitted to travel at a time. Such a traffic lane is generally demarcated between marking lines.

[0027] A road (or roadway) is defined as a set of traffic lanes. In the example considered here for illustrative purposes, this road 30 therefore comprises two traffic lanes 31 on which vehicles can travel in the same direction.

[0028] The center line T0 of the taxiway will be defined as being the geometric curve extending along the center of the taxiway 31, at an equal distance from the two lateral marking lines 34, 35.

[0029] As shown in [Fig.l], the motor vehicle 10 conventionally comprises a chassis, front steered wheels 13 and rear non-steered wheels 14, and a passenger compartment in which there is in particular a seat for the driver 40 of the vehicle and a steering wheel 12.

[0030] This motor vehicle 10 conventionally comprises a powertrain, a braking system and a steering system for turning the vehicle. Conventionally, the steering system comprises an electronically controllable power steering actuator, the powertrain comprises an electronically controllable engine control actuator, and the braking system comprises an electronically controllable braking actuator.

[0031] The motor vehicle 10 further comprises an electronic unit and / or in processing computer (hereinafter called computer 11) comprising at least one microprocessor, at least one memory and input and output interfaces.

[0032] Thanks to its input interfaces, the computer 11 is adapted to receive different input data which come from sensors or third-party computers.

[0033] Among these sensors, for example, a front camera is provided for locating the edges of the traffic lane 31 being used. Sensors are also provided that are suitable for determining the values ​​of other parameters mentioned later, relating to the position of the vehicle in its traffic lane, to the traffic lane and to the dynamics of the vehicle.

[0034] Thanks to its output interfaces, the computer is adapted to control the power steering actuator, the engine control actuator, and the braking actuator.

[0035] Thanks to its memory, the computer stores a computer application, consisting of computer programs comprising instructions whose execution by the computer allows the implementation of a function for automatically maintaining the vehicle in its lane (hereinafter called the LCA function), and more generally of the method described below.

[0036] Before describing this process in more detail, we can introduce the different variables that will be used, some of which are illustrated in [Fig.2].

[0037] The total mass of the motor vehicle will be noted “m” and will be expressed in kg.

[0038] The center of gravity of the vehicle will be noted “CG”.

[0039] The mass of the vehicle which is exerted on the front wheel set will be noted “Mf” and will be expressed in kg.

[0040] The mass of the vehicle which is exerted on the rear wheel set will be noted “Mr” and will be expressed in kg.

[0041] The wheelbase of the vehicle, that is to say the distance between the axes of these two wheel sets, will be noted “L” and will be expressed in meters.

[0042] The pneumatic stiffness of the rear wheels will be noted Cr and will be expressed in Newton / rad.

[0043] The pneumatic stiffness of the front wheels will be noted Cf and will be expressed in Newton / rad.

[0044] We can consider an orthogonal reference frame (CG, X, Y, Z) attached to the vehicle. Its origin is confused with the center of gravity CG. The X axis corresponds to the longitudinal axis of the vehicle. The Y axis corresponds to the lateral axis turned to the left of the vehicle. When the vehicle is traveling on a horizontal road, the Z axis corresponds to the vertical axis. More generally, this Z axis is the axis normal to the road.

[0045] The steering angle that the front steered wheels make with the longitudinal axis X of the motor vehicle 10 will be noted “ô” and will be expressed in rad.

[0046] It will be noted that the steering wheel angle and the steering angle ô are directly linked, with a gear ratio or even a first or second order dynamic. In the following, only the steering angle ô will be considered. to the wheels.

[0047] The steering speed of the front wheels will be noted “dô / dt”.

[0048] The longitudinal speed of the vehicle, along the X axis, will be noted v and will be expressed in m / s.

[0049] The relative heading angle between the X axis and the tangent to the centerline T0 at the center of gravity CG will be noted “W” and will be expressed in rad.

[0050] The yaw rate of vehicle 1, i.e. its rotation speed around the Z axis, will be noted “drp / dt”.

