MANUFACTURING A VEHICLE TRAJECTORY CONTROL DEVICE BY AUTOMATIC LEARNING AND SYNCHRONIZATION OF WHEEL ROTATION SPEED MEASUREMENTS

Abstract

The invention relates to a manufacturing method (100) for an electronic trajectory control device for a vehicle. The manufacturing method (100) comprises synchronizing measurements of the rotational speeds of each of the vehicle's wheels at initial times. The manufacturing method (100) comprises determining a first set of trajectory parameters at these initial times. The manufacturing method (100) comprises developing (101) candidate programs for the electronic trajectory control device by machine learning from the first set of parameters and the synchronized measurements of the rotational speeds of each of the wheels. (Fig. 1)

Description

Title of the invention: MANUFACTURE OF A VEHICLE TRAJECTORY CONTROL DEVICE BY AUTOMATIC LEARNING AND SYNCHRONIZATION OF MEASUREMENTS WHEEL ROTATION SPEED Technical field of the invention

[0001] The invention relates to the braking of a motorized road vehicle, such as a car or a van. More specifically, the invention concerns the manufacture of an electronic device for controlling the trajectory of a motorized road vehicle, in particular by synchronizing wheel speed measurements of the vehicle. Prior art

[0002] A motorized road vehicle includes brakes, at least one brake control unit and an electronic trajectory control device which is configured to control the trajectory of the motorized vehicle, in particular by differential braking between the wheels of the vehicle.

[0003] A motorized road vehicle, which is known from patent application FR 3088247, includes at least one braking control device using fuzzy logic.

[0004] A known electronic trajectory control device is generally programmed entirely by experts and computer scientists. Since the trajectory control of a motorized road vehicle involves a significant risk to the safety of the driver and passengers, the quality and reliability of the programming of the electronic trajectory control device are verified.

[0005] However, it is difficult to take into account the circumstances actually encountered by the vehicle, and many parameters whose influence is difficult to assess are not taken into account in the development of an electronically controlled trajectory.

[0006] The electronic trajectory control of a motorized road vehicle needs to be improved, particularly by increasing the safety and / or comfort of the driver and passengers. There is a need for electronic trajectory control that is reliable and robust, while also being precise. Description of the invention

[0007] The invention aims to remedy all or part of the drawbacks of the prior art. In this respect, the invention relates to a method for manufacturing a Electronic trajectory control device for a motorized road vehicle. The vehicle includes wheels, rotation speed sensors for each of the vehicle's wheels, brakes for each wheel, and an electronic trajectory control device that is configured to control the vehicle's trajectory.

[0008] According to the invention, the manufacturing process comprises synchronizing measurements of the rotational speeds of each of the vehicle's wheels at initial times, and determining a first set of vehicle trajectory parameters at those initial times. The manufacturing process includes developing candidate programs for an electronic trajectory control device by machine learning from the first set of parameters and the synchronized measurements of the rotational speeds of each of the wheels.

[0009] Thanks to synchronized measurements of wheel rotation speeds and machine learning, the manufacturing process according to the invention is capable of improving the electronic trajectory control of a motorized road vehicle. For example, the safety and comfort of the driver and passengers tend to increase. The manufacturing process facilitates the production of a reliable, robust, and precise electronic vehicle trajectory control device.

[0010] In particular, synchronized wheel speed measurements coupled with machine learning, advantageously supervised, tend to facilitate the manufacture of a reliable, robust, and precise electronic trajectory control device, by reducing software development time. Synchronized wheel speed measurements and machine learning promote the specialization and optimization of the trajectory control device for each type of vehicle.

[0011] The invention uses machine learning, particularly with a neural network, to develop candidate programs for an electronic trajectory control device. The use of machine learning to manufacture an electronic trajectory control device may seem counterintuitive at first glance, since the resulting electronic trajectory control program is difficult to analyze. In the manufacturing process as claimed, the quality and reliability of the trajectory control program are advantageously ensured by validation, for example, by a decision algorithm and / or by human supervision.

[0012] The trajectory control program of the electronic trajectory control device is improved compared to conventional control programs, notably by enabling reliable, robust, and precise trajectory control of the vehicle. The trajectory control program manufactured according to the invention tends to better take into account the actual circumstances encountered by the vehicle. and external to the vehicle. In particular, the trajectory control program strives to appropriately take into account a very large number of parameters that participate in the trajectory and whose combined influence on the vehicle's trajectory is difficult to assess.

[0013] According to one embodiment, the manufacturing process includes comparing the rotational speed measurements of the synchronized wheels at each of the first instants. Comparing the wheel rotational speeds improves trajectory control, particularly its robustness and reliability.

