WO2022243151A1

WO2022243151A1 - Method for locating autonomous vehicles

Info

Publication number: WO2022243151A1
Application number: PCT/EP2022/062901
Authority: WO
Inventors: Andrea Ancora
Original assignee: Renault S.A.S.
Priority date: 2021-05-17
Filing date: 2022-05-12
Publication date: 2022-11-24
Also published as: FR3122920A1; EP4348179A1; FR3122920B1

Abstract

The invention relates to a method for locating an autonomous vehicle (100) comprising position sensors (2), a path planning system (33), and motion actuators (41), characterised in that it comprises an iteration of the following steps: - a step (E1) of predicting the current position of the autonomous vehicle (100) which comprises receiving (E11) a previous estimated position (62) of the autonomous vehicle (100) obtained from a previous iteration of the method, receiving (E12) applied commands (67) by means of the motion actuators (41, 42, 43), calculating (E13) a commanded path (74) on the basis of the applied commands (67), and calculating (E14) an extrapolated position (71) of the autonomous vehicle (100) on the basis of said previous estimated position (62) and said commanded path (74); - an updating step (E2) which comprises calculating an updated estimated position (72) of the autonomous vehicle (100) taking into account a measured position (61) obtained from the position sensors (2) and the extrapolated position (71).

Description

TITLE: Localization process for autonomous vehicle

The invention relates to a localization method for an autonomous vehicle. The invention also relates to a location device for an autonomous vehicle. The invention also relates to a computer program implementing the mentioned method. The invention finally relates to a recording medium on which such a program is recorded.

In the context of autonomous driving systems, the precision of the location of the autonomous vehicle conditions its operational reliability and the safety of the passengers.

The location of the vehicle is generally derived from a GNSS-type satellite positioning system and vehicle movement estimates from sensors placed on the chassis.

Alternative technologies have more recently been developed to determine the position of the autonomous vehicle from multiple sensors, such as cameras, radars or lidars present on the vehicle.

Patent WO2018/175441 A1 describes a navigation system for an autonomous vehicle taking into account not only satellite positioning and sensors (speed sensors, cameras, suspension sensors) but also the planned trajectory for the vehicle. The planned trajectory is determined from the camera measurements, making it possible, for example, to estimate the shape of the road and to define—according to behaviors learned by the vehicle—a short-term trajectory adapted to the shape of the road. Thus the system described in this patent predicts the location of the autonomous vehicle by applying a processing to the data from the sensors (the processing possibly being the application of a Kalman filter or neural network modeling) to estimate the location of the vehicle along a short-term planned trajectory.

However, this solution has drawbacks, in particular its precision and its reliability can be improved.

The object of the invention is to provide a location device and method for autonomous vehicles remedying the above drawbacks and improving the location devices and methods for autonomous vehicles known from the prior art. In particular, the invention makes it possible to produce a device and a method which are simple and reliable.

To this end, the invention relates to a method for locating an autonomous vehicle comprising position sensors, a trajectory planning system, and motion actuators.

The method includes an iteration of the following steps:

- a step of predicting the current position of the autonomous vehicle comprising a reception of a previous estimated position of the autonomous vehicle resulting from a previous iteration of the method, a reception of commands applied by the movement actuators, a calculation of a trajectory controlled according to the commands applied, and a calculation of an extrapolated position of the autonomous vehicle according to said previous estimated position, and said commanded trajectory,

- an updating step comprising a calculation of an updated estimated position of the autonomous vehicle taking into account a measured position from the position sensors and the extrapolated position.

The method can comprise a step of determining a trajectory planned by the trajectory planning system, and the calculation of an extrapolated position of the autonomous vehicle can be carried out according to the planned trajectory. The method may comprise a determination of a confidence index associated with the extrapolated position as a function of the calculated difference between the planned trajectory and the commanded trajectory and the step of calculating the updated estimated position may take account of the index of confidence.

The prediction step may include the implementation of a Kalman filter with which is associated a noise covariance matrix calculated as a function of the difference between the planned trajectory and the commanded trajectory.

The calculation of the confidence index can depend on the noise covariance matrix of the Kalman filter.

The updated estimated position can be a weighted average of the measured position and of the extrapolated position, the weighting coefficients being able to be a function of the confidence index of the extrapolated position.

The invention further relates to a tracking device for an autonomous vehicle, the device comprising a computing unit, position sensors, a trajectory planning system and motion actuators. The device comprises hardware and/or software elements implementing the method as defined previously, in particular hardware and/or software elements designed to implement the method according to the invention, and/or the device comprising means of implement the method as defined above.