[0051] The lateral position error, also called lateral deviation, between the center of gravity CG of the vehicle and the center line T0 will be noted y.

[0052] The lateral speed of the vehicle will be noted “dy / dt”.

[0053] At this stage, we can also introduce a notion of “position error integral”, which corresponds to the time integral of the lateral deviations y with respect to the median line T0. This error integral will be noted “Jy.dt”.

[0054] The curvature of the center line is noted p (in m ')• This is the inverse of its radius of curvature at the level of the center of gravity CG of the motor vehicle 10.

[0055] The method for controlling the motor vehicle 10 is designed to allow this vehicle to substantially follow the center line T0 of the traffic lane, in autonomous mode (without driver intervention).

[0056] This method is implemented when the LCA function is activated.

[0057] How to activate this function will not be described here.

[0058] On the other hand, it will be possible to describe how the vehicle is kept in its traffic lane 31 when the LCA function is activated.

[0059] To establish the vehicle control law and thus regulate the steering angle δ of the motor vehicle 10 so that the latter remains centered on its center line T0, this vehicle was modeled using a bicycle model.

[0060] In such a model, the two wheels of the front axle are considered to be merged, and the same applies to the two rear wheels. The chassis of the vehicle is modeled by a body which connects the two wheel models.

[0061] The dynamics of the motor vehicle 10 is then represented by a state vector X, which is expressed here in the form:

[0062] [Math.l] idtpl dt V dy / dt dô / dt 5

[0063] It will be recalled here that the lateral deviation y corresponds to the distance between the center of gravity CG of the vehicle and the center line T0.

[0064] The equation of this median line is in practice determined on the basis of the equations representing the marking lines 33, 35 which delimit this traffic lane 31.

[0065] According to the “bicycle” model used, the equation of the system is written in the following form:

[0066] [Math.2] Y = CX

[0067] In this equation, the term ôFBK is a first component of the steering angle setpoint ô which will be transmitted to the power steering actuator. As will appear more clearly below, this component makes it possible to keep the vehicle in the center of the traffic lane 31 considering that the latter is rectilinear.

[0068] A, C, Bô and Bp are determined matrices and vectors.

[0069] The first of the Math2 equations then introduces two terms, including an open loop term Bp.p and a closed loop term Bô.ôfbk. The open loop term is intended to compensate for the steering angle ô by taking into account the curvature of the center line T0. The closed loop term allows the steering angle to be calculated by considering that the traffic lane is straight.

[0070] Y represents the measurement vector, and it therefore depends on the state X.

[0071] With reference to [Fig.3], the topology of an example of an LCA system is schematically represented by a block diagram.

[0072] This block diagram comprises a closed loop 25 and an open loop 21.

[0073] The open loop 21 has the function of taking into account the curvature of the road and to compensate for the effect of the turn on the states and the command.

[0074] The closed loop 25 has the function of keeping the vehicle in the center of its traffic lane while the latter is considered straight, that is to say rectilinear.

[0075] We therefore find the two terms introduced above.

[0076] The terms resulting from these two loops, namely the components ôFBK and ôFFD, are added together by means of an adder 27.

[0077] The steering angle ô to be transmitted to the front wheel so that the vehicle 1 moves in a bend having a known curvature p thus depends on the two previous components, so that we can write:

[0078] [Math.3] ôreq = 5FBK + ÔFFD

[0079] In [Fig.3], the state of vehicle 1 is represented by element 22. This element therefore represents the vehicle, with its sensors, its actuators, etc. A set of measured data emerges from this element 22.

[0080] The open loop 21 comprises an anticipatory element 24. This anticipatory element 24 takes into account the curvature p of the traffic lane (calculated for example from the images obtained by the camera) in order to evaluate the component ôFFD of the steering angle ô. This open loop is generally known by the Anglo-Saxon term “feed forward”.