[0014] According to one embodiment, synchronized measurements of wheel rotation speeds are compared to detect slippage of at least one of the vehicle's wheels and / or to determine a deviation between a trajectory desired by the driver and an actual trajectory of the vehicle.

[0015] According to another embodiment, the manufacturing process includes the development of candidate programs for electronic trajectory control devices by deep learning, in particular by neural network.

[0016] According to another embodiment, the manufacturing process includes the development by fuzzy logic of candidate programs for an electronic trajectory control device.

[0017] One advantage of deep learning and / or fuzzy logic is that they lend themselves well to machine learning techniques by taking into account many parameters and that the results of validation or rejection of candidate programs make it possible to use machine learning to develop these candidate programs without compromising the safety of the driver and passengers.

[0018] According to one embodiment, the first set of parameters includes vehicle model parameters, vehicle state, vehicle movement, external vehicle environment, brakes, braking control, driving and / or braking history, driving or braking styles, allocation or distribution of energy resources in the vehicle.

[0019] According to another embodiment, the first set of parameters includes a vehicle mass, a vehicle age, a vehicle speed, a vehicle yaw rate, a vehicle wheel orientation command, a vehicle acceleration, a vehicle positioning, a parameter representing wheel grip on a road, a braking force of at least one of the vehicle's brakes, a current passing through at least one of the brakes when the brake is electromechanical, and / or a current of a vehicle propulsion system when the vehicle is electric or hybrid.

[0020] According to one embodiment, the manufacturing process includes validating or rejecting candidate programs by applying at least one decision algorithm, typically through an algorithmic validation device. In addition, or alternatively, the manufacturing process includes validating or rejecting candidate programs through human supervision.

[0021] According to one embodiment, the decision algorithm includes an analysis algorithm and / or an expert system.

[0022] According to another embodiment, the decision algorithm includes a check of regulatory, technical and / or driver-related requirements.

[0023] According to one embodiment, the decision algorithm is configured to detect vehicle problems that are at least partially independent of trajectory control and that appear after the application of candidate programs.

[0024] According to another embodiment, the decision algorithm includes a testing algorithm configured to analyze the results of applying candidate programs to trajectory control situations expressed by a second set of parameters. The candidate electronic trajectory control program tends to be all the more reliable, robust, and precise.

[0025] Preferably, the decision algorithm is configured to vary the parameters of the second set of parameters in decision ranges to implement the trial algorithm.

[0026] According to one embodiment, the manufacturing process includes a synchronization of rotation speed measurements of each of the vehicle's wheels at second instants, and the determination of the second set of vehicle trajectory parameters at the second instants.

[0027] According to another embodiment, the decision algorithm includes a self-learning step configured to re-evaluate previously validated candidate programs, in the event of rejection of a candidate program deemed to be similar to the previously validated candidate programs.

[0028] According to one embodiment, the manufacturing process includes updating the candidate program after installation in a computer of the vehicle's electronic trajectory control device, in particular by continuing the automatic learning of the candidate program.

[0029] According to another embodiment, the manufacturing process includes updating a program of the electronic trajectory control device during vehicle maintenance operations.

[0030] The invention also relates to a manufacturing system for an electronic trajectory control device for a motorized road vehicle. The manufacturing system implements a manufacturing process as defined above for develop a trajectory control program for the electronic trajectory control device.

[0031] The invention also relates to a motorized road vehicle comprising wheels, rotational speed sensors for each wheel of the vehicle, brakes for each wheel, and an electronic trajectory control device configured to control the vehicle's trajectory. The electronic trajectory control device is manufactured by a manufacturing process as defined above.

[0032] According to one embodiment, the electronic trajectory control device is configured to minimize wheel slippage and to generate braking commands for the wheels by the brakes, allowing the vehicle to follow a desired trajectory determined from a vehicle wheel orientation command. The vehicle wheel orientation command corresponds, in particular, to a steering wheel angle measurement. Brief description of the figures

[0033] The present invention will be better understood upon reading the description of non-limiting examples of embodiments, with reference to the accompanying figures, which illustrate: • [Fig.l]: a method for manufacturing an electronic trajectory control device according to a first embodiment; • [Fig. 2]: a schematic representation of a neural network for the implementation of the manufacturing process according to the first embodiment; • [Fig. 3]: Implementation of program validation algorithms candidate; • [Fig. 4]: illustrates the implementation of an analysis algorithm or a algorithm based on an expert system for the validation of a candidate program; • [Fig. 5]: The operation of an expert system for the validation of a candidate program; • [Fig. 6]: a verification of critical situations for the validation of a candidate program; • [Fig. 7]: Machine learning in case of program rejection candidate; • [Fig. 8]: A schematic representation of a motorized road vehicle including an electronic trajectory control device which was manufactured by the manufacturing process according to the first embodiment; • [Fig.9]: a schematic representation of a manufacturing system for an electronic trajectory control device, implementing the manufacturing process according to the first embodiment; • [Fig. 10]: a schematic representation of a neural network for the implementation of the manufacturing process according to a second embodiment. DETAILED description of AT LEAST ONE embodiment

[0034] For clarity, identical elements are identified by identical reference signs from one figure to another.