The invention further relates to an autonomous vehicle comprising a location device according to the invention. The invention also relates to a computer program product comprising program code instructions recorded on a computer-readable medium for implementing the steps of the method as defined above when said program is running on a computer. The invention also relates to a computer program product downloadable from a communication network and/or recorded on a data carrier readable by a computer and/or executable by a computer, comprising instructions which, when the program is executed by the computer, lead it to implement the method as defined previously.

The invention also relates to a data recording medium, readable by a computer, on which is recorded a computer program comprising program code instructions for implementing the method as defined previously. The invention also relates to a computer-readable recording medium comprising instructions which, when executed by a computer, lead the latter to implement the method as defined previously.

The invention also relates to a signal from a data medium, carrying the computer program product as defined previously.

The appended drawing shows, by way of example, an embodiment of a location device according to the invention and an embodiment of a location method according to the invention.

[Fig. 1] Figure 1 shows an autonomous vehicle equipped with a tracking device.

[Fig. 2] Figure 2 shows an embodiment of the location device. [Fig. 3] FIG. 3 represents a flowchart of an embodiment of a location method.

[Fig. 4] Figure 4 represents an iteration loop of steps of the localization process.

[Fig. 5] Figure 5 represents an architecture of a localization process prediction module.

An example of an autonomous vehicle 100 equipped with a location system is described below with reference to FIG.

The autonomous vehicle 100 can be an autonomous vehicle of any type, in particular a passenger vehicle, or a utility vehicle or even a public transport vehicle.

The autonomous vehicle 100 comprises a chassis 4 and a tracking system 10.

The location system 10 mainly comprises the following elements:

- a set of 2 position sensors,

- a navigation system 3,

- and a calculation unit 7 comprising a microprocessor 1, a memory 5 and communication interfaces 6 allowing the microprocessor 1 to communicate with the set of position sensors 2 and the navigation system 3.

The set of position sensors 2 can comprise different types of sensors, including a satellite or GNSS positioning system, a set of cameras and/or lidars and/or radars. The cameras can be placed on the vehicle and also in the environment of the vehicle. Other embodiments of position sensors 2 may, for example, include communication systems between vehicles (V2V system).

The measurements from the position sensors 2 include measurements 61 of the position of the vehicle. The position measurements 61 can be absolute, for example when they come from a GNSS type system. The position measurements 61 can also be relative measurements, in particular when they relate to a movement from a known initial position, the movement of the vehicle being able for example to be calculated from measurements of rotation of the wheels of the vehicle.

The measurements from the position sensors 2 comprise measurements 68 of the environment of the autonomous vehicle. The measurements 68 include measurements of the position of the various elements present in the environment of the vehicle. The processing of measurements 68 can also provide other parameters detailing the state of an object in the scene: its attitude (angular position), its speed vector, but also other attributes such as its class (pedestrian, car, truck , cycle, etc...) or its dimensions.

The chassis 4 includes the various actuators which implement the movement of the autonomous vehicle, in particular an engine torque actuator 41, a brake actuator 42 and a steering wheel rotation actuator 43.

The chassis receives control commands 66 from the navigation system 3 in order to move the vehicle along a planned trajectory 64 by the navigation system 3.

In the rest of the document, the term “trajectory” is used to designate the temporal evolution of a state vector defining the characteristics of the movement of the motor vehicle 100. In a preferred embodiment, the state vector comprises a location, in particular x,y coordinates, longitudinal and lateral velocities and/or longitudinal and lateral accelerations and/or yaw rate. In the remainder of the document, the term “position” is used to designate either the x, y coordinates of the state vector, or the state vector as a whole.

The commands actually applied by the actuators 41, 42, 43, named in the remainder of the document "applied commands", may however differ from the control commands 66 received from the navigation system 3. Indeed, certain parameters may disturb the implementation commands 66. For example, the state of the road, the wear of the tires, or the weather conditions can interfere with the implementation of the commands 66. The chassis 4 transmits in return the applied commands 70 to the navigation system 3.

Thus the navigation system 3 is aware of the commands actually applied by the actuators. In general, the commands applied 70 by the actuators 41, 42 and 43 are very close to the command commands 66. However, there are situations where the commands applied 70 are significantly different from the command commands 66. In this case, the trajectory implemented by the autonomous vehicle 100 deviates substantially from the planned trajectory 64. The location system 10 according to the invention aims to improve the location of the autonomous vehicle, in particular in situations where the real trajectory of the vehicle autonomous deviates significantly from the planned trajectory.

