EP3811351A1 - Adaptation of the trajectory of an ego vehicle to moved extraneous objects - Google Patents

Abstract

A method (100, 200) for predicting the trajectories (2a-4a) of extraneous objects (2-4) in the surroundings (11) of an ego vehicle (1) and for determining a separate future trajectory (la), adapted thereto, for the ego vehicle (1), comprising the steps: • the extraneous objects (2-4) are identified (110); • the close-range target (2b-4b) to which the movement of each of the extraneous objects (2-4) leads and the basic rules (2c-4c) according to which this movement occurs are determined (120); • the close-range target (1b) to which the movement of the ego vehicle (1) leads and the basic rules (lc) according to which this movement occurs are determined (130); • a quality function RI-4 is respectively set up (140) for the ego vehicle (1 and for the extraneous object (2-4); • a quality measure Q1-4 is respectively set up (150) for the ego vehicle (1) and for the extraneous objects (2-4); •the optimum movement strategies π1 -4 of the ego vehicle and of the extraneous objects (2-4) which maximise the quality level Q1-4 are determined(160); • the searched-for trajectories (la-4a) are determined (170) from the optimum movement strategies πι-4. A method (100) for controlling the ego vehicle 1. An associated computer program.

Description

 description

Title:

 Adaptation of the trajectory of an ego vehicle to moving foreign objects

The present invention relates to trajectory planning for the at least partially automated method, in particular in mixed traffic with human-controlled foreign objects.

State of the art

Vehicles that are at least partially automated in traffic will not suddenly replace human-controlled vehicles and will not be isolated from human-controlled traffic on separate routes. Rather, these vehicles will have to move safely in mixed traffic with human-controlled foreign objects, whereby these foreign objects also include pedestrians as weaker road users. In the case of human-controlled foreign objects, there is always an uncertainty as to which movement action these foreign objects will carry out next. A control system for at least partially automated driving is therefore dependent on at least partially tapping the future behavior of foreign objects by observing the previous behavior.

WO 2017/197 170 A1 discloses a control unit for a moving autonomous unit, which can be a robot or a vehicle. The control unit first determines a basic trajectory with which the primary goal of the autonomous unit, such as a destination, is tracked. The basic trajectory is then modified by a security module such that a collision with people or other human-controlled units is avoided. For this purpose, the respective human-controlled

Movements predicted.

Disclosure of the invention

Within the scope of the invention, a method for predicting the trajectories of foreign objects in the environment of an ego vehicle and for determining a future trajectory for the ego vehicle adapted to them was developed.

The ego vehicle is the vehicle whose trajectory is to be acted on in order to avoid a collision with the foreign objects. The foreign objects can in particular be people or vehicles controlled by people, such as conventional motor vehicles or bicycles. However, foreign objects that cannot be controlled or can only be controlled to a limited extent are also considered, such as a vehicle that rolls away after being parked on a slope or a trailer that has torn itself away from its towing vehicle.

The foreign objects are initially identified. A time series of physical observations of the environment can be used for this, such as a sequence of camera images or a sequence of

Events that an event-based sensor outputs. As an alternative or in combination, information can be used that was received via a wireless interface of the vehicle. This information can be transmitted by the external objects themselves, for example via a vehicle-to-vehicle (V2V) interface. The information can, however, also be transmitted by an infrastructure, for example via a vehicle-to-infra structure (V 21) interface.

In this context, identifying means at least recording which foreign objects in the environment of the ego vehicle can be moved independently of one another. In this context, it is also advantageous to record what the foreign objects are in detail, but it is not absolutely necessary. It is determined to which near goal the movement of each of the

Leads foreign objects and according to which basic rules these

Movement is in progress. How this investigation is carried out in detail depends on the information available. For example, it can be extrapolated from the time course of the trajectory that certain short-range targets are more likely than others. The more additional information is used, the more accurate the prediction of the near target becomes. If, for example, it is recognized that a vehicle has set a turn signal as a foreign object, then a turning process is very likely planned. A vehicle as a foreign object can also, for example, announce its current short-range or long-range destination directly via V2V communication.

The basic rules according to which the movement of the foreign objects takes place can in particular include the rules of the road traffic regulations and also depend on the type of the foreign objects. For example

Vehicles use the lane and the right of two lanes.

