Methods for controlling a vehicle in a traffic situation, assistance systems and vehicles

Abstract

The present invention relates to a method for controlling a vehicle in a traffic situation, wherein the vehicle comprises a control unit for controlling the vehicle, at least one environment sensor, and a prediction planner which can predict future trajectories for at least one detected object, wherein several trajectories are predicted for the detected object, a probability is determined for each predicted trajectory as to whether the object will follow this trajectory or not, and the vehicle is controlled on the basis of a trajectory plan which takes into account at least one predicted trajectory of the object, and wherein several planning options are determined for the trajectory planning of the vehicle, wherein the planning options differ in that only trajectories selected on the basis of their probability are taken into account, and an evaluation of the planning option is carried out.whether a control intervention is necessary at the current time, whereby a planning option for controlling the vehicle is selected from the determined planning options based on the evaluation, by selecting a planning option that performs such an intervention if a control intervention is necessary at the current time, and selecting a planning option that changes the current control of the vehicle as little as possible if no control intervention is necessary at the current time.

Description

The present invention relates to a method for operating a vehicle, to an assistance system for a vehicle which is operated by the assistance system or AD system, and to a (partially) autonomous vehicle which comprises an assistance system or AD system according to the invention.Technological BackgroundVehicles of the generic type, such as passenger vehicles (passenger cars), trucks (trucks) or motor cycles, are increasingly equipped with driver assistance systems which, with the aid of sensor systems, can detect the environment, detect traffic situation and assist the driver, for example by a braking or steering intervention or by outputting an optical or acoustic warning. As sensor systems for detecting the environment, radar sensors, lidar sensors, camera sensors or the like are usually used. From the sensor data determined by the sensors, conclusions about the environment can subsequently be drawn.The term driving function is used in particular to subsume the assisted and (partially) automated, driver-assisted vehicle behavior in driver assistance systems. In this case, the processed sensor information is used for environment recognition in order to provide instructions for driver warning / information or for controlled steering, braking and acceleration on the basis thereof. As a result, generic driving functions can help to avoid accidents with other road users or to facilitate complicated driving maneuvers by supporting the driving task or the vehicle guidance or even performing it completely. For example, the vehicle may be maintained in the lane by an emergency brake assist (EBA), automatic emergency brake assist (AEB), emergency steer assist (ESA), automatic emergency emergency assist (AES), or active lane keeping assist with steering assist (LKA). In addition to emergency braking and emergency steering in dangerous situations, the automated braking and / or steering intervention is of great importance, in particular when a vehicle is being guided in a (fully) automated manner. Automated braking or steering is initiated in a critical environmental situation or traffic scene, in particular in the event of an imminent collision.Furthermore, in particular in fully or partially automated driving systems (AD systems; automated driving system), it is necessary to obtain knowledge about the environment in order to reliably include roads, lanes, obstacles, road users and the like in a trajectory planning of the vehicle. However, since the road users change their position and behavior with respect to the ego vehicle, it is important to predict this behavior or predict how other road users will behave in order to allow planning of their own behavior. The trajectories predicted for the road users are generally based on probabilities and are then assumed to be given during planning, which can result in conservative and, in some situations, also dangerous behavior. A further problem here is that the future behavior of the road users often cannot be predicted unambiguously. For this reason, modern prediction methods for road users generate so-called multiple prediction hypotheses, wherein a probability value is calculated for each hypothesis and all probability values sum up to 100%. This approach is also referred to as "multimodal probabilistic prediction". The predictions are usually provided in the form of trajectories which define the position of the object at a specific time for a few time steps (for example for time steps every 100 ms, over an interval of 0 to 8 seconds).With regard to the above-described prediction methods for road users, however, unclear traffic situations may occur in which it cannot be clearly predicted how, for example, the position of other road users and their behavior relative to the ego vehicle changes. Such unclear traffic situations may, however, become clear traffic situations or may become clear over time, for example, if the ego vehicle reduces the distance to relevant road users, such as a further vehicle, and more information about the respective traffic situation is thereby available. Often, the probability begins at a low level and increases with time, or it begins at a medium or high level and reduces as the ego vehicle approaches another road user, for example. In such traffic scenarios, a human driver would drive "more careful", reduce speed somewhat, and be ready to brake. However, this "caution" or conservative driving mode has some limits, since the human driver would also avoid any unnecessary disruption to the traffic flow. Another reason not to drive too conservatively is that the other road user could incorrectly interpret this, for example with the assumption that the ego vehicle wishes to do without his forward drive. There is thus an increased interest in resolving such traffic situations and controlling the (partially) autonomous ego vehicle in such a way that it mimics the human behavior.Background ArtWO 2016 119 952 A1 discloses a method for operating a control device of an automatically driving motor vehicle, wherein the location of the motor vehicle is determined, environmental data of the motor vehicle are recorded, and a control characteristic of the control device of the motor vehicle is designed such that a driving behavior of at least one other road user is influenced in a defined manner. This is intended to support a liquid driving style of the motor vehicle within a traffic situation, so that the acceptance of the occupants of the motor vehicle and