An embodiment of the navigation system 3 is described in detail in FIG. 2. The navigation system 3 receives as input, on the one hand, the measurements 68 of the environment of the autonomous vehicle, and, on the other hand, an updated estimated position 72 of the autonomous vehicle 100. The navigation system 3 includes an environmental data fusion module 31, which aggregates the data from the various sensors and relating to the environment of the autonomous vehicle 100.

The merged data 69 is then processed in a situation analysis module 32, advantageously comprising the determination of an electronic horizon of the vehicle. The electronic horizon is a local map of the elements of the driving scene including the information of the area useful for driving. In other words, the electronic horizon includes the precise positions relative to the vehicle of all the elements of the driving scene that will allow the vehicle to make the relevant decisions for driving.

The data 63 from the situation analysis module 32 are then used in a module for planning a short-term trajectory 33, to determine a short-term trajectory 64. The order of magnitude of the duration of the trajectory short term 64 is seconds.

The short-term trajectory 64 thus determined is transmitted to the control laws 34, in order to be transformed into a command order 66 transmitted to the chassis 4.

As previously explained, after transmission of the command orders 66 to the chassis 4, the latter transmits in return the applied commands 70 to the navigation system 3.

The commands applied 70 are transmitted to the computer 1 by the navigation system 3.

In one embodiment, the computer 1 makes it possible to execute software comprising the following modules: - a module 10 for determining a planned trajectory, which collaborates with the navigation module 3,

- a module 11 for predicting a current position of the autonomous vehicle 100, which collaborates with the navigation module 3,

- an update module 12, which collaborates with the set of sensors 2 and the navigation module 3.

An embodiment of the location method for an autonomous vehicle is described below with reference to FIGS. 3 to 5. The method comprises a step E0, followed by an iteration over two steps E1 and E2.

In step E0, a planned trajectory 64 is determined. The planned trajectory 64 is defined by the module for planning a short-term trajectory 33 according to the data from the situation analysis module 32. The planned trajectory 64 provides for the movements of the vehicle over a time interval of the order of a few seconds, corresponding to the sighting distance of the vehicle, that is to say the detection limit of the sensors 2, or of the electronic horizon. The planned trajectory follows an ideal curve, i.e. a curve representing a theoretical movement of the autonomous vehicle. The theoretical displacement of the vehicle can be determined by statistical models.

The planned trajectory 64 is stored in the memory 5 for its later use in the step E1.

We then move on to an iteration of steps E1 and E2.

As more specifically illustrated by Figure 4,

- the prediction step E1 predicts an extrapolated position 71 of the autonomous vehicle 100 from a previous estimated position 62, a planned trajectory 64 and a commanded trajectory 67,

- the updating step E2 calculates an updated estimated position 72 at starting from a measured position 61 and from the extrapolated position 71 resulting from step E1 of prediction.

A mode of execution of the prediction step E1 is described with reference to FIG. 5.

The first step E1 comprises a sub-step E11 of receiving an estimated position 62 of the vehicle coming from a previous iteration of the method. The estimated position 62 includes the absolute coordinates of the autonomous vehicle 100 dated at a time TN-I.

The first step E1 also includes a sub-step E12 of receiving commands 67 applied to the vehicle by the movement actuators 41, 42, 42. The commands 67 are dated at a time TCOM of command of the movement of the vehicle.

The first step E1 also comprises a sub-step E13 of calculating a commanded trajectory 74 as a function of the applied commands 67. For this, a conversion module 113 converts the commands 67 into a commanded trajectory 74.

The commanded trajectory 74 follows a noisy curve, that is to say which differs significantly from the ideal curve defined by the planned trajectory 64.

The first step E1 further comprises a sub-step E14 of calculating an extrapolated position 71. In the mode of execution described, the calculation of an extrapolated position 71 takes into account both the planned trajectory 64 and the trajectory ordered 74.

As is more specifically described by FIG. 5, the calculation of an extrapolated position 71 can implement a filter 111, taking into input a previous estimated position 62, a planned trajectory 64 and a noise covariance matrix 73.

A first mode of execution of the filter 111 can implement an extended Kalman filter, that is to say a Kalman filter based on a nonlinear model. A predicted state vector x' is then calculated according to the following Math 1 formula:

[Math 1] x' = f(x,u ) where:

- x is the state vector predicted during the previous iteration, i.e. position 62 estimated during a previous iteration,

- x' is the predicted state vector during the current iteration, corresponding to the extrapolated position 71 ,

- u is the state vector associated with the planned trajectory, i.e. the planned trajectory 64,

- f is the control function, i.e. a nonlinear function modeling the transition between two successive iterations of filter 111.

Moreover, in this first mode of execution of the filter 111 , a covariance matrix P′ associated with the predicted state vector x′ is calculated according to the following Math 2 formula.