Pedestrians, on the other hand, are, for example, required to walk on sidewalks and, when crossing paths such as traffic lights or crosswalks, for crossing the

Lane are available to use them.

Furthermore, it is determined to which near target the movement of the ego vehicle leads and according to which basic rules this movement takes place. The basic rules here can in particular include the rules of the road traffic regulations and need not be the same in all situations. For example, the permissible maximum speed is limited separately when the vehicle is pulling a trailer or driving with snow chains. The determination of the basic rules can also include, for example, an analysis of the configuration of the ego vehicle.

A quality function R _{I-4 is} set up both for the ego vehicle and for the foreign objects, which assigns a measure for an overall situation x formed from the current states of the ego vehicle and the foreign objects and a possible next movement action ai- ₄ , how good the action ai- _{4 is} in the current overall situation x for the road user under consideration. The quality function R _I-4 can in particular, for example include a measure of the extent to which the movement action ai- _{4 works} in situation x to achieve the respective short-term goal and to comply with the rules. The numerical indices ranging from 1 to 4 are not to be understood as restrictive with regard to the number of treatable foreign objects, but are merely illustrative in order to be able to explain the method using an example. In general, one can also speak of quality functions R, and next movement actions a.

The term “states” generally encompasses the quantities with which the contribution of the ego vehicle or the foreign objects to the traffic situation can be characterized. The states can in particular be positions or time derivatives thereof, that is to say speeds and

Accelerations.

Both the ego vehicle and for the foreign objects a quality measure QI ₄ is respectively positioned the x of the overall situation and the possible next move action ai ₄ in addition to the value Ri ₄ (x, ai ₄₎ and the expected value E ( P (x ')) of a distribution of the probabilities P (x') of

Assigns state changes x 'with which the other road users react to the next movement action ai- ₄ . For example, the quality measure QI- _{4 can be} a weighted sum of the value Ri- ₄ (x, ai- ₄ ) of the quality function and the expected value E (P (x ')).

Those optimal movement strategies pi- _{4 of} the ego vehicle and the foreign objects that maximize the quality measures QI- _{4 are determined} . The sought trajectories of the ego vehicle and the foreign objects are determined from the optimal movement strategies pi- ₄ .

The concept of the movement strategy generally encompasses any function pi- ₄ which assigns a numerical value ni- ₄ (x, ai- ₄ ) to an overall situation x and a next movement action ai- ₄ . The term is therefore generalized compared to the usual use of language, in which it is associated with deterministic rules. A deterministic rule can, for example, specify that if a certain overall situation x is present, exactly one next Movement action ai- _{4 is to be} carried out by the ego vehicle or is carried out by the foreign objects.

It was recognized that the behavior of the foreign objects in particular does not always follow deterministic rules. If the foreign object is controlled by a human, for example, the control is intelligent, but does not necessarily lead to the movement action that is optimal for the pursuit of the respective near target. This applies even if a human driver basically chooses the correct driving maneuver. For example, turning left from a road on which no route is explicitly marked can scatter around the ideal driving line. The vehicle will also come to a stop at the stop line every time there are a large number of braking in front of a red traffic light, but the time course of the speed may vary. For example, the driver can step on the brake pedal harder and weaker at the beginning and later unconsciously readjust the brake pressure in order to come to a stop at the right place at the end. A deeper reason for this is that the

Driving task as a whole is too complex to be carried out fully consciously. In order to be able to manage multitasking at the required speed, a learning driver must first “automate” certain processes in the subconscious.

Even the correct behavior of a pedestrian is not complete

deterministic. For example, if the pedestrian crosses the road, he will not always do so at right angles to the direction of travel, but with a random deviation from it.

The behavior of foreign objects no longer follows deterministic rules if a controlling person makes the wrong decision. For example, setting the right turn signal is no guarantee that the driver will actually turn right from one priority street to the next street and will forego the right of way compared to another vehicle coming from this street. Rather, it can also happen that the driver continues straight ahead after he realizes that he was wrong and only has to turn right one street later. A human driver can also do not react to an object that is hidden in the blind spot of its mirror. Pedestrians also deliberately ignore the obligation to use secure crossings or the waiting time at red traffic lights.

