CN112644485A

CN112644485A - Control of autonomous vehicles

Info

Publication number: CN112644485A
Application number: CN202011065890.6A
Authority: CN
Inventors: 亚历山大·卡特里尼克
Original assignee: Ford Global Technologies LLC
Current assignee: Ford Global Technologies LLC
Priority date: 2019-10-09
Filing date: 2020-09-30
Publication date: 2021-04-13
Also published as: DE102019127176A1

Abstract

The invention provides a method for controlling an autonomous vehicle at low speed by means of a model of the interaction between the vehicle and at least one unprotected road user, wherein a trajectory relating to future occurring movements of the unprotected road user is calculated, and on the basis of this control instructions are planned and the planned control instructions of the autonomous driving vehicle are guided in an information loop into the calculation of the trajectory relating to future occurring movements of the unprotected road user before the control instructions modified in response thereto are transmitted to the actuators of the autonomous vehicle. Furthermore, an autonomous vehicle for performing the method is provided.

Description

Control of autonomous vehicles

Technical Field

The invention relates to a method for controlling an autonomous vehicle (autonomenfahrzengs) by means of a model of the interaction between the vehicle and an unprotected road user.

Background

In the vicinity of parking areas unprotected road users, i.e. those who are not inside the protective body, such as passengers and/or cyclists, are particularly endangered. This applies in particular in the case of autonomous vehicles. For autonomous vehicles, detecting unprotected road users and predicting their expected movement is critical to safe operation. In this case, interaction between the autonomous vehicle and the unprotected road users is indispensable, since the actions of the vehicle may affect the unprotected road users and vice versa. An autonomous vehicle is a motor vehicle which is driven by an internal combustion engine, an electric motor and/or a hybrid drive and which can be driven without the influence of a human driver.

In road traffic, the interaction between an autonomous vehicle and an unprotected road user or the priority of one of the road users is specified by road signs. However, in parking areas, the situation is different, since here the priority of the road users is usually unclear. Thus, interaction is difficult, for example, when the autonomous vehicle wishes to drive into a parking area. If an unprotected road user wishes to grant priority to an autonomous vehicle, and thus the unprotected road user remains stationary, the wish cannot be detected by a simple control strategy; thus, the vehicle will stop. The unprotected road user may then assume that the autonomous vehicle has stopped in order to give him priority.

In conventional control strategies, future occurring movements of unprotected road users are calculated as trajectories independent of the motion of the respective autonomous vehicle. In this case, the most adverse behavior of the unprotected road user for this trajectory is generally assumed. Thus, the control strategy of an autonomous vehicle is very conservative in the sense of avoiding collisions with unprotected road users.

For example, a pre-calculated future occurring movement of a pedestrian is described in the document CN 108172025 a. Wherein this information is used in a driver assistance system for warning the driver.

In document DE 102017210394B 3 pedestrians are sensed in the parking area and their future occurring movements are predicted. Here, the input information comes from a camera, vehicle-to-vehicle communication or remote control processes. In this case, the potential trajectories are calculated with a certain probability, wherein the calculation is performed by means of an artificial intelligence method.

In document GB 2562049, artificial intelligence methods are also used to determine potential trajectories for other vehicles and pedestrians, which are input into the planning of further movements of the autonomous vehicle or of the driver warning strategy. Other methods for the calculation of autonomous vehicles and future occurring movements can be found in documents, e.g. JP 6103265B 2, US 9,604,639B 2, US 9,766,626B 1 and US 10,055,652B 2.

None of the conventional strategies take into account the interaction between the autonomous vehicle and the unprotected road users. The aim is to improve the conventional strategy for autonomous vehicle behaviour in parking areas.

Disclosure of Invention

This object is achieved by a method having the features of claim 1 and by a motor vehicle having the features of claim 10. Further advantageous embodiments and refinements of the invention emerge from the dependent claims, the figures and the embodiments. The subject matter of the embodiments and claims can be advantageously combined with one another here.

A first aspect of the invention relates to a method for controlling an autonomous vehicle at low speed by means of a mathematical model involving the interaction of the vehicle with at least one unprotected road user in the surroundings of the autonomous vehicle, the method comprising the steps of:

-sensing an unprotected road user by at least one sensor of the vehicle,

-pre-calculating future occurring movements of unprotected road users in the form of at least one trajectory,

planning control instructions relating to the vehicle travel taking into account the trajectory in order to avoid a collision,

-issuing control commands to actuators of the vehicle,

it is characterized in that the preparation method is characterized in that,

the control commands are entered into a pre-calculation of future occurring movements of unprotected road users.

The core of the invention is the interaction of the autonomous vehicle with unprotected road users. In the method according to the invention, the interaction between the autonomous vehicle and the unprotected road user is advantageously incorporated into the calculation of future occurring movements of the unprotected road user. In this case, the mathematical model is based on an artificial neural network. It is essential to the invention that the planned control actions of the autonomous vehicle are included in the prediction of the movement of the unprotected road user. As a result, the movement behavior of the autonomous vehicle, in particular in the parking area, corresponds more to the movement behavior of the human driver than a conservative automation system.