[Math 2]

P' = FPF ^T + BWB ^T where:

- P is the covariance associated with the state vector predicted during the previous iteration, i.e. the covariance associated with position 62,

- P' is the covariance associated with the predicted state vector during the current iteration, i.e. the covariance associated with the extrapolated position 71 ,

- B is the control model, i.e. the function that converts the controls on the actuators or the trajectory in space of the state vector. This function is normally known in advance and constant over time,

- W is the noise covariance matrix 73.

In the first mode of execution of the filter 111 , the noise covariance matrix 73 is used to model the uncertainty of the prediction linked to a measured deviation between the planned trajectory 64 and the commanded trajectory 74.

In order to be able to measure the difference existing between the planned trajectory 64 and the commanded trajectory 74, it is necessary to temporally synchronize these two trajectories via a synchronization module 112. From the planned trajectory 64, a synchronized planned trajectory 65 is calculated. , so that it can be compared with the commanded trajectory 74.

The first mode of execution of the filter 111 comprises a calculation of the noise covariance matrix W according to the mathematical expressions Math 3:

[Math 3] u" = Zu

W = EÇ(u" — u')(u" — u') ^T ) where:

- u is a state vector corresponding to the planned trajectory 64,

- u' is a state vector corresponding to the applied trajectory 74,

- u” is a state vector corresponding to the synchronized planned trajectory 65,

- Z is the transformation implemented in the synchronization module

112,

- E() is the estimated function. Thus, according to the first mode of execution of the filter 111 , at the end of the step E1 , an extrapolated position 71 and an associated covariance P′ are obtained. The covariance P' thus makes it possible to calculate a confidence index IC1 associated with the extrapolated position. The confidence index IC1 can, for example, be calculated from the covariance matrix, by selecting the values on the diagonal (or another similar function), which makes it possible to determine a confidence index per state variable of the positioning module.

The extrapolated position 71 corresponds to a point M belonging to the planned trajectory 64, and the confidence index IC1 reflects the probability that the autonomous vehicle 100 is actually located at point M with a given level of precision.

In other words, in the first mode of execution of the filter 111 , the controlled trajectory 74 only influences the confidence index IC1 associated with the extrapolated position 71:

- the closer the commanded trajectory 74 is to the planned trajectory 64, the higher the confidence index IC1,

- the further the commanded trajectory 74 is from the planned trajectory 64, the lower the confidence index IC1.

Other embodiments of the filter 111 could be implemented, for example by using a linear Kalman filter, or a particle filter. Alternatively, the Kalman filter could be replaced by artificial intelligence using, for example, Machine Learning algorithms or neural networks.

In addition or alternatively, in other modes of execution of step E1, the calculation of an extrapolated position 71 could take into account the commanded trajectory 74 for the calculation of the coordinates of the autonomous vehicle. In a second updating step E2, an updated estimated position 72 is calculated taking into account, on the one hand, a measured position 61 coming from the sensors 2 and, on the other hand, the extrapolated position 71 .

Step E2 includes a sub-step of receiving data from the sensors 2, and determining a measured position 61 at a given instant TMES. The reception sub-step comprises a fusion of the data coming from different sensors. In a preferred mode of execution, the data fusion makes it possible to associate a confidence index IC2 with the measured position 61. The confidence index notably takes into account the different sources of data taken into account in the data fusion. The confidence index can also take into account visibility conditions related to road infrastructure (for example, the presence of a tunnel or a sharp bend), or related to light or weather conditions. The confidence index can also take into account the environment of the vehicle, in particular areas of reflection that can disturb the measurements of the sensors.

Step E2 further comprises a sub-step of calculating an updated estimated position 72 from the measured position 61 derived from the sensors and from the extrapolated position 71 calculated in step E1.

In one embodiment, the updated estimated position 72 is a weighted average of the measured position 61 and the extrapolated position 71.

Both of the measured 61 and extrapolated 71 positions are advantageously associated with a confidence index, IC2, IC1.

The confidence index IC1 represents the degree of confidence associated with the planned trajectory 64; in particular the confidence index IC1 provides a measurement of the consistency between the planned trajectory 64 and the commanded trajectory 74.

Advantageously, the weighting coefficients of the weighted average are determined according to the confidence index IC1. For example, if the confidence index of IC1 is lower than a given threshold then the weighting coefficient associated with the extrapolated position 71 is strictly lower than the weighting coefficient associated with the measured position 61 . In other words, if the planned trajectory 64 is substantially far from the commanded trajectory 67, then the measured position 61 will have more weight than the extrapolated position 71 in the calculation of the updated estimated position 72.

