Method for determining the speed of a vehicle and processing unit enabling the implementation of such a method
An LSTM neural network trained with wheel speed data addresses the dependency on ESP/ESC by providing a reliable method for independent vehicle speed determination, improving braking security.
Patent Information
- Application Number
- FR2022001792
- Authority / Receiving Office
- FR · FR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2022-03-01
- Publication Date
- 2024-12-13
- Estimated Expiration
- 2042-03-01
AI Technical Summary
Existing vehicle safety systems, such as ABS, rely on the proper functioning of ESP/ESC for determining vehicle speed, creating a dependency that can be problematic and limiting independence in braking operations.
A method using an LSTM neural network trained with vehicle wheel speed data and optional additional parameters to determine vehicle speed independently of ESP/ESC, utilizing recurrent neurons with short and long-term memory to account for varying conditions.
Enables reliable and independent vehicle speed determination, enhancing secure braking, especially during emergency situations, without relying on external systems like ESP/ESC.
Smart Images
Figure 00000011_0000 
Figure 00000011_0001 
Figure 00000012_0000
Abstract
Description
Title of the invention: Method for determining the speed of a vehicle and processing unit enabling the implementation of such a method Technical field
[0001] The invention relates to the field of vehicles and safety systems on board such vehicles.
[0002] The invention thus more specifically relates to a method for training a neural network for determining the speed of a vehicle, a method for determining the speed of a vehicle, a processing unit and a vehicle comprising such a processing unit. State of the prior art
[0003] To optimize vehicle safety, particularly braking, it is important to know the speed of a vehicle at all times. Such a speed is necessary to optimize the operation of anti-lock braking systems, better known by the English acronym ABS, or rear wheel anti-lock braking systems, better known by the English acronym RWU.
[0004] By vehicle speed, it is understood above and in the rest of the description a linear speed of the vehicle.
[0005] Currently, this information is provided to the brake control unit by the electronic stability program / system (known by the English acronym ESP / ESC). This results in an interdependence between the two systems which can be problematic since the ABS is therefore dependent on the correct functioning of the ESP / ESC.
[0006] There would therefore be an interest in providing a brake control unit which is totally independent of the ESP / ESC and which is nevertheless able to take into account the speed of the vehicle in order to make the braking of the vehicle safer. Statement of the invention
[0007] The invention thus aims to resolve the above drawbacks and therefore aims to provide a method for determining the speed of a vehicle which can be implemented on a brake control unit and which does not require having to call upon the ESP / ECS.
[0008] To this end, the invention relates to a method for training a neural network for determining the speed of a vehicle comprising at least two wheels, the method being implemented by means of a neural network of the neural network type. recurrent neurons with short and long term memory, called LSTM neural network, the method comprising the following steps: - recovery of calibration data corresponding to different vehicle rolling situations, said calibration data including a linear speed of at least two of the vehicle's wheels, preferably each of the vehicle's wheels, and a vehicle speed, - training the first LSTM neural network, with, as input to said neural network, the calibration data recovered for at least a first part of the vehicle rolling situations, the vehicle speed being excluded from said data, and as output to said neural network a value relating to the vehicle speed.
[0009] Such a method for training a neural network is capable of enabling a determination of the speed of the vehicle based solely on the speeds of the wheels which is reliable, since it takes into account, due to the use of recurrent neurons with short and long term memory, the evolution of the rolling conditions of the vehicle and which does not require the use of an external system such as an electronic stability program / system. It is therefore possible, from a neural network trained with such a method, to provide an estimate of a reliable speed of a vehicle and this independently of the ESP / ESC.
[0010] The calibration data may further comprise at least one vehicle parameter, the at least one vehicle parameter being preferentially selected from the group comprising a signal provided by a yaw sensor, a satellite geopositioning signal, an anti-lock braking system.
[0011] In this way, the vehicle's driving situations can be fully taken into account. This results in an optimized determination of the vehicle speed.
[0012] The following step may be provided between the calibration data recovery step EL and the training step E3: - E2. separation of the different vehicle rolling situations into at least a first, a second and a third group of distinct rolling situations, the first part from which the training step corresponding to the first group of rolling situations is carried out, in which the following steps are provided after the training step E3: - E4. validation of the training of the LSTM neural network (10) from the calibration data of the second group of situations, the training step being implemented again in the case where the validation step does not allow the training to be validated from the calibration data of the second group of situations, - E6. testing of the LSTM neural network from the calibration data of the third group.
[0013] In this way the determination of the speed is validated, it follows that the speed determined from such a training method can be considered perfectly reliable.
