METHOD FOR SUPPORTING THE DRIVING OF AN AUTONOMOUS MOTOR VEHICLE ON A ROAD

Description

Domaine technique

[0001] The present invention relates to a method for assisting the driving of an autonomous motor vehicle on a road, a driving assistance device for controlling said autonomous motor vehicle, a computer program product comprising program instructions usable by said driving assistance device, and an autonomous motor vehicle comprising said driving assistance device. Technique antérieure

[0002] In recent years, advances in equipping motor vehicles with driver assistance systems have contributed to a significant improvement in road safety. The challenge for the future is to design autonomous vehicles. An autonomous vehicle is a vehicle adapted to drive on an open road with little or no driver intervention. The concept aims to develop and produce such a vehicle that can eventually operate safely on public roads, regardless of traffic generated by other vehicles or obstacles (humans, animals, trees, etc.) present on the road. The term "autonomous vehicle" here encompasses a fully autonomous vehicle in which human operator intervention is not required for driving.This concept also covers a so-called "semi-autonomous" motor vehicle equipped with automated driving assistance systems, but in which the intervention of the human operator remains generally important.

[0003] The operation of an autonomous vehicle is generally governed jointly by an onboard navigation system and a system for defining a predetermined trajectory. The onboard navigation system is designed to program a large-scale route, that is, across an entire road network. This route comprises a series of consecutive road segments adapted to connect a starting point to a destination. This series of road segments may vary dynamically depending on detected traffic conditions or predefined constraints on the road network (such as the closure of certain sections for ongoing roadworks).

[0004] The reference trajectory definition device is designed to automatically process vehicle movement trajectories on a local scale, on the order of several tens or hundreds of meters. This reference trajectory definition device is adapted to implement the route programmed by the navigation system. This implementation is achieved by determining the vehicle's position, direction, and speed over time based on movement constraints, such as dynamic constraints of the vehicle (maximum speed, longitudinal acceleration, steering angle, etc.), environmental constraints (obstacles on the road, etc.), or optimization constraints (minimizing the vehicle's lateral acceleration, for example).

[0005] In an autonomous driving context, a motor vehicle is configured to follow a predefined urban or extra-urban route by traversing a series of road segments determined by the onboard navigation system. The reference trajectory is defined by a set of Cartesian coordinates that vary over time and are calculated in real time based on the road segment at which the predefined route is reached and environmental parameters. This calculation can be performed using a road marking recognition system, radar or laser detection methods, an obstacle detection system, etc. During driving, the motor vehicle follows an actual trajectory that may differ from the reference trajectory. Indeed, during driving, the motor vehicle is subjected to a number of constraints (tire pressure, road gradient, wind force, etc.).) which may not have been taken into account by the device defining the predetermined trajectory. Therefore, there may be a discrepancy between this actual trajectory and the reference trajectory which must be minimized.

[0006] US2018 / 0348763 discloses a driver assistance method designed to optimize the control of an autonomous vehicle. This method includes a step of receiving multiple perception data points from the driving environment surrounding the vehicle and a step of generating a command to control the vehicle. This command can be generated using either a deliberative or an intuitive method. In the deliberative method, the command is generated from a predefined set of driving rules. In the intuitive method, the command is generated from a machine learning model. The choice of decision-making method, deliberative or intuitive, is made based on the collected data. In complex situations, the method is adapted to use the deliberative method. In more familiar situations, the method is adapted to use the intuitive method.The driver assistance process then repeats itself, switching from one decision-making mode to another. While this process enables adaptive driver assistance, it requires maintaining and updating two parallel decision-making systems: one based on rules and the other on machine learning. These two systems must be perfectly synchronized because the process may rely on one or the other depending on the selected operating mode. This can lead to delays in the process, which can be critical for certain maneuvers. Furthermore, the machine learning described in this document requires complete relearning with each new data processing, which can result in varying performance from one cycle to the next.

[0007] Therefore, there is a need to offer a more responsive driving assistance method that is also more precise in responding to the command to be given.

[0008] US patent application 2020 / 0218271 A1 describes a method in which the driving characteristics of an autonomous vehicle are adjusted based on a history of the user's satisfaction level with the vehicle. US patent application 2020 / 0276988 A1 describes a method in which the driving characteristics of an autonomous vehicle are adjusted based on risk predictions. Exposé de l'invention

[0009] The present invention aims to address at least partially this need.