[0081] Considering the bicycle model in steady state (with dX / dt = 0), and assuming the vehicle is in the center of its lane (dy / dt = 0 and y=0) and in an established turn (dô / dt=0), we can then write:

[0082] [Math.4] ^FFD~ + VSVV2}

[0083] In this equation, Vsv is the vehicle's own understeer gradient, which is classically defined by the following expression:

[0084] [Math.5]

[0085] In other words, the component ôFFD of the steering angle ô has a term pL which is determined as a function of the curvature of the bend and the architecture of the vehicle, and a term which makes it possible to take into account the drift of the vehicle when bending.

[0086] The closed loop 25 comprises an observer element 26 which makes it possible to observe the state Xobs of the motor vehicle 10.

[0087] It also includes a comparator 28 making it possible to differentiate between a reference state Xref and this observed state Xobs. This difference forms an error XeiT.

[0088] It finally comprises a controller 20 which collects the signal delivered by the comparator 28 and generates the component ôFBK of the steering angle ô.

[0089] This controller 20 in practice comprises a gain Ks in the form of a vector which, once multiplied by the error Xeir, makes it possible to calculate the component ôFBK.

[0090] The state observer 26 collects a measurement vector Ymes (comprising measured values, such as the steering angle) and delivers the observed state Xobs.

[0091] The state representation implemented by the state observer 26 is based on the bicycle model of the vehicle.

[0092] This state observer 26 is used to estimate the unmeasured values ​​of the model parameters. These unmeasured values ​​are for example dy / dt and dô / dt.

[0093] It should be noted here that, as a variant, this observer could be dispensed with if all the values ​​were measured.

[0094] Noting x as the estimate of the state vector X, the observer equation can be written as:

[0095] [Math.6] X = (A-LpC)X + B^fbk + LpY

[0096] with Lp a gain value associated with the observer element 26.

[0097] As a general rule, the control implemented by the closed loop 25 aims to minimize the state vector X around a zero reference state Xref corresponding to a straight line. In other words, the nominal reference values ​​(under nominal driving conditions) are such that one can write:

[0098] [Math.7]

[0099] In summary, when the LCA function is activated and operating in nominal mode, it is intended to measure the variables of the measurement vector Ymes and the curvature p of the traffic lane.

[0100] This curvature makes it possible to calculate the ôFFD component.

[0101] The measured variables make it possible, thanks to the state observer 26, to determine the values ​​Xobs>i of the parameters of the state vector X (also called state variables and noted: rp, drp / dt, y, dy / dt, ô, dô / dt, Jy.dt).

[0102] The vector XeiT making the difference between these observed values ​​Xobs>i and the corresponding nominal reference values ​​Xref.i then makes it possible to determine the component ôFBK, and therefore to deduce the steering angle setpoint ô.

[0103] In [Fig. 1], a section of traffic lane is shown on a bend, and a trajectory T1 which is superimposed on the center line T0.

[0104] When the LCA function is activated and operating in nominal mode (as explained above), the motor vehicle 10 is driven so as to follow this trajectory T1 when turning.

[0105] In practice, this trajectory T1 differs from that generally followed by a driver of a motor vehicle.

[0106] Indeed, drivers of motor vehicles have ways of driving which can vary from one another. There are thus a large number of driving modes when cornering, which cause vehicles to follow different trajectories.

[0107] For simplicity, we propose here to list nine standard driving modes, that is to say nine typical trajectories, defined in the following manner (see [Fig.4]): - the aforementioned median trajectory Tl, - the ideal trajectory T2 having a constant radius of curvature, - the left shift trajectory T3, used in a straight line to shift the vehicle to the left of its lane, - the right shift trajectory T4, used in a straight line to shift the vehicle to the right of its lane, - the sport trajectory T5, in which the steering is at its maximum before the bend, - the emergency trajectory T6 in which the vehicle brakes in a straight line before turning sharply and suddenly after the bend, - the outer trajectory T7 in which the vehicle follows the left marking line in a right turn (and vice versa), - the inner trajectory T8 in which the vehicle follows the right-hand marking line in a right turn (and vice versa), - the instinctive trajectory T9 in which the vehicle dives into the bend and steers sharply after the bend.

