WO2019011536A1

WO2019011536A1 - Method, device, computer program and a machine-readable storage medium for operating a vehicle

Info

Publication number: WO2019011536A1
Application number: PCT/EP2018/064859
Authority: WO
Inventors: Thomas Lich; Thomas GUSSNER; Thomas Schlender
Original assignee: Robert Bosch Gmbh
Priority date: 2017-07-11
Filing date: 2018-06-06
Publication date: 2019-01-17
Also published as: DE102017211815A1

Abstract

A method (400) for operating a vehicle (1), comprising the steps: - detecting (401) a collision object (2), particularly a road user at increased risk, particularly a pedestrian; - determining (402) a stay probability of the collision object (2), particularly according to a prediction model for the collision object (2); - determining (403) an evasion trajectory (20, 20') in order to avoid a collision with the collision object (2) or a collision trajectory according to the determined stay probability; - controlling (404) the vehicle (1) such that the vehicle (1) is guided, automatically at least in part, along the determined evasion trajectory (20, 20') or the collision trajectory.

Description

description

title

Method, device, computer program and a machine-readable

Storage medium for operating a vehicle

State of the art

There are already many systems in the market that can prevent or reduce accidents with other road users, in particular with vehicles in longitudinal traffic and with vulnerable road users, known as Vulnerable Road Users (VRU), by means of automatic emergency braking.

However, especially at higher speeds, a collision may u. be avoided by an evasive maneuver when an emergency braking reduces the consequences of accidents only by reducing the speed.

From cellar, CG. et al: "Active Pedestrian Safety by Automatic Braking and Evasive Steering," IEEE Transactions on Intelligent Transportation Systems Vol 12 (4), 2011 systems are known that can avoid accidents by an automatic or driver-initiated evasive maneuver.

An important question that determines the feasibility of an evasive maneuver is how the collision object, e.g. the increased vulnerable road user, e.g. the pedestrian to be evaded will move in the future. Models already exist for this purpose

Describe the movement behavior of the collision object. However, these models are static in nature and take into account that at one possible

Tripping decision for an automatic emergency stop or one

Dodge the collision object, if it is, for example, a pedestrian acts, suddenly stops or flees only in a worst case

Consideration. The different reaction possibilities of the collision object, in particular if it is a pedestrian, are therefore not considered differentiated. Also the movement history of the

Collision object neglected because there are dependencies between future and historical movement behavior that can assist in a decision-making (Dodge or brakes). Thus, it can occur when initiating full braking or avoidance operations due to - possibly

Inappropriate - Worst-case consideration to come that much harder than necessary action is taken.

Disclosure of the invention

Against this background, the present invention provides and provides a method, apparatus, computer program, and machine-readable

Storage medium for operating a vehicle.

Would the possible reactions of the collision object or its

Considering movement history also in the planning of avoidance trajectories, a safe avoidance maneuver would only be possible in very few cases and would generally be too wide, since the most unfavorable possible change in movement of the collision object would have to be calculated.

Against this background, the present invention provides a method of operating a vehicle, comprising the steps of:

• detecting a collision object;

Determining a probability of residence of the collision object,

in particular depending on a prediction model for the collision object;

Determining a avoidance trajectory for avoiding a collision with the collision object or a collision trajectory as a function of the determined probability of residence; Controlling the vehicle such that the vehicle is at least partially automated along the determined avoidance trajectory or

Collision trajectory is performed.

In the present case, a collision object is understood to mean an object detected by the sensor system of the vehicle with which a collision is likely to take place if neither the movement of the vehicle nor, if possible, the movement of the object change. It is advantageous if, from the set of recognized objects only those for the method of the present invention are pursued, which approach the vehicle in critical.

In this case, a critical approximation can be understood as an approximation in which the approximation parameters, such as type of object, direction, distance, speed, etc., are mapped to a criticality value and this

Criticality value exceeds a predetermined threshold for a critical approach. As collision object can also be an increased risk

Road users are understood.

Under an increased vulnerable road users are presently understood all those participants in the road that are not surrounded by a protective shell of a body. These include u.a. Motorcyclists, cyclists and, in particular, pedestrians.

In the present case, a prediction model for the collision object is a

Understood calculation rule, which derives based on measurements of, for example, sensors of the vehicle a probability of residence of the collision object.

In the present case, an avoidance trajectory is understood as a movement path for the vehicle which is suitable for avoiding a collision with the collision object with high probability.

In the present case, a collision trajectory is understood to mean a movement path for the vehicle which is suitable in the event that a collision with the vehicle

Collision object based on the underlying model is inevitable, too a collision whose consequences are minimal, based on the underlying model.

In situations where a collision is unavoidable, the result of the collision can be mitigated by appropriately influencing the movement of the vehicle. For example. could, if the collision object is a vehicle, this vehicle could be made such that optimal absorption of the collision energy is enabled. As a result, the consequences for the occupants of the vehicles involved can be minimized.