Furthermore, even the behavior of the ego vehicle is somewhat probabilistic. If, for example, a certain brake pressure is applied to the brake cylinder of the brake system for stopping, this can result

The deceleration of the ego vehicle can vary, for example, depending on the condition of the road and the temperature and water content of the brake fluid.

By generalizing the changes in state of the other road users to a probability distribution P (x ') and by

Movement strategies pi- _{4 of} all road users can also be probabilistic, the reaction of the ego vehicle to the overall situation x can thus be refined so that it is more likely to actually be traffic-friendly and in particular to avoid collisions. In a way, the predictive driving that every human driver has to learn in the driving school is technically simulated so that a system for at least partially automated driving can cope with the driving task at least as well as a human driver.

In a particularly advantageous embodiment, quality measures Q _{1-4 are} chosen whose optima with respect to the movement strategies pi- _{4 are given} by the Bellman optimum. In a way, this is a combination of recursive definition and mutual coupling of the quality measures Q _I-4 .

For example, with an infinite time horizon, the quality measures Q, in the final optimized state Q * the shape

where the p * (- ΐ) are the optimal movement strategies of the others

Are road users whose index is other than i and

V ^* (x ') = softmax Q ^* (x', a '). The expected value E runs through the probabilistic state transitions and the strategies of the other road users whose index is different from i. It is given by

In a further particularly advantageous embodiment, the optimal movement strategies pi- _{4 are determined} on the condition that they are independent of one another with the same history H ¹ :

If a Boltzmann-Gibbs distribution continues as the distribution of the

Probabilities P (x ') of state changes x' is selected, then the road users choose their movement strategy according to the principle of maximum entropy:

p (a ₁ (t) | H ') <x exp (ß ^* (x (i), a (t))} (3).

Equations (1) to (3) form a set of M coupled equations, where M is the number of road users considered. The equations can be summarized as

 Q = T, (Q: "Q, \ i [M] <4).

Herein is T, the right side of equation (1). Equation (4) has exactly one optimal solution Q *, which is available with the following algorithm:

Algorithm 1 - MMCE-I

Enter Ht, {Rj}

1: Initialize: Q, ⁰ and Q-, ⁰

2: s = 0

 3: while convergence do

4: for j e -i do

5: Q _j ^{s + 1} Ti (Qi ^{s + 1} , Qi ^s ) 6: end for

 9: end while

Edition Q ^s

When the time horizon is finite, the problem is slightly different. The quality function Q of the i-th road user has the form in the fully optimized state at the time step te [t, t + T]

with the constraint that

is the value of the quality function R, in the final optimized state at the end of the time horizon. Analogous to the case of the finite time horizon, the expected value again depends on the strategies of the other road users, which in turn are advantageously Boltzmann-distributed:

 Therefore equation (5) can be written as:

 W = u {r, nG '\ ί ^ [M \ («where U is the right side of equation (5). For example, an optimal solution can be obtained with the following algorithm:

Algorithm 2 - MMCE-F

Enter {Rj, R _{j F} }, T

1: for j = l, ..., M do

2: V _j ^{t + T} <- R _{j, F}

 3: end for

4: Initialize: Q ⁰ <- [QI ⁰ , ..., QM °]

 5: for k = T-1, ..., 0 do

6: s = 0

7: while convergence do 8: for i = 1, M do

9: Qi ^{s + 1} <- Ui (Qi ^s , Vi ^{t + K + 1} )

 10: end for

 11: s <- s + l

 12: end while

13: QVQ ^S

 14: for j = 1, M do

15: Vjt + k <- softmax _aj U _j (Q- _j °, V _j ^{t + K + 1} )

 16: end for

 17: end for

Edition {V ¹ , V ^{t + T} }

Within the scope of the invention, a further method for predicting the trajectories of foreign objects in the environment of an ego vehicle, and for determining a future trajectory for the ego vehicle adapted to it, was developed. This process initially begins like the previously described process, i.e. the foreign objects are identified and the short-range targets and the basic rules of movement are determined for both the ego vehicle and for the foreign objects.

In contrast to the method described above, a feature function F _{I-4 is} set up for the ego vehicle as well as for the foreign objects such that the application of F _I-4 to a set of qi- ₄ still free parameters is a quality function R _{I -4} supplies, said quality function R x _I-4 an overall situation formed from the current states of the ego vehicle and the foreign objects and a possible next move action ai ₄ assigns a measure of how well the action ai ₄ in the current

Overall situation x for the road user under consideration. The quality function R _I-4 can in particular include, for example, a measure of the extent to which the movement action ai- ₄ in situation x works towards the achievement of the respective short-term goal and compliance with the rules.