[0010] More specifically, the present invention aims to improve the control of an autonomous vehicle on a road.

[0011] A first object of the invention relates to a method of assisting the driving of an autonomous motor vehicle on a road according to claim 1.

[0012] The invention proposes a multi-evolutionary decision-making method capable of generating the most appropriate response by combining, in real time, all available information with knowledge acquired in similar situations in the past. This method offers two main functionalities to meet these requirements. Firstly, the method takes into account approximate data for three main parameters affecting the performance of the autonomous vehicle: the vehicle's condition, the road conditions, and interaction with the environment. Secondly, the method adapts the decision to the current situation by comparing it with other previous situations the vehicle has encountered. This learning allows the vehicle's control to be adjusted accordingly.

[0013] According to the invention, the plurality of data received comprises a first type of data, a second type of data and a third type of data, the data of the first type of data relating to an evolution of the vehicle, the data of the second type of data relating to an evolution of the road, the data of the third type of data relating to an evolution of the environment.

[0014] In a particular embodiment, the data of the first data type includes data selected from the following list of data: data on mass change; data on tire change; data on the change of state of the autonomous motor vehicle.

[0015] In a particular embodiment, the data of the second data type includes data selected from the following list of data: weather conditions data; road type data; road condition data.

[0016] In one particular embodiment, the data of the third data type includes data selected from the following list of data: Free driving data; nominal driving data; emergency driving data.

[0017] In a particular embodiment, the steps of receiving and processing the plurality of data, the step of generating a control response and the step of sending the command are carried out according to a fuzzy logic.

[0018] The invention proposes a fuzzy logic-based system that is subtly updated / modified based on the results of vehicle actions. This eliminates the need for a complete relearning process each time. Such a complete relearning process can, in fact, produce significantly different performance from one test to the next.

[0019] Another object of the invention relates to a device for assisting the driving of an autonomous motor vehicle according to claim 6.

[0020] Thus, thanks to this decision-making system, autonomous driving will have the ability to mimic the reactions of the human driver during complex scenarios by creating a learning base to optimize the response in real time.

[0021] Another object of the invention relates to a computer program product comprising program instructions that can be used by the driver assistance device according to a previous object, which, when executed or interpreted by said driver assistance device, trigger the implementation of the driver assistance method according to another previous object in an autonomous motor vehicle.

[0022] Another object of the invention relates to an autonomous vehicle comprising a driving assistance device according to one of the preceding objects.

[0023] The present invention will be better understood upon reading the detailed description of embodiments taken by way of non-limiting examples and illustrated by the accompanying drawings, in which: There figure 1 is a schematic view illustrating an autonomous motor vehicle according to the invention; The figure 2 illustrates the autonomous automobile vehicle of the figure 1 operating on a road; The figure 3 is a diagram illustrating a driver assistance device for the motor vehicle of the figure 1 on the road to figure 2 ; There figure 4 is a diagram detailing a fuzzy system belonging to the driver assistance system of the figure 3 ; There figure 5 is a diagram detailing the operation of a reinforcement learning agent belonging to the driver assistance system of the figure 3 ; There figure 6 is a diagram illustrating different stages of a method for assisting the driving of a motor vehicle. figure 1 said control method being implemented by the driver assistance system of the figure 3 .

[0024] The invention is not limited to the embodiments and variants shown, and other embodiments and variants will be obvious to a person skilled in the art.

[0025] In the different figures, identical or similar elements bear the same references.

[0026] There figure 1 The diagram schematically represents a top view of a motor vehicle 10 according to the present invention. This motor vehicle 10 comprises a front section, a rear section, a roof, an interior, and steering means (not shown). The motor vehicle also comprises a chassis and one or more body panels mounted or fixed to said chassis.

[0027] Vehicle 10 is an autonomous vehicle. By autonomous vehicle, we mean a fully or semi-autonomous vehicle. There are, in fact, several levels of autonomy in a vehicle.

[0028] In the first level, known as level 1, the vehicle is responsible for a limited number of driving operations associated with that vehicle. The driver remains responsible for most driving control operations. In level 1, acceleration and / or braking (cruise control, etc.) are controlled by the vehicle. This level 1 corresponds to a level of driver assistance.