Such movement paths (evasion trajectory, collision trajectory) can be converted into sequences for the means for longitudinal or lateral acceleration of the vehicle. In this case, under a longitudinal acceleration, both the positive acceleration, i. the actual increase in the speed of the

Vehicle, as well as the negative acceleration, i. actually reducing the speed, i. a slowing down of the vehicle, understood. In the present case, transverse acceleration is understood to mean the movement of the vehicle transversely to the longitudinal orientation of the vehicle. Depending on the drive concept of the vehicle, the lateral acceleration may also be due to steering or braking

Acceleration actions of individual wheels of the vehicle take place.

According to one embodiment of the method, the motion history of the collision object is taken into account in the prediction model.

The consideration of the movement history in the prediction model leads to an improved determination of the probability of residence of the collision object. So that a more optimal avoidance trajectory can be determined.

An evasion trajectory is considered to be more optimal in accordance with the present invention if it is capable of minimizing the probability of collision with the collision object while minimizing the effect on the movement of the vehicle. A collision trajectory is considered to be more optimal in accordance with the present invention if it is capable of minimizing the collision sequences.

According to one embodiment of the method, the predicted model considers the predicted impact point of the collision object on the vehicle.

The consideration of the predicted impact point leads to an improved determination of the probability of the collision object, since it has been shown that depending on the predicted impact point, the future behavior of the collision object, especially if the collision object is a pedestrian, is significantly influenced. When considering the predicted impact point, certain movement states of the pedestrian or, more generally, of the collision object may possibly be disregarded, this reduces the calculation resources for the avoidance trajectory and leads overall to a more accurate determination of the avoidance trajectory.

According to one embodiment of the method, in the prediction model, the collision area of the vehicle in which the predicted impact point of the vehicle

Collision object on the vehicle is considered.

To further minimize computational resources, the predicated one may be

Impact point be assigned to a collision area (Near Side, Center Area, Far Side) of the vehicle. This reduces the computation resources because the predicated behavior of the collision object is uniform within a collision area.

According to one embodiment of the method, in the step of determining the avoidance trajectory or the collision trajectory by means of a method for

Multi-objective optimization determined. It is beneficial as goals of minimizing the

Collision probability with the collision object or the minimization of the total evasion width or the minimization of the lateral acceleration or, in the case of an unavoidable collision, to use the minimization of the collision sequences. This embodiment has the advantage that methods for multi-objective optimization can be carried out efficiently.

According to one embodiment of the method, the method for

Multi-objective optimization also has at least one of the following constraints:

• The probability of collision remains below a predetermined one

Limit value for collision risk;

• stay in a predetermined driving corridor;

• remaining the lateral acceleration below a predetermined limit for the lateral acceleration;

• The remaining of the transverse offset below a predetermined limit value for the transverse offset;

• Finding the collision sequences below a predetermined limit for the collision sequences.

According to one embodiment of the method, the avoidance trajectory or the collision trajectory is represented as an nth-order polynomial, in particular of the 5th order.

An nth-order polynomial can be solved iteratively quickly by known efficient methods.

According to one embodiment of the method, the collision probability is determined as a function of a passing point.

In the present case, the passing point is understood to be the point at which the vehicle moves during the evasive maneuver or the collision at the height of the vehicle

Pedestrian is located.

According to one embodiment of the method, the prediction model is updatable.

This embodiment of the method offers the advantage that the prediction model can be adapted in advance. For example. then, if more recent findings are available or if more up-to-date data is available. The updating of the prediction model can take place, for example, during a service visit to a workshop. Also conceivable is an implementation of the updatability by a so-called. Flash Over the Air method. It will be over

Radio transmission technologies corresponding updates performed on the vehicle or vehicles.

Another aspect of the present invention is an apparatus adapted to carry out an embodiment of the method of the present invention.

In such a device may be a controller or a

Control unit network of the vehicle act. In a control unit network, various aspects of the method are performed by different controllers. Where necessary, the controllers exchange data with each other. The data exchange takes place via one of the communication networks in the vehicle, for example via the CAN or FlexRay bus.

Another aspect of the present invention is a computer program configured to carry out an embodiment of the method of the present invention.

Another aspect of the present invention is a machine-readable one

Storage medium on which the computer program according to the present invention is stored.

The invention will be explained in more detail with reference to embodiments and drawings.

drawings

Show it 1 is an illustration of a prediction model according to the present invention;

FIG. 2 shows an example of a prediction model according to the present invention; FIG.

3 shows another example of a prediction model according to the present invention

Invention;

4 is a flowchart of a method according to the present invention.

In the following, the prediction model for pedestrians will first be described. It then describes how parameters for the exact design of the prediction model can be determined with the aid of findings from accident research. Finally, it describes how a determination of a

Evidence trajectory based on the described model can be performed.