[0014] The invention further relates to a method for determining the speed of a vehicle comprising at least two wheels, the method being implemented by means of a neural network of the short and long term memory recurrent neural network type, called an LSTM neural network, having been previously trained within the framework of a method according to the invention: - El 1. recovery of the linear speed of at least two wheels of the vehicle, preferably of each of the wheels, of said vehicles, - E12. determination of the value relative to the speed of the vehicle from the LSTM neural network by providing it as input the linear speed of at least two wheels of the vehicle, preferably of all the wheels of the vehicle.
[0015] Such a determination method makes it possible, due to the learning method according to the invention, to obtain a particularly reliable vehicle speed. It is thus possible, from such a reliable speed, to provide particularly secure braking, in particular during emergency braking using electric parking brakes of a vehicle.
[0016] The LSTM neural network having been previously trained within the framework of a method according to the invention in which the calibration data further comprise at least one parameter of the vehicle, during step E1 1. of recovering the linear speed of at least two wheels it can furthermore be recovered the at least one parameter of the vehicle.
[0017] After step E12. of determining the speed of the vehicle, the model determined during the prior training can further be corrected to take into account the data recovered during the implementation of the determination method.
[0018] In this way it is possible to take into account changes in the vehicle's traffic conditions, such as, for example, tire wear.
[0019] The invention further relates to a processing unit of a vehicle comprising at least two wheels, the processing unit comprising a neural network, in which the neural network is a neural network of the short and long term memory recurrent neural network type, called an LSTM neural network, said LSTM neural network having been trained according to a training method according to the invention, the processing unit being configured to recover the linear speed of at least two wheels of the vehicle and to determine, from said linear speeds, a value relating to the speed of the vehicle.
[0020] The vehicle may comprise at least four wheels including at least two rear wheels, the processing unit being a rear brake control unit adapted to control rear brakes of the vehicle which equip said at least two rear wheels, said rear brakes preferably being electric parking brakes of the vehicle.
[0021] The invention further relates to a vehicle comprising at least two and one processing unit according to the invention.
[0022] Such a vehicle benefits from the advantages linked to the reliable speed determination enabled by the learning method according to the invention. Brief description of the drawings
[0023] The present invention will be better understood on reading the description of exemplary embodiments, given purely for informational purposes and in no way limiting, with reference to the appended drawings in which:
[0024] [Fig.l] schematically illustrates a vehicle comprising a processing unit according to the invention;
[0025] [Fig.2] schematically illustrates a neural network as implemented in the scope of the invention;
[0026] [Fig.3] illustrates a flowchart of a training method according to the invention of a neural network as illustrated in [Fig.2];
[0027] [Fig.4] illustrates a flowchart of a method for determining a speed of the vehicle according to the invention.
[0028] Identical, similar or equivalent parts of the different figures bear the same numerical references so as to facilitate the transition from one figure to another.
[0029] The different parts represented in the figures are not necessarily on a uniform scale, in order to make the figures more readable.
[0030] The different possibilities (variants and embodiments) must be understood as not being mutually exclusive and can be combined with each other. Description of the embodiments
[0031] [Fig.l] illustrates a vehicle 1, such as an automobile, equipped with a rear electric braking system, a processing unit of which is capable of implementing a speed determination method according to the invention.
[0032] It will be noted that in the present case and as illustrated in [Fig.l], the vehicle 1 being a four-wheeled vehicle 2A, 2B, 2C, 2D, such as an automobile, it comprises two rear electric brakes 3C, 3D capable of acting as a parking brake. Of course, even if the invention is particularly advantageous in the context of such an application, it is not limited to this configuration alone and it is perfectly applicable to vehicles comprising a number of wheels other than four and to the use of brakes having a configuration other than this one. It will also be noted that if the present invention is particularly suitable for being implemented implemented by a brake control unit, as is the case in the present embodiment, it can be implemented by another type of processing unit equipping a vehicle, without departing from the scope of the invention.
[0033] As shown in [Fig.l], the vehicle comprises, in addition to the four wheels 2A, 2B, 2C, 2D: - the rear electric brakes 3C, 3D capable of acting as a parking brake, - wheel speed sensors 4A, 4B, 4C, 4D each equipping a respective wheel 2A, 2B, 2C, 2D of the vehicle, each wheel speed sensor 4A, 4B, 4C, 4D being capable of providing a linear speed V4A, V4B, V4C, V4D of the corresponding wheel 2A, 2B, 2C, 2D, - one or more possible pieces of equipment of the vehicle 6, such as a yaw sensor and a geo-positioning signal, - a control unit 5 for the rear electric brakes 3C, 3D.