[0035] Fig. 1 illustrates a manufacturing method 100 for an electronic trajectory control device 110. The manufacturing method 100 according to the first embodiment or according to the second embodiment comprises three steps.

[0036] The first step 101 is a step of developing candidate programs y, y', y'' for an electronic trajectory control device 110 using a machine learning algorithm, based in particular on the rotational speed of wheels 119 that are synchronized at initial times and on vehicle trajectory parameters x that are determined at initial times. The first step 101 is carried out in the laboratory, in a factory, and / or on the road. The first step 101 is the main step of the manufacturing process 100 according to the invention.

[0037] The second step 102 is a validation step of at least one candidate program y, y', y" which was developed during the first step 101. The validation of candidate programs y, y', y" is carried out in particular by human supervision 38 or by analysis algorithms 7, tests 8 and / or expert systems 9 such as those represented in [Fig.3].

[0038] The third step 103 is an implementation of a candidate program validated in the second step 102 in a computer 109 of an electronic trajectory control device 110 for a motorized road vehicle 120, which is shown in [Fig. 8]. The program of the electronic trajectory control device 110 is capable of being updated. It is preferably specific to the type of vehicle 120 in which it is installed, and is optimized for each series of vehicles 120.

[0039] With reference to [Fig. 8] and [Fig. 9], the first step 101 is carried out by a candidate program manufacturing device 104, which is equipped with a computer such as a processor. The candidate program manufacturing device 104 is configured to perform machine learning operations. This machine learning is, for example, supervised, unsupervised, and / or reinforcement learning. The machine learning is, in particular, partially based on logic and knowledge, for example, on an expert system 9 or algorithms. of analyses 7 such as those represented in [Fig. 3], and / or on a statistical approach such as fuzzy logic, Bayesian estimation, search and optimization trees. The manufacturing device 104 preferably develops the candidate programs y, y', y" by deep learning and / or fuzzy logic.

[0040] The validation or rejection results of candidate programs y, y', y” by the manufacturing device 104 are transmitted to a trajectory control program storage device 105. The storage device 105 contains a database 106 of potential candidate programs y, y', y” and a database 6 of validated candidate programs.

[0041] The candidate program y, recorded in the candidate program database 106, is transmitted to an algorithmic validation device 107, which includes at least one computer that is, for example, shared with the manufacturing device 104. The algorithmic validation device 107 performs algorithmic calculations to validate the candidate programs y, y', y”, for example, using analysis algorithms 7, testing algorithms 8, and / or expert systems 9 such as those shown in [Fig. 3]. In addition, or alternatively, at least some candidate programs y, y', y”, may be validated by human supervision 38, for example, by a team of computer scientists and experts. The validated programs y, y', y”, are recorded in the database 6. The contents of the database 6 are then, for example, written to a memory 108 of a computer 109 of an electronic trajectory control device 110.The candidate program manufacturing device 104, the storage device 105, and possibly the algorithmic validation device 107, together form a manufacturing system for an electronic vehicle trajectory control device.

[0042] The electronic trajectory control device 110 includes the memory 108 and the computer 109 in which is implemented at least one trajectory control program which has been validated by the manufacturing process 100. The electronic trajectory control device 110 is configured to control the trajectory of the vehicle 120, in particular by differential braking of the wheels 119 of the vehicle 120 and / or by modifying an engine torque which is generated by a propulsion system 121 of the vehicle 120. The electronic trajectory control device 110 is in particular configured to individually control each wheel 119 of the vehicle.

[0043] The electronic trajectory control device 110 is configured to compare synchronized measurements of wheel rotation speeds to detect slippage of at least one of the vehicle's wheels. The electronic trajectory control device 110 is configured to determine a deviation between a trajectory desired by the driver and the actual trajectory of the vehicle.

[0044] The electronic trajectory control device 110 is configured for minimize the slip of each of the wheels 119. It is configured to generate wheel braking commands by the brakes 111 by allowing the vehicle 120 to follow a desired trajectory which is determined from a wheel orientation command a of the vehicle which corresponds in particular to a measurement of steering wheel angle or steering column.