The future position of a pedestrian is determined by probability distributions Pi (x, y) for several equidistant times t, up to a maximum

Prediction horizon t _n shown. These distributions are a weighted sum of normal distributions with standard deviations Ok and mean values k

V ^ x. y) = ELiW _fc , / v (x, y, a _fc ^ _fc ).

It applies

Σ ₌ Λ = 1 ·

The three components Wi, W2, Wa of the distribution represent three possible and exemplary states of movement of the pedestrian:

• stopping;

• continue at constant speed; • Accelerate.

It is conceivable and possible to include further components in the prediction model

which represented further states of motion of the pedestrian; z. B. going backwards.

The standard deviations σκ, on the other hand, result from the prediction uncertainty for the respective state of motion resulting from the uncertainty of the

Position measurement and velocity measurement propagated over the prediction period result. Distributions are also assumed for the states in which the pedestrian accelerates or stops.

Possible distributions may be, for example, Tiemann, N. et al: "Predictive Pedestrian

Protection - Situation Analysis with a Pedestrian Motion Model ", AVEC 2010

be removed.

Fig. 1 shows the illustration of a prediction model according to the present invention.

In the situation illustrated in FIG. 1, there is an increased probability of collision of the vehicle 1 with the road user 2 at heightened risk.

The illustrated prediction model is based on the three previously introduced

Dimension Stuck Wi, Continue W2, Accelerate W3.

Each dimension Wi, W2, W3 is assigned a weight which reflects the probability with which the collision object will be in the respective movement state or will change into it.

The parameters can be collected based on accident data from near misses or real accidents. In addition to the predicted distribution functions for each state of motion is an essential part of the invention, the choice of weights Wk, which in turn has a significant impact on the outcome of

Determination of evasion trajectories. The weights Wk are not fixed, but chosen depending on the situation.

For example, the weights are chosen depending on the predicted impact point y _C0 n. The function rule ^w k = ^k = CoII fy ^ ² ^'3 with the predicted collision point y _co n can empirically for example. Be determined based on the analysis of real accident.

As particularly relevant for the future probability of residence Pi (x, y) of the pedestrian, the assignment of the impact point y _C0 n has resulted in a collision area. In this case, the front area of the vehicle can, for example, be divided into three areas. Center Area, ie the central area of the vehicle front, as well as the Near Side and the Far Side. The near side is the side area of the

Vehicle front, which is closer to the pedestrian. That is, when the pedestrian approaches the vehicle from the right, the area to the right of the center area. The far side is corresponding to the side area of the vehicle front, which is located farther away from the pedestrian.

It has been shown that it is advantageous to create the weights Wk depending on empirical observations of the behavior of increased vulnerable road users (VRU), for example pedestrians, in real accidents.

In addition to the collision point y _co n, other influencing variables z can also be used, so that

Wk = f ^~ {y _colv z ₁ , z ₂ , ...), k = 1,2,3 holds.

Examples of further influencing variables z are:

• Head orientation of the pedestrian

• Width of the roadway

• oncoming traffic • Approach angle to the road

• Crosswalk

• Traffic light system

• Curve radius

· Light and visibility

• Driving on inner-city or extra-urban roads

• preceding traffic (traffic density)

Velocity of the EGO Vehicle The determination of the avoidance trajectories or collision trajectory is carried out according to an embodiment of the present invention by means of a multi-objective optimization. Typical optimization goals are included

• minimizing the risk of collision with the pedestrian;

• Minimization of the total evasion width;

· Minimization of lateral acceleration;

• in case of an inevitable collision, minimizing collision consequences.

In addition to the optimization of the above goals, constraints must be met. A typical selection constraints are:

· The collision risk remains smaller than a predetermined limit (maximum value);

• stay in a driving corridor;

• lateral acceleration remains below a predetermined limit;

• Maximum lateral offset is less than a predetermined limit

• In case of an inevitable collision, the collision sequences remain

under a predetermined limit.

An important aspect of the present invention is the minimization of the

Kollisionsrisikos or the restriction of the same to a maximum value. In the present case, a collision risk is understood to mean a value to which

collision-relevant factors, such as sensor values, are mapped. For this value, a limit can be predetermined which, when using the

Avoidance trajectory on the vehicle may not be exceeded. The risk of collision with the pedestrian to be evaded can be determined according to the present invention depending on the passing point (XPP, VPP).

With the pedestrian model described above results from a given Passierpunkt a collision risk by means of the integration of

Probability density p (x, y) over the corresponding

Stay area of the vehicle.

The travel corridor can be determined in such a way that it is guaranteed that no (other) collision with static and dynamic obstacles, e.g. Oncoming traffic, takes place,

The predetermined limit for the lateral acceleration can be physically motivated. The limit values for stable guidance of the vehicle must be taken into account, e.g. Stiction, loads.