[0029] In a second level, known as Level 2, the vehicle is adapted to collect information (for example, via one or more driver assistance systems, sensors, etc.) about the external environment (the road around the vehicle, the road surface, traffic, ambient conditions). At this Level 2, the autonomous vehicle is adapted to use the collected information to control certain driving operations (for example, steering, acceleration, and / or braking). This Level 2 represents a level of partial automation for the vehicle. It is important to note that for both Level 1 and Level 2, the driver must retain full control of the driving operations performed by the autonomous vehicle.

[0030] In a third level, called level 3, the driver delegates all driving operations to the motor vehicle except when the motor vehicle requests the driver to act or intervene to control one or more of these driving operations. This level 3 is a level of conditional automation.

[0031] In a fourth level, known as level 4, the driver no longer manages any driving operations. The vehicle then controls all driving operations, including when the driver does not respond to a request for intervention. Level 4 is a high level of automation.

[0032] In a fifth level, known as level 5, the vehicle controls all driving operations. During driving, the vehicle monitors road traffic, moving objects on the road (humans, animals), stationary obstacles, and the road itself. At this level 5, no interaction with a human driver is required. Level 5 represents the level of complete automation. It should be noted that for levels 3 through 5, the vehicle is adapted to monitor driving operations as well as its external environment.

[0033] To perform the various driving operations at these different levels of autonomy, the vehicle 10 includes a number of sensors such as: a front radar 11A, 11B; a rear radar 12A, 12B; an ultrasonic sensor 13; a video camera 14; a LIDAR 15; a GPS antenna 16.

[0034] The front radar comprises two front radar elements 11A and 11B arranged on the front of the vehicle on either side of an axis of symmetry of said vehicle 10. The front radar has a detection zone 111 at the front of the vehicle. It is thus adapted to detect the positions of surrounding objects. It allows the speed of the vehicle to be measured. The information gathered by the front radar 11A and 11B is particularly useful for implementing certain driving operations such as emergency braking or lane departure warning.

[0035] The rear radar comprises two rear radar elements 12A and 12B positioned on the rear of the vehicle, on either side of the vehicle's axis of symmetry 10. The rear radar has a detection zone 112 at the rear of the vehicle. It is thus adapted to detect the positions of surrounding objects. It allows the speed of other vehicles following the vehicle 10 to be measured. The information gathered by the rear radar 12A and 12B is particularly useful for implementing certain driving operations such as emergency braking or lane departure warning.

[0036] The ultrasonic sensor 13 is located on the front of the vehicle between the two front radar elements 11A and 11B. The ultrasonic sensor 13 has a detection zone 113 that is significantly smaller than the detection zone 111 of the front radar 11A and 11B. This ultrasonic sensor 13 is therefore suitable for detecting very close obstacles. The information gathered is particularly useful for implementing driving maneuvers such as maintaining a safe following distance from another vehicle directly in front of the vehicle 10.

[0037] Video camera 14 is located here behind the rearview mirror. The information gathered by this video camera 14 is particularly useful for carrying out certain driving operations such as deciphering road signs, identifying curbs and center lines, and detecting moving objects on the road (humans, animals).

[0038] The LIDAR 15 (for "Light Detection and Ranging") is a sensor located here on the roof of the vehicle 10. It enables laser remote sensing measurements. Laser remote sensing is a remote measurement technique based on analyzing the properties of a light beam reflected back to its emitter. The LIDAR has a fairly large detection area 115, for example, on the order of 50 meters in diameter. It allows for continuous 360° scanning of the environment to create a 3D map. The information gathered by the LIDAR 15 is useful for implementing certain driving operations such as obstacle detection, including at night.

[0039] The GPS antenna 16 is located on the rear of the vehicle 10. It receives GPS (Global Positioning System) signals. Navigation data can then be updated in the vehicle 10 based on these GPS signals.

[0040] The motor vehicle also includes a central computer 17 adapted to process the various data from the sensors 11A, 11B, 12A, 12B, 13, 14, 15, 16. The central computer 17 and the sensors 11A, 11B, 12A, 12B, 13, 14, 15, 16 are connected in the motor vehicle by one or more networks (not shown) of the CAN bus type (for "Controller Area Network" in English) for the transport of said sensor data.