[0045] With reference in particular to [Fig. 8], the motorized road vehicle 120 is typically a mass-produced automobile. The vehicle 120 comprises a steering wheel 123, the electronic trajectory control device 110, a braking control device 122, a propulsion system 121, and preferably four wheels 119, each equipped with a brake 111 and sensors 118. When the vehicle 120 is electric or hybrid, the propulsion system 121 is also configured to brake the vehicle 120 by regenerative braking. In this case, the electronic trajectory control device 110 is specifically configured to control regenerative braking by the propulsion system 121.

[0046] The sensors 118 include wheel speed sensors 119, for example, Hall effect sensors or more precise special sensors such as tachometers. Each tachometer is, for example, mechanical, optical, and / or eddy current. The wheel speeds are measured synchronously at the first instants instead of being interpolated. The vehicle 120 also preferably includes other sensors such as an overall vehicle speed sensor v, a yaw rate sensor, an acceleration sensor, a wheel orientation sensor a, a steering wheel angle sensor, or a steering column angle sensor, and / or a braking force sensor. The signals from the sensors 118 are transmitted to the electronic trajectory control device 110 and / or the braking control device 122.

[0047] The brake control device 122 includes a central computer and / or local electronic control units which are part of the brakes 111 of the vehicle 120. The brake control device 122 and the brakes 111 together form a braking system 112 of the vehicle.

[0048] The central computer is for example the computer 109 which is common to the braking control device 122 and the electronic trajectory control device 110. In addition or alternatively, the electronic trajectory control device 110 is decentralized between the local electronic control units which each form part of one of the brakes 111.

[0049] Each brake 111 comprises a control portion 114, as well as a mechanical portion 116 and / or a hydraulic portion. Each brake 111 is configured to brake one of the wheels of the vehicle 120, in service braking, in parking braking and / or during safety braking. The control unit 114 includes, for example, at least one microcontroller and / or a microprocessor. The control unit 114 is controlled by the electronic trajectory control device 110 or by the central control unit of the braking control device 122, during normal vehicle operation 120.

[0050] The control part 114 of these brakes 111 is, for example, configured to control the mechanical part 116 at least partially independently of the computer 109 in the event of a failure of the computer 109. The candidate program y, y', y" which has been validated by the manufacturing process 100 is then, in particular, at least partially implemented in the control part 114. In this case, the control part 114 includes, for example, a microprocessor serving as a computer.

[0051] Preferably, at least two of the brakes 111 located on the same axle of the vehicle are electromechanical brakes, each comprising a geared motor, i.e., a mechanical part 116, and a local electronic control unit. At least one sensor 118 measures the electrical current and / or power supplying the electromechanical brake 111. When the propulsion system 121 is electric or hybrid, the sensor 118 also detects the direction of the current and / or electrical power between the electromechanical brake 111 and the propulsion system 121, in order to detect any regenerative braking or acceleration.

[0052] The first step 101 of developing candidate programs y, y', y" is preferably implemented by an artificial neural network 1 which is described with reference to [Fig. 2]. Parameters from a first set, taking values ​​denoted xl to x9 and denoted x, are provided to neurons 2 of an input layer 3 of the network 1. The neurons 2 are, in general, logic processors capable of processing the input signals they receive and transmitting output signals corresponding to the logic processes to other neurons 2, belonging in principle to subsequent layers of the network 1, called intermediate layers 4, and which combine the output signals of the neurons 2 of the preceding layers in various ways.Typically, but not exclusively, the processes performed by the neurons 2 of input layer 3 may include weighting by numerical coefficients or discrimination based on threshold values, and the combinations may include logical operations (AND, OR, etc.), numerical combinations, or others. The layers 4 of network 1 gradually shape the candidate programs y. Neural networks 1 can have as many intermediate layers 4 of neurons 2 as necessary, which combine the output signals of the preceding layers according to the same principle, for example, between 3 and 1,000 layers, preferably between 30 and 100 layers, for example, 80 layers. The number of layers depends on the available computing power and the complexity of the data to be processed. When it has a high number of layers, the neural network 1 is said to be deep learning.

[0053] In each of the embodiments shown, the neural network 1 comprises an output layer 5 containing as many neurons 2 as there are wheels 119, each representing the behavior of one of the vehicle's wheels. More generally, the neural network 1 has as many neurons 2 in the output layer 5 as there are brakes 111 on the vehicle, and each neuron represents the behavior of one of the brakes 111. The neural network 1 develops a candidate program y for the electronic trajectory control device 110.