[0108] The invention then proposes to program the computer 11 so that it can recognize what type of trajectory the vehicle follows when the LCA function is inactive and the driver is driving the vehicle (from the list of trajectories given above), then it can steer the vehicle in a bend using the type of trajectory used by the driver previously so as not to change the habits felt by the latter.

[0109] To recognize the driver's preferred trajectory type, the invention proposes to rely on a neural network.

[0110] To configure the neural network so that it provides satisfactory data, it is planned to implement three successive steps: - a step of developing a database, - a step of learning the neural network using this database, and - a saving of the neural network in the calculator 11.

[0111] In the remainder of this presentation, we will first describe how the database will be developed, then how the neural network will be trained, and finally how the computer will be used in practice to determine the type of trajectory taken when turning by the driver of the vehicle.

[0112] The first step firstly comprises a simulation operation which consists, for a natural person hereinafter called “operator”, of driving a test vehicle on several bends, then a second operation of storing in the database the data collected during the first operation.

[0113] The maneuvers could be performed on a real motor vehicle. However, for ecological, cost and precision reasons, these maneuvers could also be performed virtually.

[0114] During all maneuvers, trajectory data is recorded.

[0115] In practice, this data includes: - a time indicator t,

[0116] - a lateral distance between the center of gravity of the motor vehicle 10 and the line left edge 33, - a lateral distance between the center of gravity of the motor vehicle 10 and the right edge line 35,

[0117] - the curvature of the left edge line 33, at the level of the center of gravity CG of the motor vehicle 10 (expressed in m1),

[0118] - the curvature of the straight edge line 35, at the level of the center of gravity CG of the motor vehicle 10 (expressed in m1),

[0119] - the speed v of the motor vehicle 10, - the lateral acceleration (along the Y axis) of the motor vehicle 10,

[0120] - the distance between the center of gravity CG of the motor vehicle 10 and the point further from the left edge line 33 detected by the camera,

[0121] - the distance between the center of gravity CG of the motor vehicle 10 and the point further from the right edge line 35 detected by the camera.

[0122] These trajectory data will be used as input data for the neural network. They are considered to allow the selection of a type of trajectory used in each bend. For example, it will be noted that the distance between the vehicle and the most distant detectable point from a lane edge line will be shorter the tighter the bend.

[0123] These data are recorded at regular time steps, for example every hundredth of a second.

[0124] For each time step, the operator associates with this input data a data of exit, namely the type of trajectory taken, chosen from among the nine types of trajectory illustrated in [Fig.4].

[0125] This data is then recorded in a database comprising as many records (i.e. lines) as there will be time steps recorded, and comprising ten fields (i.e. ten columns) corresponding to the nine input data and the output data.

[0126] It would of course be possible to enrich the database with other fields, but here the nine input data are considered sufficient to obtain reliable results.

[0127] Once the database has been obtained, the second learning step consists of providing the database to the neural network so that it can train itself to select the type of trajectory taken by the driver when turning, taking into account the aforementioned input data.

[0128] The neural network used is preferably of the recurrent type. Thus, it determines the type of trajectory taken based on the input data measured not at a single time step but at several time steps.

[0129] To give it this recurrent character, the neural network then has ten inputs, namely the nine aforementioned input data and the type of trajectory selected at the previous time step.

[0130] We could use a network of the MTSF - LSTM type (from the English “Multivariate Time Series Forecasting - Long Short-Term Memory”).

[0131] However, to reduce the calculations without harming the reliability of the results, we will instead use a “GRU” (from the English “Gated Recourent Unit”) type algorithm here.

[0132] The input layer of the neural network comprises ten artificial neurons receiving the ten input data.

[0133] The output layer here comprises a single artificial neuron which provides the output data, namely the type of trajectory recognized.

[0134] In this case, this neural network will have two hidden layers.

[0135] The first hidden layer here comprises one hundred artificial neurons which integrate an internal memory mechanism and an update of their settings. This recurrent network architecture makes it possible to reduce the effects of gradient disappearance on the lower layers and therefore the slowdown of learning.