The feature function F _I-4 can, for example, embody properties and destinations of the respective road user, such as the destination to which a pedestrian is moving, or his walking speed. At a In addition to the destination, the vehicle can, for example, include the requirement that the journey should be safe, smooth and comfortable in the feature function F _I-4 . The feature function F _1-4 can therefore in particular be composed, for example, of several parts which relate to different goals, wherein these goals can also be opposite. The set qi- ₄ of parameters can then embody, for example, the weights with which different goals and requirements are contained in the final quality function R _I-4 . The set qi- ₄ of parameters can in particular be present, for example, as a vector of parameters and contain, for example, coefficients with which a linear combination of different targets contained in the feature function F _I-4 enters the quality function R _I-4 .

The movement strategies pi- _{4 of} the ego vehicle and the foreign objects are determined as those strategies which lead to a maximum causal entropy H (ai- ₄ || x) of the movement actions ai- _{4 of} the ego vehicle and the foreign objects in the overall situation x , The trajectories sought are determined from the movement strategies pi- ₄ .

The final result obtained has the same advantages as the result obtained according to the previously described method. The advantage of this method in particular is that even less information about the respective road users is required for the determination of the parameter set qi- ₄ than for the direct determination of the quality function R _I-4 . Any additional information, regardless of the source, can be considered on the other side in the parameter set qi- ₄ . The free parameters qi ₄ are pi- in the optimization in response to movement strategies ₄ determined.

The causal entropy H (ai _-4 || x) can be written as

The maximum of the causal entropy H (ai _-4 || x) with respect to the movement strategies pi- ₄ is advantageously determined under the boundary condition that both the ego vehicle and the foreign objects have the expected value of the respective feature function F _I-4 over all possible overall situations x and all possible next movement actions ai- _{4 are} equal to the mean value of the feature functions F1-4 observed empirically in the previous trajectories. This mean can be empirical, especially across all

observed situations x and movements ai- _{4 are} formed:

 In conjunction with the other boundary conditions that the optimal

Motion strategies pi- ₄ with the same history H ^{l are} independent of one another and that they are each statistically distributed around a strategy that maximizes the respective quality function RI- ₄ can be used using

a recursive solution to equation (7) can be given:

Here is

 The boundary condition is Z, (T +1) = 1 for all road users.

The recursive solution is similar to the fully optimized quality measure Q according to equation (5). Wi ^T (H ^T , ai (x)) plays the role of the quality measure Q ,, and the

Quality functions R are composed of the characteristic functions F as a linear combination. Ultimately, an “inverse reinforcement learning” can therefore be carried out from the perspective of the ego vehicle, ie, if the quality function Ri of the ego vehicle is known, only by observing the others

Road users whose quality functions R _{2-4 are} developed. This can be done using the following algorithm, for example:

Algorithm 3: MMCE-IRL for the ego vehicle

 9: end for

In a further particularly advantageous embodiment, the

Foreign objects are each classified in terms of their type, and the respective quality function R2-4 or the respective feature function F2-4 is selected on the basis of this type. In this way, the determination of the final trajectories of the foreign objects, and thus also the ones adapted to them

Trajectory of the ego vehicle, converge faster and arrive at a more precise result. In particular, as previously explained, the basic rules of motion may depend on the type of object. The classification can be made on the basis of the physical observations and / or on the basis of the information received via the wireless interface.

As previously explained, the determination of the trajectory of the ego vehicle, which is adapted to the presence of moving foreign objects, is not an end in itself, but aims to improve the suitability of at least partially automated vehicles, especially for mixed traffic with human-controlled foreign objects. The invention therefore relates also to a method for controlling an ego vehicle in a

Traffic situation with moving foreign objects in the environment of the ego vehicle.

In this method, the trajectory of the ego vehicle, which is adapted to the behavior of the foreign objects, is determined using one of the methods described above. The adapted trajectory is transmitted to a movement planner of the ego vehicle. A control program for a drive system, a steering system and / or a brake system of the ego vehicle is determined by the motion planner, the control program being designed to bring the actual behavior of the vehicle within the system limits as well as possible into agreement with the determined trajectory.