[0054] The first set of parameters x includes parameters xl of the vehicle model, such as mass, dimensions, wheel characteristics, etc. The first set includes parameters x2 of the brake model or its nature, for example, front or rear wheel, service, parking, and / or safety braking. The first set includes parameters x3 of the braking control, and in particular of the actuation of a parking brake lever or button or a brake pedal, such as the application force, abruptness, and braking duration. The braking control parameters x3 include, for example, measurements of the intensity and / or power passing through the brakes 111 of each wheel. The braking control parameters x3 include, for example, measurements of the intensity and / or power of the drive system 121. The first set includes parameters x4 of vehicle motion, including the rotational speeds of each wheel.The rotational speed measurements of each wheel are synchronized with each other and preferably with those of the other parameters x1 to x3 and x5 to x9. The rotational speeds of each wheel are compared with each other at each initial instant. The vehicle motion parameters x4 also include the overall speed v of the vehicle, its acceleration, its yaw rate, a steering direction setpoint a for the vehicle's wheels which is determined in particular by a measurement of the steering wheel angle and / or the steering column, etc.

[0055] The first set includes x5 external environment parameters that provide information about the environment external to the vehicle or its interactions with the vehicle. The x5 external environment parameters include, for example, the outside temperature, wheel grip on the road, vehicle positioning 120, the positioning of a nearby vehicle, road traffic density, images from an external camera, etc. The first set includes x6 parameters from any sensor 118 reflecting the vehicle's condition, such as tire pressure, tire wear, tire type, brake fluid level, brake wear status, battery voltage or charge, etc. The first set includes x7 parameters for allocating and managing immediately available energy resources for to control the vehicle, possibly modifying them to favor trajectory control, which will notably take priority over other systems, such as damping, oil pressure allocation, etc. When the vehicle 120 is electric or hybrid, the x7 parameters for energy resource distribution and management are used with reference to the brakes 111 and the propulsion system 121.

[0056] The first set includes x8 parameters of driving, trajectory, or braking history to take into account past events such as immediately preceding braking or other driving parameters that may provide an indication of the road conditions. The x8 parameters of driving, trajectory, or braking history provide information, for example, on the appropriateness of anti-lock braking or traction control by the electronic trajectory control device 110. The first set includes x9 parameters of braking or driving style, which are preferably classified, for example, into different types of trajectories that are presented to the neural network 1 to benefit from its self-learning capabilities. These x9 parameters of braking or driving style take into account, for example, the driver's driving style, expertise, and responsiveness.

[0057] In general, all conceivable relevant parameters x can be introduced into the neural network 1 and taken into account in various very different ways to develop candidate programs y for vehicle trajectory control. All parameters x are, for example, determined synchronously, by measurement or interpolation, at initial times that are, for example, ten milliseconds apart. Analogous parameters x that were determined at different initial times are, for example, compared with each other in the first step 101.

[0058] The second step 102 of the manufacturing process 100, shown in [Fig. 1], relates to the validation of the candidate programs obtained in the first step 101. Like the first step 101, it is carried out in the factory, in the laboratory, or on the road. There are various ways to validate or reject each candidate program obtained previously. These options can also be combined, as shown in [Fig. 3].

[0059] The validation methods are either human or algorithmic. Algorithmic validation methods include, for example, analysis algorithms 7, testing algorithms 8, and / or expert system algorithms 9. The results of these different algorithms are provided to a decision module 10, which can produce a validation decision 11 for candidate program y, in principle if the results of all the validation algorithms 7, 8, and 9 are favorable, or, conversely, a rejection decision 13 for candidate program y. The latter may, however, be preceded by a correction attempt step 12 for the candidate program y, which then re-enters the system. The corrected program form y' is at least as determined by validation algorithms 7, 8, and 9, which had rejected the candidate program y before correction. In case of a further failure by decision module 10, the rejection 13 of the candidate program y, y' is final.

[0060] With reference to [Fig. 7], self-learning 14 is likely to be performed in the event of rejection of the candidate program y. Candidate programs, which have been previously validated and stored in database 6 and which are analogous to the rejected candidate program y, are searched for by mathematical similarity or distance calculations according to a distance evaluation step 15. A subset of candidate programs with sufficient similarity to the rejected candidate program y is extracted 16 from database 6. The programs in this subset are again subjected to the validation algorithms 7, 8, and 9 in a validation submission step 17.Candidate programs are maintained in database 6 if validation is renewed at a maintenance stage 18, or they are removed from database 6 if rejected at a database deletion stage 19.