[0034] The control unit 5, or processing unit, is configured to control the electric brakes and is preferably adapted to implement an anti-lock wheel system, such as ABS or RWU. Similarly, in accordance with the invention, the control unit 5 is able to determine a speed of the vehicle 1 from the linear speed V4A, V4B, V4C, V4D of the wheels 2A, 2B, 2C, 2D. To do this, the control unit 5 is therefore in direct or indirect communication with at least some of the wheel speed sensors 4A, 4B, 4C, 4D to recover the corresponding linear speeds V4A, V4B, V4C, V4D.
[0035] the control unit 5 comprises a neural network 10 of the short and long term memory recurrent neural network type, called LSTM neural network 10, as illustrated in [Fig.2]. As shown in [Fig.2], the neural network comprises a plurality of LSTM neurons 10A interconnected and arranged in layers L1, L2, LN including an input layer L1, intended for the recovery of input data, including in particular and an output layer LN, the remaining layers L2 being called hidden layers.
[0036] In order to allow a determination of the speed of the vehicle, the LSTM neural network 10 can be trained from a training method for determining the speed Vi of a vehicle 1, the principle steps of which are illustrated in the flowchart shown in [Fig.3]. Such a training method comprises the following steps: - EL recovery of calibration data corresponding to different rolling situations of the vehicle 1, said calibration data including a linear speed V4A, V4B, V4C, V4D of the wheels 2A, 2B, 2C, 2D of the vehicle 1 and a speed Vi of the vehicle 1, - E3. training of the LSTM neural network 10, with, as input to said LSTM neural network 10, the calibration data recovered for at least a first part of the driving situations of the vehicle 1, the speed Vi of the vehicle 1 being excluded from said data, and as output to said LSTM neural network 10 a value relating to the speed Vi of the vehicle 1.
[0037] It will be noted that, in a conventional configuration, corresponding to that of the flowchart of [Fig. 3], the method further comprises, between the step EL of retrieving calibration data and the step E3. of training, the following step: - E2. separation of the different rolling situations of the vehicle 1 into at least a first, a second and a third group of distinct rolling situations, the first part from which the training step is carried out corresponding to the first group of rolling situations, and after the step E3. of training the following steps: - E4. validation of the training of the LSTM neural network (10) from the calibration data of the second group of situations, the training step being implemented again in the case where the validation step does not allow the training to be validated from the calibration data of the second group of situations, - E6. testing of the LSTM neural network from the calibration data of the third group.
[0038] It will be noted that, as indicated in step E4., after the validation step E4., a verification E5. of the result of the validation step is provided in such a way as to re-implement the training step E3. in the case where said training is not validated during the validation step E4.
[0039] It will be noted that to perfect the training, during step E3. training of the LSTM neural network 10, the calibration data may further comprise at least one parameter of the vehicle 1, such as a signal provided by a yaw sensor, a satellite geo-positioning signal or even an anti-lock braking system. To do this, as shown in [Fig.l], the processing unit 5 is able to interrogate, directly or indirectly, one or more of the possible equipment of the vehicle 6.
[0040] In the context of such training, the different driving situations of the vehicle 1 are chosen to correspond, in the most exhaustive manner possible, to the driving situations likely to be encountered by the vehicle, including in particular normal driving in different types of weather and surfaces (different types of asphalt, dirt roads, concrete, this in dry, wet conditions and in the presence of ice) and driving with anti-lock braking system activated or deactivated in different types of weather and surfaces as well.
[0041] With an LSTM neural network 10 thus trained, the control unit 5 is able to implement a method for determining the speed of the vehicle 1. A method for determining the speed of the vehicle comprises, as shown in the flowchart illustrated in [Fig.4], the following steps: - El 1. recovery of the linear speed V4A, V4B, V4C, V4D of each of the wheels 2A, 2B, 2C, 2D, of said vehicle 1, - E12. determination of the value relative to the speed Vi of vehicle 1 from the LSTM 10 neural network by providing it as input the linear speed V4A, V4B, V4C, V4D of the wheels 2A, 2B, 2C, 2D of vehicle 1.
[0042] It will be noted that in the context of step El 1. of recovering the linear speed (V4A, V4B, V4C, V4D) of each wheel 2A, 2B, 2C, 2D, at least one parameter of the vehicle 1 can also be recovered.
[0043] The value relating to the speed determined within the framework of such a method for determining the speed of the vehicle can be selected from the group comprising: an estimated speed of the vehicle 1, a value proportional to such an estimated speed of the vehicle 1 or a multiplicative factor with respect to a linear speed V4A, V4B, V4C, V4D of the wheel among the linear speeds of the linear wheel V4A, V4B, V4C, V4D.