The drive system, steering system and / or braking system is controlled in accordance with the control program.

The method can be implemented in any existing control device of the ego vehicle, since, thanks to the internal networking via CAN bus, access to those recorded with a sensor system or obtained via the wireless interface is typically possible from anywhere in the vehicle

There is information about foreign objects in the vehicle environment. The motion planner can also be controlled from anywhere in the vehicle via the CAN bus. The method can be implemented, for example, in the form of software that can be sold as an update or upgrade for such a control device and, in this respect, represents a separate product. Therefore, the invention also relates to a computer program with machine-readable

Instructions that when on a computer, and / or on a

Control device to be executed, cause the computer and / or the control device to carry out a method provided by the invention. The invention also relates to a machine-readable

Data medium or a download product with the computer program.

Further measures improving the invention are shown below together with the description of the preferred exemplary embodiments of the invention with reference to figures. Exemplary embodiments It shows:

Figure 1 embodiment of the method 100;

FIG. 2 embodiment of the method 200;

FIG. 3 embodiment of the method 300;

Figure 4 Exemplary traffic scene with ego vehicle 1 and three

human-controlled foreign objects 2-4.

FIG. 1 shows an exemplary embodiment of the method 100. In step 110, a time series 11a-11c of physical observations of the surroundings 11 of the ego vehicle 1 (not shown in FIG. 1) is processed together with information 12a that was received via the wireless interface 12. This information 12a comes from the foreign objects 2-4 in the vehicle environment 11 itself, and / or from an infrastructure 5. In step 110, the foreign objects 2-4 are identified, i.e. it is found that there are three foreign objects 2-4 that move in different ways.

Foreign objects 2-4 are classified in step 115 according to types 2d-4d. In step 120, the short-range targets 2b-4b aimed at by the foreign objects 2-4 are predicted, and the basic rules 2c-4c are determined according to which the movement of the foreign objects 2-4 takes place. Analogously to this, it is determined in step 130 to which short-range target 1b the movement of the ego vehicle 1 leads and according to which basic rules 1c this movement takes place.

In step 140, the respective quality function R _{I-4 is} set up for the ego vehicle 1 and for the foreign objects 2-4 on the basis of the available information, with the respective type 2d-4d according to the optional substep 141 of the foreign object 2-4 can be used if this was determined in the optional step 115.

In step 150, the quality functions R _{I-4 are expanded} to quality measures Q _I-4 , which also include the expected value E (P (x ')) of a distribution of the

Contains probabilities P (x ') of state changes x' and insofar also the quality measures Q _{I-4 are} coupled to one another. Quality steps Q _{I-4 are} selected in accordance with substep 151, the optima of which for the movement strategies pi- _{4 are given} by the Bellman optimum. According to sub-step 152, a Boltzmann-Gibbs distribution is selected as the distribution of the probabilities P (x ') of changes in state x'.

In step 160, those movement strategies pi- _{4 of} the ego vehicle and of the foreign objects 2-4 are determined which maximize the quality measures Q _I-4 . Finally, in step 170, the sought trajectories 2a-4a become

Foreign objects 2-4 and the target trajectory la of the ego vehicle 1 adapted to them are determined.

FIG. 2 shows an exemplary embodiment of method 200. Steps 210, 215, 220 and 230 are identical to steps 110, 115, 120 and 130 of the

Procedure 100.

In step 240 of the method 200, in contrast to step 140 of the method 100, a complete quality function R _{I-4 is not} determined, but instead

Feature functions F _I-4 , which are parameterized with a set of qi- ₄ still free parameters and only form the complete quality function R _I-4 in connection with these parameters qi- ₄ . If the types 2d-4d of the foreign objects 2-4 were determined in step 215, these can be used in the optional sub-step 241 to select the respective feature function F _2-4 .

In step 250, the movement strategies pi- _{4 of} the ego vehicle and the foreign objects are determined as those strategies that maximize the maximum causal entropy. At the same time, the parameters qi- _{4 of} the feature functions F _{I-4 are also} determined. According to sub-step 251, a Boundary condition specified that a recursive determination of the

Movement strategies pi- ₄ enables.