[0136] The second hidden layer, called densification, here comprises a single artificial neuron which receives the data from the hundred aforementioned neurons.

[0137] The output layer, which gives the trajectory type, integrates a “softmax” type function, allowing the results to be normalized. Here, it produces the following equation:

[0138] [Math.8] 7 .... ew<

[0139] where Zi is the output, and Wi is the i element of the input vector W, and Wk is the k element of the input vector W.

[0140] The neural network learning step then makes it possible to configure the different artificial neurons, so that the neural network can determine the type of trajectory taking into account the input data measured in real time.

[0141] This step will not be described in detail here. It can only be specified that the learning algorithm will preferably be of the “Adam” type, which is an extension of the stochastic gradient method. This method indeed presents, in the context of the invention, a high speed of convergence. It uses a cost function which is the average of the errors in absolute value, called here MAE and defined by the following equation:

[0142] [Math.9] MAE = ±yn Iv.-yl - d

[0143] Where n is the number of points used. More precisely, it is the number of points on which we want to calculate an MAE average, that is to say the number of predictions whose precision we want to measure (i.e. for example the set of values ​​predicted with the training samples).

[0144] y. is the expected value previously obtained in a training data set, and y; the output value of the neural network.

[0145] Once the neural network is properly configured, this neural network is implemented in the computer of the motor vehicle 10.

[0146] It may be noted that before this, it will be possible to verify that the neural network is working well by establishing a second database of the same type as the first but carried out on other turns, and by verifying that the results obtained are reliable.

[0147] The least squares method can be used to rate the neural network in order to check whether it is sufficiently reliable or not.

[0148] Then, when a driver is at the wheel of the motor vehicle 10 and is driving along bends, the computer 11 implements the steps illustrated in [Fig.5].

[0149] During the first step El, the LC A function is considered to be on standby so that the driver remains in control of the maneuver.

[0150] The LCA function is at this stage parameterized in nominal mode (the reference values ​​of the state variables are notably zero, as described by the equation Math.7).

[0151] At this stage, it is planned that the computer 11 stores the input data at each time step.

[0152] On the basis of this data and / or other information, the computer 11 is then able to detect, during a step E2, whether the motor vehicle 100 is entering a bend or starting a shifting maneuver in its lane (in order to shift from the center line to the right or left of its traffic lane, without leaving it).

[0153] In this eventuality, during a step E3, the computer 11 uses the neural network, supplied by the measured input data in order to determine the type of trajectory used on the bend taken by the driver.

[0154] This type of trajectory (sport, median, etc.) is associated with a “target trajectory” illustrated in [Fig. 4]. The computer 11 can then determine, during a step E4, a lateral deviation between the center of gravity CG of the motor vehicle 100 and the target trajectory (along the Y axis).

[0155] It will be noted that here, this target trajectory is not defined with respect to physical quantities but is defined during the labeling step of the data used to train the neural network. It is nevertheless possible to compare the real trajectory and the target trajectory, by comparing the real measurements of curvature, lateral deviation and heading angle with the control law parameterization associated with this target trajectory, via an open-loop simulation for example.

[0156] If this deviation remains below a threshold over several time steps or with an adjusted duration delay (typically 0.5 seconds), the process continues directly with step E6 described below. Otherwise, the computer determines whether the vehicle is at the end of the turn (step E5).

[0157] If the vehicle has not reached the end of the turn, the process loops to step E3. Otherwise, it continues with step E6.

[0158] During this step E6, the computer considers the last type of trajectory selected. Indeed, thanks to the above looping, if the type of trajectory detected by the neural network changes during the turn, it will be the last one which will then be considered as being the correct one.

[0159] From then on, the computer can determine whether or not it is necessary to modify the parameterization of the LCA function, and how, taking into account the type of trajectory selected. How this parameterization is carried out will be described later.

[0160] During step E7, the computer checks whether the vehicle is in a straight line. As long as this is not the case, the process loops to step E6 in order to wait for the next straight line. The idea here is not to change the setting of the LCA function when cornering, but to wait until the vehicle is on a straight line to do so.