[0041] There figure 2 illustrates the autonomous automobile vehicle of the figure 1 on road 20. To simplify this illustration, sensors 11A, 11B, 12A, 12B, 13, 14, 15 have not been shown on the figure 2 Route 20 is bordered by two curbs 21. It is further divided into two lanes separated by multiple median strips 22. On the figure 2 The vehicle is traveling in the right lane, following an actual trajectory P. This actual trajectory P is offset laterally by a distance DL relative to a reference trajectory T. The reference trajectory T was previously calculated by the navigation system of the motor vehicle 10, notably from GPS signal data received by the GPS antenna 16. This reference trajectory T extends midway between the curb 21 and the central lines 22. The invention aims to help the vehicle 10 follow the reference trajectory T as closely as possible. To this end, the motor vehicle 10 includes a driver assistance device 30 illustrated in the figure. figure 3 This driver assistance device 30 is, in a preferred embodiment, integrated directly into the central computer 16 of the motor vehicle. figure 1 .

[0042] There figure 3 illustrates in more detail the driver assistance system 30. As already mentioned, this driver assistance system 30 is adapted to control the movements of the autonomous motor vehicle 10. It includes: a vehicle evolution module 301; a road evolution module 302; an environment evolution module 303; a fuzzy system 304; a reinforcement learning agent 305; an actuator control block 306; a Human-Machine interface 307.

[0043] The vehicle's evolution module 301 is designed to provide a first type of data, data1. This first type of data, data1, provides information on any potential situation that could modify the dynamics of vehicle 10. The evolution module 301 includes: a mass change block 3011; a tire change block 3012; a vehicle state change block 3013;

[0044] The mass change block 3011 is designed to provide mass change data. This data concerns the mass of the autonomous vehicle 10 and its distribution within the vehicle (passengers, cargo, etc.). Mass change data is important because the characteristics associated with the vehicle's mass can completely alter the vehicle's dynamics and influence the most appropriate decision to be made in a given situation. Block 3011 is therefore responsible for analyzing any deviations from nominal values.

[0045] The 3012 tire change unit is designed to provide tire change data. Tires are the points of contact between the vehicle 10 and the road 20. The associated data is therefore important because it allows for a good estimation of the relationship between lateral forces and the slip angle, i.e., cornering stiffness. This stiffness can be evaluated using an observer to measure the difference between nominal / factory values ​​and real-time updated values.

[0046] The 3013 vehicle state change block is suitable for providing data on any unexpected change in the motor vehicle that may alter the dynamics of that vehicle, such as a sensor failure, a puncture, etc.

[0047] The data of the first data type data1 of the evolution module 301 is therefore a combination of data from these different blocks 3011, 3012, 3013.

[0048] The road evolution module 302 is adapted to provide a second type of data, data2. This second type of data, data2, provides information on any potential situation that could alter the road condition. Road conditions have a significant impact on vehicle handling. The evolution module 302 therefore monitors any potential changes that may result from the road layout.

[0049] The 302 evolution module includes: a weather conditions block 3021; ​​a road type block 3022; a road state block 3023.

[0050] The 3021 weather conditions module is designed to manage changing weather conditions, including snow, ice, drought, and rain. Module 3021 can therefore adapt its response to the specific situation. This change can be measured by a temperature sensor installed on vehicle 10 or by any other means.

[0051] The 3022 road type block is designed to handle changes in the road surface on which vehicle 10 is traveling. Indeed, grip changes considerably when the vehicle is driving on concrete, asphalt, gravel, cobblestones, or any other surface, thus altering the vehicle's dynamics. This change can be measured by a front camera to detect the driving surface, by a detailed digital map including this information, or by any other means.

[0052] The 3023 road condition block is adapted to detect any sudden and unexpected changes in the road surface, such as a pothole, oil slick, or any other hazardous situation that could impact vehicle dynamics.

[0053] The data of the second type of data data2 of the evolution module 302 is therefore a combination of data from these different blocks 3021, 3022, 3023.

[0054] The 303 environment evolution module is adapted to provide a third type of data, data3. This third type of data, data3, provides information on interactions with third parties such as pedestrians, other motor vehicles (cars, motorcycles, trucks, etc.).

[0055] Module 303 on environmental evolution includes: a 3031 free control block; a 3032 nominal control block; a 3033 emergency control block.

[0056] The 3031 free driving block includes all cases where no interaction is present. This block thus provides free driving data covering all driving scenarios on lightly congested highways or any other potential situation in which no third party is within the vehicle's electronic horizon, i.e., the field of vision of the obstacle sensors.