The predetermined limit for the transverse offset may, for example, be based on the typical or current track width.

An efficient n-order polynomial has proved to be an efficient way to determine the avoidance trajectory. In particular, a polynomial of the 5th order.

If one uses a polynomial of the 5th order for the avoidance trajectory or collision trajectory, one obtains

y (x (t)) = ₀ + a _t x + a ₂ x ² + ... + ₅ x ⁵ where x (t) = v ^■ t where v is the instantaneous vehicle speed. This is how it works

Optimization problem can be solved iteratively. The result is the coefficients a, of the polynomial. FIGS. 2 and 3 show by way of example two different situations

different values for the weights Wi, W2, W3 and as a result different evasion trajectories.

In the first case (Figure 2), it is much more likely that the pedestrian 2 stops (Wi) or continues to run at the same speed (W2) as that the pedestrian 2 accelerates (W3). With an appropriately selected barrier for the

The probability of collision can thereby be implicitly ignored if the pedestrian accelerates (W3).

In the first case (FIG. 2), therefore, an evasion trajectory 20 is selected which provides the vehicle with a lateral offset to the left. In the second case (FIG. 3), the situation is reversed. Here the case can be neglected that the pedestrian stops (Wi), as this is unlikely in relation to the other two dimensions.

In the first case (FIG. 3), an evasion trajectory 20 'is selected as the evasion trajectory 20', which provides a lateral offset to the right for the vehicle.

The corresponding probabilities can be based on empirical data and, accordingly, applied as parameters and updated as needed or in the presence of new findings or more recent data.

The table in Fig. 4 shows a flow chart of a method according to the present invention.

In step 401, a collision object 2, in particular an increased risk

Road users (VRU), such as a pedestrian, on the sensors of the

Vehicle 1 or the like recognized. After the detection of a collision object 2, it is necessary to determine an optimal counteraction which, if possible, should be used to prevent the collision while minimizing the risk of further traffic, including the driver and the occupants of the vehicle 1. For this purpose, first in step 402 a probability of residence of

Collision object 2 determines. If the determination of the probability of residence is based on the prediction model according to the present invention, a particularly accurate determination of the probability of residence takes place

Consideration of the most relevant factors for the determination of the same.

In step 403, based on the determined probability of residence of the collision object, an optimal avoidance trajectory 20, 20 'can be determined or determined.

By means of the determined avoidance trajectory, the corresponding longitudinal and lateral acceleration means of the vehicle 1 are controlled in step 404 such that the vehicle 1 substantially follows the evasion trajectory.

Claims

claims

A method (400) for operating a vehicle (1), comprising the steps of:

• detecting (401) a collision object (2), in particular an increased vulnerable road user, in particular a pedestrian;

Determining (402) a residence probability of the collision object (2), in particular dependent on a prediction model for the collision object (2);

Determining (403) an evasion trajectory (20, 20 ') for avoiding a collision with the collision object (2) or a collision trajectory as a function of the determined probability of residence;

• Controlling (404) of the vehicle (1) in such a way that the vehicle (1) is guided at least partly automatically along the determined avoidance trajectory (20, 20 ') or collision trajectory.

The method (400) of claim 1, wherein in the prediction model the

Movement history of the collision object (2) is taken into account.

3. Method (400) according to claim 1 or 2, wherein in the prediction model the predicted impact point of the collision object (2) on the vehicle (1) is taken into account.

The method (400) of claim 3, wherein in the prediction model the

Collision area of the vehicle (1), in which the predicted impact point of the collision object (2) on the vehicle (1) is taken into account.

5. Method (400) according to claim 1, wherein in the step of determining the evasion trajectory (20, 20 ') or the collision trajectory is determined by means of a method for multi-objective optimization, in particular with the goals minimizing the collision probability with the

Collision object and / or minimization of the total evasion width and / or minimization of the lateral acceleration and / or minimization of the expected collision sequences.

The method (400) of claim 5, wherein the method of multi-objective optimization further includes at least one of the following constraints:

• The probability of collision remains below a predetermined collision risk limit.

• stay in a predetermined driving corridor;

• Finding the lateral acceleration below a predetermined

Limit value for the lateral acceleration;

7. The method (400) according to one of the preceding claims, wherein the

Ausweichtrajektorie (20, 20 ') or collision trajectory is represented as an n-th order polynomial, in particular 5-th order.

8. The method (400) according to any one of claims 5 to 7, wherein the

Collision probability is determined depending on a passing point.

9. The method (400) according to any one of the preceding claims, wherein the

Predictive model is updatable.

Apparatus for setting up a method (400) according to any one of

previous claims.

11. Computer program that is configured to execute a method (400) according to one of claims 1 to 9. A machine-readable storage medium on which the computer program according to claim 11 is stored.