[0161] When this is the case, the computer checks whether the mission is finished (step E8).

[0162] If so, the process starts again at step E1. Otherwise, during a final step E9, the computer updates the LCA function settings so that, if activated by the driver, this function is correctly configured to approach the next bend. The process then loops to the second step E2.

[0163] In other words, if the mission is finished, we want the process to go back through the mission start and LCA function initialization step to its initial settings, to ensure that the vehicle always starts a mission with the same settings, so that the client has a reference point.

[0164] If the mission is not finished, we consider that the learned turn type must be injected into the control law parameters, which is why we update the LCA function settings.

[0165] It should be noted in this regard that a mission starts when the engine is started and stops when the engine is completely stopped (excluding “start and stop”).

[0166] To parameterize the LCA function, it was chosen, for simplification, to simply modify at least one of the reference values ​​introduced in the Math.7 equation and / or to modify the gain applied to at least one of the state variables % drp / dt, y, dy / dt, ô, dô / dt, Jy.dt. As a variant or in addition, one could also have modified the value of the underturn gradient Vsv used in the Math.4 equation

[0167] Here, depending on the type of trajectory selected, it was chosen to modify the gain or the reference value of one or other of the three state variables which are: - the heading angle rp, - the lateral deviation y, and - the integral of the lateral deviation Jy.dt.

[0168] To modify the gain of one of these variables, it is sufficient to modify variables of the matrix A of the equation Math.2.

[0169] More specifically, the following criteria can be used to modify the setting of the LCA function.

[0170] If the selected trajectory is of the “median trajectory Tl” type, the parameter setting does not change (the reference values ​​remain zero, and the gain matrix A remains unchanged compared to the nominal mode).

[0171] If the selected trajectory is of the “ideal trajectory T2” type, the gain applied to the three aforementioned state variables (rp, y, Jy.dt) is increased by a first percentage, here 10%.

[0172] If the selected trajectory is of the type “left offset trajectory T3”, the reference lateral deviation yref is made non-zero. Its value is preferably equal to half the difference between the width of the traffic lane and the width of the vehicle.

[0173] If the selected trajectory is of the type “right offset trajectory T4”, the reference lateral deviation yref is made non-zero. Its value is preferably equal to half the difference between the width of the vehicle and the width of the traffic lane.

[0174] If the selected trajectory is of the “T5 sport trajectory” type, the gain applied to the three aforementioned state variables (ip, y, Jy.dt) is increased by a second percentage strictly greater than the first, here 20%.

[0175] If the selected trajectory is of the “T6 emergency trajectory” type, the gain applied to the three aforementioned state variables (rp, y, Jy.dt) is increased by a third percentage strictly greater than the second, here 30%.

[0176] If the selected trajectory is of the “external trajectory T7” type, the gain applied to the three aforementioned state variables (rp, y, Jy.dt) is increased by a fourth percentage strictly lower than the first, here 5%.

[0177] If the selected trajectory is of the “inner trajectory T8” type, the gain applied to the three aforementioned state variables (rp, y, Jy.dt) is increased by a fifth percentage strictly lower than the first, here 5%.

[0178] If the selected trajectory is of the “instinctive trajectory T9” type, the parameter setting does not change (the reference values ​​remain zero, and the gain matrix A remains unchanged compared to the nominal mode).

[0179] As described above, in step E4, it is planned to determine the difference between the trajectory of the motor vehicle 10 and the target trajectory (that corresponding to the type of trajectory selected).

[0180] It could be predicted that, if the gap is too large, no type of trajectory is selected.

[0181] It will also be possible to provide for constructing a new type of trajectory to be added to the list illustrated in [Fig.4], when it is detected that the driver regularly takes trajectories which do not correspond to any of those in the list.

[0182] It could thus be provided, for example, that when the lateral difference between the trajectory of the vehicle and each trajectory illustrated in [Fig.4] is greater than a threshold, typically 10 cm, the computer stores in its memory the input data recorded along the bend.