[0057] The nominal driving block 3032 encompasses all potential scenarios and how they are handled. This block provides nominal driving data that reflects different behaviors depending on the driving scenario, such as a roundabout, an intersection, a lane reduction, interaction with traffic lights, an exit lane, etc. Block 3032 can include specific processing for each scenario based on use cases or a general approach using a generalization concept in which all vehicles are treated as a single lane.

[0058] Emergency driving block 3033 is designed to handle all sudden and unexpected changes during driving. This block provides emergency driving data for situations such as a police stop, a fire truck, an unexpected pedestrian, or any other sudden situation within the normal driving scenarios where the vehicle must modify its pre-established decision rules.

[0059] The data of the third type of data data3 of the module 303 of environmental evolution is therefore a combination of data from these different blocks 3031, 3032, 3033.

[0060] The data of the first type of data data1, the data of the second type of data data 2 and the data of the third type of data data 3 determine a current situation Sa in which the autonomous motor vehicle 10 is located.

[0061] The driver assistance system 30 also includes the fuzzy system 304, the reinforcement learning agent 305, the actuator control block 306, and the human-machine interface 307. The fuzzy system 304 is adapted to deliver a first command K1 to the actuator control block 306 and a second command K2 to the human-machine interface 307. These commands K1 and K2 are generated from the current situation resulting from the first type of data data1, the second type of data data2, and the third type of data data3. The fuzzy system 304 is also adapted to transmit the first command K1 and the second command K2 resulting from the current situation to the reinforcement learning agent 305. The fuzzy system 304 is adapted to receive a new control response Rep2 from the reinforcement learning agent 305.This new control response Rep2 is optimized compared to the current situation Sa. The fuzzy system 304 will be described in more detail later using the . figure 4 .

[0062] Reinforcement learning agent 305 is adapted to receive the K1 and K2 commands generated by the fuzzy system 304. Reinforcement learning agent 305 is also adapted to receive quality information (not shown) on the outcome of the first K1 command to control the autonomous vehicle. Based on this quality information, reinforcement learning agent 305 delivers the new control response Rep2 to the fuzzy system 304. The operation of reinforcement learning agent 305 will be described in more detail later using the figure 5 .

[0063] The actuator control block 306 is adapted to receive the first command K1.

[0064] Based on this command K1, an actuator is used to control the trajectory of the autonomous motor vehicle 10. This actuator is, for example, a steering column of the autonomous motor vehicle.

[0065] The 307 Human-Machine Interface is designed to communicate with a human driver. It can therefore include audible or visual signals. These signals are activated, for example, in the event of a hazard requiring a sudden change of trajectory for the autonomous vehicle.

[0066] There figure 4 illustrates in more detail the fuzzy system 304 of the driver assistance device 30. This fuzzy system 304 includes: a fuzzification block 3011; an inference engine 3012; a database 3013; a defuzzification block 3014.

[0067] The fuzzification block 3011 is designed to receive and process multiple data points of different data types (data1, data2, data3) in order to determine the current situation Sa faced by the autonomous vehicle 10. Within this block, the multiple data points of different data types (data1, data2, data3) are translated into linguistic values ​​representing the current situation Sa. This translation depends on membership functions that represent the subjective knowledge of the human driver. The current situation Sa is then transmitted to the database 3013 and the inference engine 3012.

[0068] Database 3013 is designed to store inference rules. These inference rules were formulated based on human experience. For example, if the vehicle's mass is "heavy" and the tires are "worn," the risk is "high" and the decision is "conservative." Database 3013 is designed to transmit the appropriate inference rule (Rule) to the inference engine (3012) for the current situation (Sa). Furthermore, database 3013 is designed to receive the new control response (Rep2) from the reinforcement learning agent (305). This new response (Rep2) allows the inference rules to be updated to optimize the response for the current situation (Sa) should it occur again in the future.

[0069] The 3012 inference engine is adapted to receive the inference rule Rule from the 3013 database. The 3012 inference engine then applies this inference rule to the current situation Sa to generate an appropriate control response Rep1. For example, the 3012 inference engine might generate a control response Rep1 of the form: the vehicle mass is heavy and the tires are worn, the risk is high, and the decision is conservative.