Here, the surroundings of the autonomous vehicle comprise areas which may collide with unprotected road users when the vehicle is moving in a planned manner.

Low speed means that the autonomous vehicle travels slowly in a manner suitable for parking area conditions, compared to traveling on a free road. For example, a low speed may be considered as a walking speed, e.g. 4-6 km/h.

The method according to the invention can be explained on the basis of two modules. In a first module, future occurring movements of unprotected road users are calculated. In this case, one or more (N ≧ 1) trajectories are determined, wherein a probability level is preferably calculated for each pre-calculated trajectory of unprotected road users. In the second module, a decision is made as to which control commands are issued to the respective actuators of the vehicle. The control command is preferably selected from the group consisting of an actuating operation of the steering system, an actuating operation of the power control of the driver, and an actuating operation of the brake. The planned control commands of the autonomous vehicle are then input into the movement prediction of the unprotected road user. In the case of an internal combustion engine, the control command may also relate to a throttle. In an alternative embodiment, the method may also be performed using a module having the features of both modules described above.

The convergence action of the control commands is preferably checked before issuing to the respective actuators. This advantageously ensures that collisions with unprotected road users can be avoided. In this case, the method is preferably repeated starting from a pre-calculation of future occurring movements of the unprotected road user without convergence. This means that there is an iteration between module 1 and module 2 until the control instruction is no longer altered and convergence is thus achieved.

The method is preferably carried out in a parking area. This method is particularly advantageous in parking areas, since the behavior of unprotected road users in complex surroundings comprises more behavior than conventional methods due to the desired interaction between the autonomous vehicle and the unprotected road user. This increases the safety of unprotected road users and improves the performance of the autonomous vehicle in terms of comfort and time required to complete operations in the parking area.

The unprotected road user is preferably a pedestrian. The method is particularly advantageously suitable for continuing and ending the starting movement of the autonomous vehicle on the basis of the interaction between the autonomous vehicle and an unprotected road user, and at the same time ensuring the safety of pedestrians.

The sensor for sensing unprotected road users is preferably selected from the group comprising a camera, a radar, a lidar and devices for vehicle-to-vehicle, vehicle-to-pedestrian and vehicle-to-cyclist communication. Unprotected road users can periodically transmit status data, for example, via a mobile radio.

In this method, the autonomous vehicle preferably moves at an approximately constant speed. As a result, unprotected road users advantageously gain a clear understanding of the planned movement of the vehicle. Unlike the conventional method in which the autonomous vehicle travels at a variable speed and stops traveling upon sensing an unprotected road user, the movement of the vehicle is relatively quickly and safely ended here.

A second aspect of the invention relates to an autonomous vehicle having a control device in which a first module is used to calculate in advance the trajectory of an unprotected road user by means of a mathematical model based on an artificial neural network, and a second module is used to calculate control instructions for the movement of the autonomous vehicle, and the first and second modules are designed to control the method according to the invention.

The advantages of the autonomous vehicle correspond to the advantages of the method according to the invention.

Drawings

The invention will be explained in more detail with reference to the drawings, in which

Figure 1 shows traffic situation in a parking area of an embodiment of a vehicle according to the invention,

figure 2 shows a flow chart of a conventional method for controlling maneuvering of an autonomous vehicle,

fig. 3 shows a flow chart of an embodiment of a method according to the invention, an

Fig. 4 shows a modular illustration of an apparatus for performing the method according to the invention.

Detailed Description

Fig. 1 shows traffic conditions in a parking area 1. Here, the parking area 1 includes a plurality of parking spaces. The first vehicle 10 is parked between two vehicles 11 and the other vehicle opposite. The first vehicle 10 is an autonomous vehicle. The direction of movement of the first vehicle 10 is shown by the dashed arrow from the first vehicle 10.

The first vehicle 10 has at least one sensor 20 in the rear region, which sensor 20 is designed to sense the surroundings of the first vehicle 10. The sensor 20 may be a camera, radar or lidar, but is not so limited. In this case, the above-described embodiments of the sensor 20 may also all be present and arranged at different locations on the first vehicle 10, i.e. in the immediate vicinity of the front region, for example on the rear region and on the side of the first vehicle 10.

Furthermore, the first vehicle 10 is designed for vehicle-to-vehicle communication, vehicle-to-pedestrian communication, and vehicle-to-cyclist communication. Here, the corresponding devices of the first vehicle 10 are also included under the term sensor 20. Unprotected road users can periodically transmit status data, for example, via a mobile radio.

The sensor 20 is connected to a control device 30. A first module 31 is implemented in the control device 30, which first module 31 is used to pre-calculate the trajectory of an unprotected road user by means of a mathematical model based on an artificial neural network. Furthermore, a second module 32 for calculating a control command for moving the first vehicle is implemented in the control device 30. The

modules

31, 32 are understood to be separate software interacting with each other in the control device 30, wherein the function of the second module 32 is based on the information of the first module 31. The

modules

31, 32 may also be spatially separated, i.e. for example the first module may also be arranged outside the control device 30, for example on a hard disk.