[0061] Figure 4 illustrates an analysis algorithm 7 for validating candidate programs y. The candidate program y is checked successively according to various criteria, such as regulatory requirements or standards in step 20, technical requirements for compatibility with the braking system in step 21, or requirements for interaction with other vehicle systems such as resource allocation in step 22, and customer requirements or expectations, for example regarding the desired braking style, in step 23. If the candidate program y passes these checks successfully, it is validated in step 24, or rejected otherwise in step 25. These different requirements can be checked in any order, being independent of each other, and they can be supplemented by other checks.

[0062] A test algorithm 8 by simulation of trajectory control programs can apply the candidate program y to a trajectory control device model along with signals representing a second set of parameters of the vehicle or its environment, which characterize certain trajectory control situations deemed particularly relevant. The second set of parameters may or may not include parameters already present in the first set, used in the first step 101. These parameters of the second set may also be representative of sensor signals present in the vehicle, or other indications. Exhaustiveness is sought, that is to say, recreating all plausible or possible situations.

[0063] The description of an expert system algorithm 9 is now given with reference to [Fig. 5]. The expert system 26 itself comprises a vehicle sensor signal simulator 27, a computer 28, and an analyzer 29, and it interacts with a model of an electronic trajectory control device 30 and models of other vehicle systems 31, which are typically systems interacting with the electronic trajectory control device 110. The sensor signal simulator 27 and the computer 28 provide their output signals to the model of the electronic trajectory control device 30 and to the model of other systems 31. And the model of the electronic trajectory control device 30, which is either the physical device itself or a simulation of this device on a computer, provides representative trajectory control signals to the computer 28 and the analyzer 29.

[0064] The computer 28 can transmit, in particular, the candidate program y or a sequence of candidate programs y to the electronic trajectory control device model 30 to simulate an electronic trajectory control situation, which is also defined by the sensor signal simulator 27. The results of the electronic trajectory control device model 30 provided to the computer 28 allow the candidate program y to be adjusted, and the analyzer 29 verifies that the braking is satisfactory. The model of the other systems 31 indicates whether the electronic trajectory control is functioning satisfactorily for these other systems as well.

[0065] The manufacturing process 100 may also include the following main steps, shown in [Fig. 6]. First, a check is carried out in step 32 for critical situations, represented by specific values ​​provided by the sensor signal simulator 27 and the computer 28. If the analyzer 29 concludes that the electronic trajectory control is satisfactory, then in step 33, a check is carried out for problems detected by the model 31 of other systems, which may, for example, cause overheating of the brakes 111. These problems of the vehicle 120 are typically at least partially independent of the trajectory control and appear, for example, after the application of candidate programs y, y', y''. Finally, a last important check may include variations in input parameters x in step 34.The manufacturing process 100 is repeated with different values ​​of the signals provided by the simulator 27 within the ranges that these signals can take. The choice of these different values ​​can be random, or guided by detections of potential problems by the analyzer 29 or the model 31 of the other systems. If all the checks are conclusive, a validation of the candidate program is performed at a further step 35; otherwise, the candidate program is rejected at a step 36. As with... the test algorithm 8, here simulations of trajectory control situations are carried out, but with interactivity between the computer and the electronic trajectory control device model 30, and a measurement of influence on other systems.

[0066] This second main step 102 of the process can be made more efficient by also applying the candidate programs y to other vehicle or trajectory control system models, according to step 37 shown in [Fig. 3], once the candidate programs y have been validated. The validation algorithms 7, 8, and 9 will be repeated for these other models. The corresponding candidate programs, denoted y", may, however, be identical to the programs y just validated. It is understood that the implementation of validation / error detection in trajectory control software developed without the implementation of machine learning, for example, by writing the program by a human operator, advantageously using a software development environment, or by algorithmic logic using the expert system of Figures 5 and / or 6, does not fall outside the scope of the present invention.

[0067] The third main step 103 of the method consists of recording at least one validated electronic trajectory control device program in the memory 108 of the computer 109 of the electronic trajectory control device 110. The validated candidate program is configured there to control the trajectory of the vehicle 120, in particular by differential braking of the wheels 119 of the vehicle 120 and / or by acting on an engine torque which is generated by the propulsion system 121. It takes into account a maximum of input parameters x, it individually and simultaneously controls the braking of each of the wheels 119 of the vehicle to control the trajectory of the vehicle 120, it prevents in particular the slippage of each wheel 119 during braking or acceleration of the vehicle 120.

[0068] When the electronic trajectory control device 110 contains at least one duly validated candidate program, it can be installed in the vehicle 120. With reference more specifically to the first embodiment and to [Fig. 2], the machine learning of the trajectory control program is stopped when the candidate program y, y', y” has been validated and implemented in the vehicle's electronic trajectory control device 110. The trajectory control program of the electronic trajectory control device 110 can nevertheless be updated, for example, during maintenance of the vehicle 120 in a garage.