In step 260, analogously to step 170 of the method 100, the sought trajectories 2a-4a of the foreign objects 2-4 and the target trajectory la of the ego vehicle 1 adapted to them are determined from the movement strategies pi- ₄ .

FIG. 3 shows an exemplary embodiment of the method 300. In step 310, the target trajectory 1 a for the ego vehicle 1, which is adapted to the behavior of the foreign objects 2-4 in the environment 11 of the ego vehicle 1, is determined using the method 100 or 200. This adapted trajectory la is transmitted to the movement planner 13 of the ego vehicle 1 in step 320. In step 330, a control program 13a for a drive system 24, a steering system 15 and / or a brake system 16 of the ego vehicle 1 is determined by the movement planner 13.

In this context, it is important that the term trajectory generally refers to a path in combined space and time coordinates. This means that a trajectory is not just a change in the

Direction of movement can be changed, but also by changing the speed, such as braking, waiting and starting again later.

In step 340, the drive system 14, the steering system 15, or the

Brake system 16, driven according to the control program 13a.

FIG. 4 shows a complex traffic scene in which the described methods 100, 200, 300 can be used advantageously. On the right

The lane of a road 50 drives the ego vehicle 1 straight in the direction of the near destination lb.

The first foreign object 2 is a further vehicle, the blinker 2e of which indicates that its driver intends to turn into the side street 51 leading to the near target 2b of the vehicle 2. The second foreign object 3 is another vehicle which, from the perspective of the ego vehicle 1, is traveling straight ahead on the opposite lane of the road 50 in the direction of its near destination 3b. The third foreign object 4 is a pedestrian who is a short-range target 4b from his point of view

on the opposite side of road 50.

In the situation shown in FIG. 4, the pedestrian 4 must use the crossing 52 over the road 50, which at the same time causes the driver of the vehicle 3 to

Waiting obliges. In principle, therefore, the driver of vehicle 2 can accelerate directly and, as intended, turn left, which would be optimal for his rapid achievement of short-range target 2b. Accordingly, the ego vehicle 1 would have free travel in its lane at least up to the crossing 52. A control method under the simplifying assumption that the driver of the

Vehicle 2 that will do the best for him would accelerate ego vehicle 1. However, if the driver of vehicle 2 misjudges the situation in that he first has to let vehicle 3 pass in oncoming traffic (which would also be correct without pedestrian 4 on crossing 52), the ego vehicle drives onto the vehicle from behind 2 on. The

Methods according to the invention enable such uncertainties to be taken into account. For example, the speed for the onward journey can be limited to such an extent that in the event that the vehicle 2 actually stops, a collision can still be prevented with full braking.