[0183] From then on, in after-sales service, it will be possible to reconstruct each trajectory used by the driver in order to determine whether he regularly follows a trajectory different from the types of trajectory stored in the computer (those of [Fig.4]). For this purpose, a clustering algorithm typically of the “K-means” type will be used to carry out this classification.

[0184] If this is the case, it will be possible to retrain the neural network by adding new records corresponding to the input data recorded along the turns in which the driver used this type of trajectory.

[0185] Thus, the list can be enriched with a new type of trajectory so that, during subsequent driving, the vehicle can follow, when the LCA function is activated, this new type of trajectory.

[0186] The present invention is in no way limited to the embodiment described and shown, but those skilled in the art will be able to provide any variant in accordance with the invention.

Claims

Claims

1. Method for controlling a motor vehicle (10) traveling on a traffic lane (31) of a road (30), comprising: H when an automatic centering function of the motor vehicle (10) in said traffic lane (31) is not activated and the motor vehicle (10) is controlled by a driver (40), a step of acquiring data relating to the trajectory of the motor vehicle (10) when turning relative to edge lines (33, 35) of said traffic lane (31) and / or relative to the edge lines (33, 35), H when said automatic centering function is activated, steps of: - determining a center line (T0) of said traffic lane (31), - calculating, by a computer (11) on board said motor vehicle (10), a control instruction for the motor vehicle (10) as a function of said center line (T0) and determined parameters,and - control by the computer (11) of a steering actuator of said motor vehicle (10) according to said control instruction, characterized in that: - provision is made to select, by means of a neural network of the recurrent type, a cornering driving mode as a function of said data acquired when the motor vehicle (10) is controlled by the driver (40), and in that - when the automatic centering function of the motor vehicle (10) is activated, at least part of the parameters are determined as a function of the cornering driving mode selected.,

2. A driving method according to claim 1, wherein said neural network is of the GRU type.

3. A driving method according to one of claims 1 and 2, in which a prior operation of parameterizing the computer (11) is provided, comprising: - a step of developing a database associating test trajectories with cornering driving modes, - a step of learning the neural network using said database, and - saving the neural network in the computer (11).

4. A control method according to one of claims 1 to 3, in which said data are chosen from at least: - a time indicator (t), - a distance between the motor vehicle (10) and at least one of the edge lines (33, 35), - a curvature datum of at least one of the edge lines (33, 35), - a datum relating to the dynamics of the motor vehicle (10), - a distance between the motor vehicle (10) and the detected point of at least one of the edge lines (33, 35) which is the furthest from the motor vehicle (10).

5. Driving method according to one of claims 1 to 4, in which the cornering driving mode is selected based at least on the data acquired in the last corner taken by the motor vehicle (10) before activation of the automatic centering function of the motor vehicle (10).

6. Driving method according to one of claims 1 to 5, in which the driving mode is selected from a list, from at least five predetermined and distinct cornering driving modes, and preferably from at least nine cornering driving modes.

7. Driving method according to claim 6, in which, if at least in one bend, the trajectory of the motor vehicle (10) did not correspond to any of the bend driving modes in said list, provision is made in after-sales service to add to said list at least one new bend driving mode predetermined and distinct from the other bend driving modes.

8. Piloting method according to one of claims 1 to 7, wherein said parameter is a reference value to be reached and / or a gain associated with: - a lateral deviation (y) between the motor vehicle (10) and the center line (T0), and / or - a heading angle (W) between the motor vehicle (10) and the center line (T0), and / or - a position error integral (Jy.dt) between the motor vehicle (10) and the center line (T0).

9. Motor vehicle (10) comprising: - acquisition means adapted to acquire, when the motor vehicle (10) is moving on a traffic lane (31), data relating to edge lines (33, 35) of said traffic lane (31), and - a steering actuator adapted to control the motor vehicle (10), characterized in that it further comprises a computer (11) programmed to implement a control method in accordance with one of claims 1 to 8.

Images