The unprotected road user in parking area 1 is a pedestrian 40. The direction of movement of the pedestrian 40 is illustrated by the trajectories T1, T2, T3 in the form of dashed arrows from the pedestrian. The trajectories T1, T2, T3 here mean possible movement paths that the pedestrian may follow from its current position. In this case, the probability level of the trajectory actually followed by the pedestrian is calculated for the trajectory. Here, the first trajectory T1 has the highest probability level, the second trajectory T2 has the second highest probability level, and the third trajectory T3 has the lowest probability level.

FIG. 2 illustrates a flow chart of a conventional method for controlling movement of an autonomous vehicle in the ambient environment of a parking area. The vehicle moves out of its parking position at low speed. In this case, in a first step S1, an unprotected road user (e.g., pedestrian 40) is detected by a sensor of the vehicle. The sensor data is transmitted to a control device of the vehicle. In a second step S2, future movements of the unprotected road user are calculated in the form of at least one trajectory, in particular independent of the vehicle. In this case, it is also possible to calculate a plurality of trajectories and the probability that the unprotected road user can actually follow the respective trajectory. In the third step S3, control commands relating to vehicle travel are planned while taking into account the trajectory in order to avoid a collision. In this case, the trajectory with the highest probability to be followed by the unprotected road user is selected. In a fourth step S4, corresponding control commands are issued to the actuators of the vehicle to control, for example, steering, acceleration and braking of the vehicle.

If there is no convergence between the effects of the planned control commands, in contrast to the conventional method according to fig. 2, in an embodiment of the method according to the invention the information loop returns from step S3 to step S2 according to the illustration in fig. 4. In this case, information about the planned control instructions is input into the calculation of the trajectory and its probability level. The predicted trajectory of the movement of the pedestrian 40 and its corresponding probability level are modified based on the planned movement of the first vehicle 10. The control instruction is subsequently modified in step S3. In this case, the pedestrian 40 and the first vehicle 10 change their actions, for example, using game theory, until a persistent state (balance) occurs. Only then is the control command transmitted to the respective actuators of the first vehicle 10 in step S4.

A functional modular approach of the method according to the invention is shown in fig. 4. During a low speed parking area maneuver, the sensor 20 provides data to the first module 31. The first module 31 processes the data into an artificial neural network. The trajectory of future occurring movements of the unprotected road user is predicted based on the interaction between the first vehicle 10 and the pedestrian 40, and the probability of such trajectory is calculated. These calculations are provided to the second module 32. In the second module 32, control commands relating to further movement of the first vehicle 10 while avoiding a collision are calculated. In this case, the decision is made assuming that the unprotected road user of the first vehicle 10 is allowed to drive past. The control commands are input into the calculations of the first module 31 (feedback loop) before they are issued to the respective actuators. And then issues a control command to the actuator. The first vehicle 10 ends its movement at a low speed but at the same time at a speed which is as constant as possible.

List of reference numerals

1 parking area

10 first vehicle

11 other vehicles

20 sensor

30 control device

31 first module

32 second module

40 pedestrian

Claims

1. A method for controlling an autonomous vehicle (10) at low speed by means of a mathematical model involving the interaction between the vehicle (10) and at least one unprotected road user (40) in the surroundings of the autonomous vehicle (10), said method comprising the steps of:

-sensing the unprotected road user (40) by at least one sensor (20) of the vehicle (10),

-pre-calculating future occurring movements of the unprotected road user (40) in the form of at least one trajectory (T1, T2, T3),

-planning control commands relating to the vehicle travel taking into account said trajectory (T1, T2, T3) in order to avoid a collision;

-issuing control commands to actuators of the vehicle (10),

it is characterized in that the preparation method is characterized in that,

the control commands are entered into a pre-calculation of future occurring movements of the unprotected road user (40).

2. Method according to claim 1, wherein a probability level is calculated for each pre-calculated trajectory (T1, T2, T3) of the unprotected road user (40).

3. A method according to any of the preceding claims, wherein the control instruction is selected from the group comprising an actuation operation of a steering system, an actuation operation of a power control of a drive and an actuation operation of a brake.

4. Method according to any of the preceding claims, wherein the convergence effect of the control commands is checked.

5. The method according to claim 4, wherein the method is repeated starting from a pre-calculated future occurrence of movements of the unprotected road user (40) without convergence.

6. The method according to any one of the preceding claims, wherein the method is performed in a parking area (1).

7. The method according to any one of the preceding claims, wherein the unprotected road user (40) is a pedestrian (40).

8. Method according to any one of the preceding claims, wherein the sensor (20) for selecting the unprotected road user (40) is selected from the group comprising cameras, radars, lidar, and devices for vehicle-to-vehicle, vehicle-to-pedestrian and vehicle-to-cyclist communication.

9. A method according to any one of the foregoing claims, in which the vehicle (10) is moving at an approximately constant speed.

10. Autonomous vehicle (10) with a control device (30), in which control device (30) a first module (31) is used to calculate beforehand the trajectory (T1, T2, T3) of an unprotected road user (40) through a mathematical model based on an artificial neural network, and a second module (32) is used to calculate control commands for the movement of the autonomous vehicle (10), and the first module (31) and the second module (32) are designed to control a method according to any one of claims 1-9.