[0069] Figure 10 illustrates a method for manufacturing an electronic trajectory control device according to a second embodiment, in which the modification of the trajectory control program is intended to continue after the installation of a validated candidate program in the electronic trajectory control device 110.

[0070] This modification of the trajectory control program is carried out, for example, by machine learning, in particular by deep learning using a neural network 1. This machine learning is, for example, supervised, unsupervised, and / or reinforcement learning. The machine learning is notably partially based on logic and knowledge, for example on an expert system 9 or analysis algorithms 7 such as those shown in [Fig. 3], and / or on a statistical approach such as fuzzy logic, Bayesian estimation, search trees, and optimization. The manufacturing device 104 preferably develops the candidate programs y, y', y" by deep learning and / or fuzzy logic.

[0071] In addition or alternatively, the modification of the trajectory control program is carried out by human supervision 38, for example by computer scientists and experts in the trajectory control of motorized road vehicles.

[0072] Due to the modification of the trajectory control program, the trajectory control program needs to be validated again, for example by validation algorithms such as the analysis algorithm 7, the testing algorithm 8 and / or the expert system 9 and similar to those shown in [Fig. 3]. The vehicle 120 then includes, in particular, its own electronic trajectory control device manufacturing system which includes the candidate program manufacturing device 104, the storage device 105 and the algorithmic validation device 107 as shown in [Fig. 9], which allows, for example, the trajectory control program of the vehicle 120 to be continuously or regularly updated in order to continuously improve the electronic trajectory control device 110 of the vehicle.

[0073] Of course, various modifications can be made by a person skilled in the art to the invention which has just been described without going out of the scope of the disclosure of the invention.

[0074] Alternatively, the vehicle 120 comprises three wheels such as a bicycle or more than four wheels such as a truck. LIST OF DIGITAL REFERENCES 1 Neural network 2 Neurons 3 Input layers 4 Intermediate layers 5 Output layers 6 Database 7 Analysis Algorithm 8 Testing Algorithm 9 Expert System 10 Decision Module 11 Validation of Candidate Program 12 Correction of Previously Proposed Candidate Program 13 Rejection of Candidate Program 14 Self-Learning in Case of Rejection 15 Distance Evaluation 16 Extraction of a Subset 17 Submission to Validation Algorithms 18 Retention in the Database 19 Deletion from the Database 20 Regulatory Requirements 21 Braking System Compatibility Requirements 22 Non-Braking System Compatibility Requirements 23 Customer Requirements 24 Validation of Candidate Program 25 Rejection of Candidate Program 26 Expert System 27 Sensor Signal Simulator 28 Expert System Calculator 29 Expert System Analyzer 30 Electronic Trajectory Control Device Model 31 Non-Electronic Trajectory Control Device Models 32 Critical Situation Checks 33 Problem Detection Check 34 Variation ofInput parameters 35 Validation of candidate program 36 Rejection of candidate program 37 Application to other vehicle models 38 Human supervision 100 Manufacturing program 101 First stage 102 Second stage 103 Third stage and update 104 Candidate program manufacturing device 105 Trajectory control program storage device 106 Candidate program database 107 Algorithmic validation device 108 Memory 109 Computer 110 Electronic trajectory control device 111 Electromechanical brake 112 Braking system 114 Brake control part 116 Mechanical brake part 118 Sensor 119 Wheel 120 Vehicle 121 Propulsion system 122 Braking control device 123 Steering wheel V Overall vehicle speed a Vehicle wheel angle y, y', y” Candidate program xl to x9 Input parameters xl Vehicle model parameters x2 Brake parameters - brake model or type x3 Braking control parameters x4 Vehicle movement parameters x5 Vehicle external environment parameters x6 Vehicle status parameters x7 Energy resource allocation or distribution parameters x8 Driving, trajectory, or braking history parameters x9 Driving or braking style parameters

Claims

Demands

1. A method for manufacturing (100) an electronic trajectory control device (110) for a motorized road vehicle (120), the vehicle (120) comprising wheels (119), rotational speed sensors (118) for each of the wheels (119) of the vehicle, brakes (111) for each of the wheels (119), and an electronic trajectory control device (110) that is configured to control the trajectory of the vehicle (120), characterized in that the manufacturing method (100) comprises synchronizing rotational speed measurements of each of the wheels (119) of the vehicle at first instants, and determining a first set of trajectory parameters (x) of the vehicle (120) at first instants, in that the manufacturing method (100) comprises developing (101) candidate programs (y, y',y") of an electronic trajectory control device (110) by machine learning from the first set of parameters (x) and synchronized measurements of the rotational speeds of each of the wheels (119).