Claims

Expectations 1. Method (100) for predicting the trajectories (2a-4a) of Foreign objects (2-4) in the environment (11) of a ego vehicle (1), as well as to determine an adapted future trajectory (la) for the ego vehicle (1), with the steps:  • from a time series (lla-llc) of physical observations of the environment (11), and / or from the foreign objects (2-4) themselves and / or from an infrastructure (5) via a wireless interface (12) of the vehicle (1) received information (12a), the foreign objects (2-4) are identified (110);  • It is determined (120) which short-range target (2b-4b) the movement of each of the foreign objects (2-4) leads to and which basic rules (2c-4c) this movement takes place;  • it is determined (130) to which short-range target (lb) the movement of the ego vehicle (1) leads and according to which basic rules (lc) this movement takes place; • For the ego vehicle (1) and for the foreign objects (2-4), a quality function R _{I-4 is} set up (140), which is one of the current states of the ego vehicle (1) and the foreign objects (2- x 4) formed situation and a possible next move action ai ₄ assigns a measure of how well the action ai ₄ in the current  Overall situation x for the road user under consideration (1-4); • A quality measure Q _{I-4 is} set up (150) for the ego vehicle (1) as well as for the foreign objects (2-4), which in addition to the value of the overall situation x and the possible next movement action ai- ₄ Ri- ₄ (x, ai- ₄ ) also assigns the expected value E (P (x ')) to a distribution of the probabilities P (x') of state changes x ' which the other road users (2-4; 1, 3, 4; 1-3) react to the next movement action ai- ₄ ; • those optimal movement strategies pi- _{4 of} the ego vehicle and the foreign objects (2-4) are determined (160) that maximize the quality measures Q _I-4 ; • The optimal movement strategies pi- ₄ are used  Trajectories (la-4a) determined (170). 2. The method (100) according to claim 1, wherein quality measures Q _{I-4 are} selected (151) whose optima with respect to the movement strategies pi- _{4 are given} by the Bellman optimum. 3. The method (100) according to any one of claims 1 to 2, wherein one Boltzmann-Gibbs distribution is chosen as the distribution of the probabilities P (x ') of state changes x' (152). 4. Method (200) for predicting the trajectories (2a-4a) of Foreign objects (2-4) in the environment (11) of an ego vehicle (1), as well as for determining an adapted future trajectory (la) for the ego vehicle (1), with the steps:  • from a time series (lla-llc) of physical observations of the environment (11), and / or from the foreign objects (2-4) themselves and / or from an infrastructure (5) via a wireless interface (12) of the vehicle (1) received information (12a), the foreign objects (2-4) are identified (210);  • It is determined (220) which short-range target (2b-4b) the movement of each of the foreign objects (2-4) leads to and which basic rules (2c-4c) this movement takes place;  • it is determined (230) to which short-range target (lb) the movement of the ego vehicle (1) leads and according to which basic rules (lc) this movement takes place; • For the ego vehicle (1) and for the foreign objects (2-4), a feature function F _{I-4 is} set up (240) in such a way that the Applying F _I-4 to a set of qi- ₄ still free parameters yields a quality function R _I-4 , this quality function R _{I-4 being} one of the current states of the ego vehicle (1) and the foreign objects (2-4) formed overall situation x and a possible next one Move action ai ₄ assigns a measure of how well the action ai ₄ in the current overall situation for the x under consideration  Is a road user (1-4); • the movement strategies pi- _{4 of} the ego vehicle (1) and the Foreign objects (2-4) are determined as those strategies (250) that lead to a maximum causal entropy H (ai- ₄ || x) of the movement actions ai- _{4 of} the ego vehicle (1) and the foreign objects (2-4) in the  Overall situation x lead; • The trajectories sought (la-4a) are determined from the movement strategies pi- ₄ (260). 5. The method (200) according to claim 4, wherein the maximum of the causal entropy H (ai- ₄ || x) with respect to the movement strategies pi- ₄ below the Boundary condition is determined (251) that both for the ego vehicle (1) and for the foreign objects (2-4) the expected value of the respective Feature function F _I-4 over all possible overall situations x and all possible next movement actions ai- _{4 is} equal to the mean value of the feature functions F _I-4 observed empirically in the previous trajectories. 6. The method (100, 200) according to any one of claims 1 to 5, wherein the optimal movement strategies pi- ₄ are determined under the boundary condition (160, 250) that they are independent of one another with the same history H ¹ . 7. The method (100, 200) according to any one of claims 1 to 6, wherein the optimal movement strategies pi- ₄ are determined under the boundary condition (160, 250) that they are each statistically distributed around a strategy that the respective quality function R _{I -4} maximized. 8. The method (100, 200) according to one of claims 1 to 7, wherein the foreign objects (2-4) are each classified in terms of their type (2d-4d) (115, 215) and wherein the respective quality function R _2-4 , or the respective Feature function F _2-4 , based on this type (2d-4d) is selected (141, 241). 9. Method (300) for controlling an ego vehicle (1) in a traffic situation with moving foreign objects (2-4) in the vicinity of the ego vehicle (1) with the steps:  • the trajectory (la) of the ego vehicle (1) adapted to the behavior of the foreign objects (2-4) is determined (310) using a method (100, 200) according to one of claims 1 to 8;  • the adapted trajectory (la) is transmitted (320) to a movement planner (13) of the ego vehicle (1);  • The motion planner (13) determines (330) a control program (13a) for a drive system (14), a steering system (15) and / or a brake system (16) of the ego vehicle (1), the control program (13a ) is designed to bring the actual behavior of the vehicle (1) as close as possible to the trajectory (la) within the system limits;  • The drive system (14), steering system (15) and / or brake system (16) is controlled (340) in accordance with the control program (13a). 10. A computer program containing machine-readable instructions which, when executed on a computer and / or on a control device, cause the computer and / or the control device to do so Execute method (100, 200, 300) according to one of claims 1 to 9.