[0070] The defuzzification block 3014 is adapted to receive the appropriate control response Rep1. From this control response Rep1, the defuzzification block 3014 transmits the first command K1 to the actuator control block 306 and the second command K2 to the Human-Machine Interface 307. The first command K1 and the second command K2 are also transmitted to the reinforcement learning agent 305. The defuzzification block 3014 implements the Center of Area (CoA) method to provide an average result from a conclusion of each given rule in the inference process. The first command K1 goes from fuzzy to clear values, thus providing a measure of the risk of the situation and the strategy to follow. The first command K1 includes a measurement value for the risk of the current maneuver and a decision value for the driving strategy to follow.The two values ​​lie within the range [0, 1] where 0 signifies the lowest risk and the most effective strategy, and 1 signifies the highest risk and the need to adopt the most cautious strategy. A mathematical representation of the CoA method is as follows: Out put = ∑ w i O i ∑ w i

[0071] Where wi represents the weight assigned to each rule evaluation in the inference process, and O i represents the singleton values ​​assigned to the output language values. The output corresponds to the output of the 3014 defuzzification block.

[0072] There figure 5 details the operation of the reinforcement learning agent 305.

[0073] At each cycle, the learning agent 305 receives information about the state of the environment (data1, data2, data3). From this information, agent 305 selects a decision to execute that modifies the state of the environment and results in a reward. The problem that learning agent 305 seeks to solve is composed of the following elements: a set S of possible environmental states, a set A of possible actions a, and a set R of rewards r. However, it is important to emphasize that, despite the immediate reward r, the real objective is to find a policy or action tactic that maximizes a final reward rf. Considering that driving is a learning process, the application of this technique fits perfectly within the vehicle's autonomous decision-making module. For learning, learning agent 305 utilizes Q-learning.Q-learning is a reinforcement learning method based on calculating a quality value Q associated with the actions available in a given state. This quality estimation is performed iteratively, i.e., by updating a value Q(s,a) at each cycle where action a is selected in that state. The update is done using the following equation: . Q s t a t = Q s t a t + α r t + β a t + 1 max Q s t + 1 a t + 1 − Q s t a t

[0074] In which st and at are the state and action taken at time t, and rt is the reward obtained for that action. The variable α is a parameter known as the learning rate, which refers to the speed at which the algorithm adjusts its knowledge based on experience. The value of this variable α is between 0 and 1, with this value being proportional to the desired rate. The variable β is a discount factor for the algorithm that establishes the priority of the long-term reward over the immediate reward. The value of this variable β is between 0 and 1, where 1 represents the highest priority of the long-term reward. Once the function Q converges to its final value, the optimal policy of the system can be obtained by selecting, in each state, the action with the maximum value of Q. This policy is expressed by the term max at+1 (Q(st+1, at+1)).

[0075] The main advantage of this reward-based approach is that it does not depend on the user defining a pre-defined evolutionary model or control policy, but is based directly on the system's interaction with its environment and the reward received at each stage. Furthermore, it perfectly aligns with the learning objective of this device, as it only intervenes when encountering a situation it has already faced, rather than using learning for new situations based on the multi-evolutionary decision block. This mirrors human behavior, where more aggressive maneuvers are employed for driving situations encountered in the past, while a more conservative approach is used for new driving situations / scenarios.

[0076] The method of assisting the driving of the motor vehicle shown in Figure 10 is described below in support of the figures 1 à 5 and of the figure 6 .

[0077] On the figure 6 , in a reception step E1, a plurality of data data1, data2, data3 are received by the fuzzification block 3011.

[0078] In a processing step E2, the received data data1, data2, data3 are processed by the fuzzification block 3011 to determine the current situation Sa of the autonomous motor vehicle 10.

[0079] In an E3 step, a control response Rep1 is generated by the inference engine 3012 with respect to the current situation Sa.

[0080] In step E4, a command K1 is sent by the defuzzification block 3014 to control the autonomous vehicle 10. This command K1 is a function of the response to be given Rep1. During this step E4, the command K1 is also transmitted to the reinforcement learning agent 305.

[0081] In a reinforcement generation step E5, the new control response Rep2 is generated by the reinforcement learning agent 305. This new control response Rep2 is based on quality information about the outcome of the command K1 for controlling vehicle 10. It is optimized relative to the current situation Sa. This new control response Rep2 is then transmitted to the database 3013 for updating the inference rules.

[0082] The receiving steps E1 and processing E2 of the plurality of data data1, data2, data3, the step E3 of generating a control response Rep1 and the step E4 of sending the command K1 are carried out according to a fuzzy logic.