In addition, the calculation of the weighting coefficients could take into account the two confidence indices IC1, IC2, in particular the relative level of the confidence indices IC1, IC2. For example, the weighting coefficient of each position 61, 71 could be proportional to the confidence index IC2, IC1 respectively associated with said position. Alternatively, this method of calculation could only apply if the ratio between the highest confidence index and the lowest confidence index is below a given threshold, for example 120%. Beyond the given threshold, one could for example assign a very low coefficient, or even zero, to the position of the lowest confidence index.

Thus, the calculation of a weighted average between these two positions aims to improve the precision of the location, in particular by taking into account the inherent uncertainty on the one hand in the measurements of the sensors, and on the other hand in a position determined by extrapolation from a planned trajectory.

Finally, the location method according to the invention has the advantage of detecting and processing as soon as possible the differences between the trajectory planned and the commanded trajectory, i.e. the trajectory actually implemented by the vehicle's motion actuators. In other words, instead of considering that the vehicle is moving on an ideal trajectory, defined according to statistical models by trajectory planning, the deviation between the commanded trajectory and the planned trajectory is evaluated.

This processing is all the more relevant since, unlike the planned trajectory which is defined over a time window of the order of a few seconds, the commanded trajectory is updated every 10 milliseconds, which is comparable to the update frequency measurements from the sensors.

Furthermore, while the data from the sensors describe a past situation, the commanded trajectory corresponds to a very short-term extrapolation of the future position of the vehicle from the commands applied by the motion actuators.

In an autonomous vehicle context, the objective of the invention is to apply a planned trajectory while improving the prediction of the position of the autonomous vehicle. Thus, instead of being in a reactive mode (in which one reacts to measurements), the invention makes it possible to be proactive by taking into account the maneuvers to come.

In alternative embodiments, the commanded trajectory could be used in the calculation of the vehicle's position itself.

Claims

1. Method for locating an autonomous vehicle (100) comprising position sensors (2), a trajectory planning system (33), and movement actuators (41), characterized in that it comprises an iteration of the following steps:

- a step (E1) of predicting the current position of the autonomous vehicle (100) comprising a reception (E11) of a previous estimated position (62) of the autonomous vehicle (100) resulting from a previous iteration of the method, a reception (E12) of commands applied (67) by the movement actuators (41, 42, 43), a calculation (E13) of a commanded trajectory (74) according to the commands applied (67), and a calculation (E14) an extrapolated position (71) of the autonomous vehicle (100) as a function of said previous estimated position (62), and of said commanded trajectory (74),

- an updating step (E2) comprising a calculation of an updated estimated position (72) of the autonomous vehicle (100) taking into account a measured position (61) from the position sensors (2) and the extrapolated position (71 ).

2. Method according to the preceding claim, characterized in that it comprises a step (E0) of determining a planned trajectory (64) by the trajectory planning system (33), and in that said calculation (E14) of an extrapolated position (71) of the autonomous vehicle (100) is carried out according to said planned trajectory (64).

3. Method according to the preceding claim, characterized in that it comprises a determination of a confidence index (IC1) associated with the extrapolated position (71) according to the difference calculated between the planned trajectory (64) and the commanded trajectory (74) and in that the step of calculating the updated estimated position (72) takes account of the confidence index (IC1).

4. Method according to claim 2 or 3 characterized in that the prediction step (E1) comprises the implementation of a Kalman filter (111) with which is associated a noise covariance matrix (73) calculated as a function the difference between the planned trajectory (64) and the commanded trajectory (74).

5. Method according to the preceding claim, characterized in that the calculation of the confidence index (IC1) depends on the noise covariance matrix of the Kalman filter (111).

6. Method according to one of claims 3 to 5, characterized in that the updated estimated position (72) is a weighted average of the measured position (61) and of the extrapolated position (71), the weighting coefficients being a function the confidence index (CI1) of the extrapolated position (71).

7. Localization device (10) for an autonomous vehicle, the device comprising a calculation unit (7), position sensors (2), a trajectory planning system (33) and movement actuators (41, 42, 43) implementing the method according to one of the preceding claims, in particular a calculation unit (7) capable of implementing the method according to one of the preceding claims.

8. Autonomous vehicle (100) comprising a location device according to the preceding claim.

9. Computer program product comprising program code instructions recorded on a computer-readable medium to implement the steps of the method according to any of the claims 1 to 6 when said program runs on a computer.

10. Support (5) for recording data, readable by a computer, on which is recorded a computer program comprising program code instructions for implementing the method according to one of claims 1 to 6.