2. A manufacturing method (100) according to the preceding claim, wherein the manufacturing method (100) comprises comparing the rotational speed measurements of the wheels (119) synchronized at each of the first instants, to detect a slip of at least one of the wheels of the vehicle and / or to determine a deviation between a trajectory desired by the driver and an actual trajectory of the vehicle.

3. A manufacturing method (100) according to any one of the preceding claims, wherein the manufacturing method (100) comprises the development (101) of candidate programs (y, y', y") of an electronic trajectory control device (110) by deep learning, in particular by neural network, and / or wherein the manufacturing method (100) comprises the development (101) by fuzzy logic of candidate programs (y, y', y") of an electronic trajectory control device (110).

4. A manufacturing method (100) according to any one of the preceding claims, wherein the first set of parameters (x) comprises parameters (xl) of vehicle model (120), vehicle state (x6), vehicle motion (x4), vehicle external environment (x5), brakes (x2), braking control (x3), driving and / or braking history (x8), driving or braking styles (x9), allocation or distribution of energy resources (x7) in the vehicle (120).

5. A manufacturing method (100) according to the preceding claim, wherein the first set of parameters (x) includes a vehicle mass (120), a vehicle age (120), an overall vehicle speed (v), a vehicle yaw rate (120), a vehicle wheel orientation setpoint (a), a vehicle acceleration (120), a vehicle positioning (120), a parameter representing wheel grip (119) on a road, a braking force of at least one of the vehicle brakes (111), a current through at least one of the brakes (111) when the brake (111) is electromechanical, and / or a current of a vehicle propulsion system (121) when the vehicle (120) is electric or hybrid.

6. A manufacturing method (100) according to any one of the preceding claims, wherein the manufacturing method (100) comprises the validation (102) or rejection of candidate programs (y, y', y") by applying at least one decision algorithm (7, 8, 9) and / or by human supervision.

7. A manufacturing method (100) according to the preceding claim, wherein the decision algorithm (7, 8, 9) comprises an analysis algorithm (7) and / or an expert system (9), the decision algorithm (7, 8, 9) comprising a check of regulatory (20), technical and / or driver-related (21, 22) requirements, and / or the decision algorithm (7, 8, 9) being configured to detect vehicle (120) problems (33) at least partially independent of the trajectory control and appearing after application of candidate programs (y, y', y").

8. A manufacturing method (100) according to any one of the preceding claims, wherein the decision algorithm (7, 8, 9) comprises a test algorithm (8) which is configured to analyze application results of candidate programs (y, y', y") to trajectory control situations expressed by a second set of parameters (x), the decision algorithm (7, 8, 9) preferably being configured to vary the parameters (x) of the second set of parameters (x) in decision ranges to implement the test algorithm.

9. Manufacturing method (100) according to the preceding claim, wherein the manufacturing method (100) comprises a synchronization of rotational speed measurements of each of the wheels (119) of the vehicle at second instants, and the determination of the second set of trajectory parameters (x) of the vehicle (120) at the second instants.

10. A manufacturing method (100) according to any one of the preceding claims, wherein the decision algorithm (7, 8, 9) includes a self-learning step (14) configured to re-evaluate previously validated candidate programs (y, y', y"), in the event of rejection of a candidate program judged to be similar to the previously validated candidate programs (y, y', y").

11. A manufacturing method (100) according to any one of the preceding claims, wherein the manufacturing method (100) includes updating (103) the candidate program after installation in a computer (109) of the vehicle's electronic trajectory control device (110), in particular by continuing the automatic learning of the candidate program, and / or wherein the manufacturing method (100) includes updating a program of the electronic trajectory control device (110) during vehicle maintenance operations (120).

12. Motorized road vehicle (120) comprising wheels (119), rotation speed sensors (118) for each of the wheels (119) of the vehicle, brakes (111) for each of the wheels (119), and an electronic trajectory control device (110) which is configured to control the trajectory of the vehicle (120), characterized in that the electronic trajectory control device (110) is manufactured by a manufacturing process (100) according to any one of claims 1 to 11.

13. Vehicle (120) according to the preceding claim, wherein the electronic trajectory control device (110) is configured to minimize the slippage of each of the wheels (119) and to generate wheel braking commands by the brakes (111) by enabling the vehicle (120) to follow a desired trajectory which is determined from a wheel orientation command (a) of the vehicle corresponding in particular to a steering wheel angle measurement.

Images