[0083] The vehicle driver assistance method 10 and the associated driver assistance system 30 thus make it possible to have: an intelligent decision-making system linked to a vehicle evolution module relating to any potential situation likely to modify the vehicle's dynamics; a learning implemented by fuzzy logic and reinforcement learning.

[0084] The invention also relates to a computer program product comprising program instructions that can be used by the driver assistance device 30, which, when executed or interpreted by said driver assistance device 30, trigger the implementation of the driver assistance method.

[0085] The invention also relates to a motor vehicle 10 comprising the driver assistance device 30. This motor vehicle is a passenger vehicle, as shown in the figure 1 and the figure 2 Alternatively, a motor vehicle is a completely different type of vehicle such as a bus or a truck.

[0086] The invention is not limited to the embodiments and variants shown, and other embodiments and variants will be obvious to a person skilled in the art.

Claims

1. Method for assisting in the driving of an autonomous motor vehicle (10) on a road, said driving assistance method comprising: - a step (E1) of reception of a plurality of data (data1, data2, data3), said data comprising a first type of data (data1), a second type of data (data2) and a third type of data (data3), the data of the first type of data concerning an evolution of the vehicle, the data of the second type of data concerning an evolution of the road, the data of the third type of data concerning an evolution of the environment; - a step (E2) of processing of the plurality of the received data (data1, data2, data3) to determine a current situation (Sa) of the autonomous motor vehicle (10); - a step (E3) of transmission, by a database (3013) to an inference engine (3012), of an inference rule suited to said current situation (Sa) and of generation, by the inference engine (3012), of a control response (Rep1) from said inference rule; - a step (E4) of sending of a command (K1) to control the path of the autonomous motor vehicle (10), said command (K1) being dependent on said response to be made (Rep1); - the method being characterized in that it comprises, based on an item of quality information on the result of the command to control the path of the autonomous motor vehicle (10), a step (E5) of generation, by reinforcement learning, of a new control response (Rep2), said new control response (Rep2) being used to update the inference rules of the database (3013).

2. Driving assistance method according to Claim 1, wherein the data of the first type of data (data1) comprise data selected from among the list of the following data: - data on change of weight; - data on change of tyre; - data on change of the status of the autonomous motor vehicle.

3. Driving assistance method according to either of the preceding claims, wherein the data of the second type of data (data2) comprise data selected from among the list of the following data: - data on weather conditions; - data on the type of road; - data on the state of the road.

4. Driving assistance method according to any one of the preceding claims, wherein the data of the third type of data (data3) comprise data selected from among the list of the following data: - free driving data; - nominal driving data; - emergency driving data.

5. Driving assistance method according to any one of the preceding claims, wherein the steps of reception and of processing of the plurality of data (data1, data2, data3), the step of generation of a control response (Rep1) and the step of sending of the command (K1) are performed according to a fuzzy logic.

6. Device for assisting in the driving of an autonomous motor vehicle on a road, said device (30) comprising: - a fuzzy system (304) comprising: - a fuzzying block (3011) suitable for receiving and processing a plurality of data (data1, data2, data3), in order to determine a current situation (Sa) of the autonomous motor vehicle (10), said data comprising a first type of data (data1), a second type of data (data2) and a third type of data (data3), the data of the first type of data concerning an evolution of the vehicle, the data of the second type of data concerning an evolution of the road, the data of the third type of data concerning an evolution of the environment; - an inference engine (3012) suitable for receiving an inference rule suited to the current situation (Sa) from a database (3013) in order to generate a control response (Rep1) from said inference rule; - a defuzzying block (3014) suitable for commanding an actuator (K1) in order to control the path of the autonomous motor vehicle (10), according to the control response (Rep1) to be made; - a reinforcement learning agent (305), said agent being suitable for generating a new response (Rep2) according to an item of quality information on the result of the command (K1) for the control of the path of said autonomous motor vehicle (10), said new response being used to update the inference rules of the database (3013).

7. Computer program product comprising program instructions that can be used by the driving assistance device (30) of Claim 6, which, when they are executed or interpreted by said driving assistance device (30), trigger the implementation of the driving assistance method according to any one of Claims 1 to 6 in an autonomous motor vehicle (10).

8. Autonomous motor vehicle comprising a driving assistance device (30) according to Claim 6.