Claims
Claims
1. Method for training a neural network (10) for determining the instantaneous speed (Vi) of a vehicle (1) comprising at least two wheels (2A, 2B, 2C, 2D), the method being implemented by means of a neural network (10) of the short and long term memory recurrent neural network type, called LSTM neural network, the method comprising the following steps: - E1. recovery of calibration data corresponding to different rolling situations of the vehicle (1), said calibration data including a linear speed (V4A, V4B, V4C, V4D) of at least two of the wheels (2A, 2B, 2C, 2D) of the vehicle (1), preferably of each of the wheels (2A, 2B, 2C, 2D) of the vehicle (1), and an instantaneous speed (Vi) of the vehicle (1), - E3.training the LSTM neural network (10), with, as input to said LSTM neural network (10), the calibration data recovered for at least a first part of the rolling situations of the vehicle (1), the instantaneous speed (Vi) of the vehicle (1) being excluded from said data, and as output to said LSTM neural network (10) a value relating to the instantaneous speed (Vi) of the vehicle (1).
2. Training method according to claim 1, wherein the calibration data further comprises at least one parameter of the vehicle (1), the at least one parameter of the vehicle being preferentially selected from the group comprising a signal provided by a yaw sensor, a satellite geo-positioning signal, an anti-lock braking system.
3. Training method according to one of claims 1 or 2, in which the following step is provided between the calibration data recovery step EL and the training step E3: - E2. separation of the different rolling situations of the vehicle (1) into at least a first, a second and a third group of distinct rolling situations, the first part of the rolling situations of the vehicle (1) from which the training step is carried out corresponding to the first group of rolling situations, in which the following steps are provided after the training step E3: - E4. validation of the training of the LSTM neural network (10) from the calibration data of the second group of situations, the training step being implemented again in the case where the validation step does not allow the training to be validated from the calibration data of the second group of situations, - E6. testing of the LSTM neural network from the calibration data of the third group.
4. Method for determining the speed (Vi) of a vehicle comprising at least two wheels (2A, 2B, 2C, 2D), the method being implemented by means of a neural network (10) of the short and long term memory recurrent neural network type, called LSTM neural network, having been previously trained in the context of a method according to any one of claims 1 to 3: - Eli. recovery of the linear speed (V4A, V4B, V4C, V4D) of at least two wheels (2A, 2B, 2C, 2D) of the vehicle (1), preferably of each of the wheels (2A, 2B, 2C, 2D), of said vehicle (1), - E12. determining the value relative to the instantaneous speed (Vi) of the vehicle (1) from the LSTM neural network (10) by providing it as input the linear speed (V4A, V4B, V4C, V4D) of at least two wheels (2A, 2B, 2C, 2D) of the vehicle (1), preferably of all the wheels (2A, 2B, 2C, 2D) of the vehicle (1).
5. Method for determining the instantaneous speed (Vi) of a vehicle (1) according to claim 4, in which the LSTM neural network (10) has been previously trained in the context of a method according to claim 2 taken alone or in combination with claim 3 or 4, in which during step El 1. of recovering the linear speed (V4A, V4B, V4C, V4D) of at least two wheels, the at least one parameter of the vehicle (1) is also recovered.
6. Method for determining the instantaneous speed (Vi) of a vehicle (1) according to claim 4 or 5, in which after step E12. of determining the speed of the vehicle (1), the model determined during the prior training is further corrected to take into account the data recovered during the implementation of the determination method.
7. Brake control unit (5) of a vehicle (1) comprising at least two wheels (2A, 2B, 2C, 2D), the processing unit (5) comprising a neural network (10), wherein the neural network (10) is a neural network of the short and long term memory recurrent neural network type, called an LSTM neural network, said LSTM neural network (10) having been trained according to a training method according to any one of claims 1 to 4, the processing unit (5) being configured to recover the linear speed (V4A, V4B, V4C, V4D) of at least two wheels (2A, 2B, 2C, 2D) of the vehicle (1) and to determine, from said linear speeds (V4A, V4B, V4C, V4D) a value relating to the instantaneous speed (Vi) of the vehicle (1).
8. Brake control unit (5) according to claim 7, wherein the at least two wheels (2A, 2B, 2C, 2D) are rear wheels (2C, 2D), the processing unit (5) being a rear brake control unit adapted to control rear brakes (3C, 3D) of the vehicle (1) which equip said at least two rear wheels (2C, 2D), said rear brakes (3C, 3D) preferably being electric parking brakes of the vehicle (1).
9. Vehicle (1) comprising at least two wheels (2A, 2B, 2C, 2D) and a brake control unit (5) according to claim 7 or 8.