also the acceptance of other road users is positively influenced.Furthermore, DE 10 2019 003 557 A1 describes a method for operating a vehicle, in which at least one control parameter for controlling an automated driving operation is adapted to a determined level of trust of a driver of the vehicle, wherein the level of trust is determined on the basis of at least one physiological parameter of the driver detected during the automated driving operation. For this purpose, cortical brain activity and / or an electrical skin conductance resistance of the driver are detected as physiological parameters. This allows an objective determination of a reaction of the driver during the human automated driving operation, so that the automated driving operation is adapted to the current driver's level of trust in order to increase the acceptance of the driver assistance system for the driver.EP 2 681 085 B1 discloses a method for predicting and adapting movement trajectories of a motor vehicle for assisting the driver in his driving task and / or for preventing a collision or mitigation of accident consequences, in which an intersection set of movement trajectories required by situation, which are determined with the aid of a surroundings sensor system, and of physically possible movement trajectories which result from the driving dynamic properties of the motor vehicle and the coefficient of friction which is established between the tire and the roadway up to a maximum possible limit coefficient of friction, is thereby formed for predicting and adapting the movement trajectories of the motor vehicle, wherein only movement trajectories which are located within the intersection set are taken into account.Furthermore, DE 10 2020 200 183 A1 describes a method for creating a probabilistic free space map with static and dynamic objects, in which static objects of a perception range polygon are retrieved from an existing environment model, predicted trajectories of dynamic objects are collected, the static objects of the perception range polygon and the predicted trajectories are merged in a first free space map and a maximum prediction time, a current prediction time, prediction time steps, confidence ranges (around the static and dynamic objects) and at least one uncertain range around at least one static or dynamic object are defined. A first probabilistic free space map for the current prediction time and at least one further free space map for at least one prediction time step are then generated, which are subsequently evaluated.Object of the present inventionProceeding from the prior art, the object of the present invention is now to provide a method of the generic type with which a human driving behavior can be simulated, with the result that the acceptance of the occupants is increased without reducing the safety in the respective traffic situation.Solution of the ProblemThe above object is achieved by the entire teaching of claim 1 and the subordinate claims. Practical embodiments of the invention are claimed in the dependent claims.In the method according to the invention for controlling an ego vehicle in a traffic situation, the ego vehicle comprises a control device for controlling the ego vehicle, at least one environment sensor for object detection and environment detection, and a prediction planner which can predict future trajectories for at least one detected object. The prediction planner can be part of the control device (e.g. installed thereon as a software module) in the same way as a trajectory planner. For the detected object, a plurality of possible future trajectories are predicted or predictions about its future movement are made, wherein for each predicted trajectory a probability is determined whether or not the object will follow this trajectory. The ego vehicle is then controlled on the basis of a trajectory planning (which is calculated, for example, by the trajectory planner) which takes into account at least one of the predicted trajectories of the object. A plurality of planning options are then determined for the trajectory planning of the ego vehicle, wherein the planning options differ in that only predicted trajectories are taken into account in each case, which are selected on the basis of their probability. Furthermore, the planning option is evaluated as to whether intervention in the control of the ego vehicle is necessary at the current point in time (e.g. since a collision or an accident is imminent if the ego vehicle continues to travel unchanged and the object follows its predicted trajectory). Advantageously, a planning option for controlling the ego vehicle is then selected from the determined planning options on the basis of the evaluation by selecting a planning option which carries out such an intervention, provided that intervention in the control of the ego vehicle is necessary at the current point in time, and selecting a planning option which changes the current control of the ego vehicle as little as possible or is as comfortable as possible at the current point in time, and thereby increasing the acceptance of the occupants (for example by avoiding unnecessary braking processes which could subsequently be interpreted as incorrect behavior by the occupants).Preferably, the planning options may also differ in that different planning strategies and / or planning parameters are taken into account in each case, insofar as intervention time and ease of intervention are varied, wherein the planning options, which are more likely to use only probable predictions (i.e. are essentially based on predictions that have a high probability value, e.g. >60% or >70% or >80% or >90%), provide an early and comfortable intervention (i.e. no excessively strong steering or braking or a slow speed reduction or avoidance movement that is perceptible to the occupants), and planning options, which are more likely also to use unlikely predictions (i.e. are essentially based on predictions that have a low probability value, For example, <50% or <40% or <30% or <20% or <10%) use a late (i.e. for example only shortly before a hazardous situation, e.g. a collision; determinable, e.g. over a time until the collision or TTC-timo-to-collision; the behavior is similar to that of an emergency brake assist or emergency steering assist) and therefore cause a more uncomfortable intervention.The two opposing behaviors can also be described situationally in such a way that, for example, a vehicle standing by far ahead at a traffic light (i.e., the prediction of this vehicle is very safe) causes a very early slow deceleration. In contrast, another vehicle that is cutting in in an instantaneous manner in front of the ego vehicle and that ignores the priority rules generates a violent braking (or possibly causes an emergency braking.Expediently, as an intervention in the control of the ego vehicle, a speed control (e.g. braking, accelerating or the like) and / or a lateral movement (e.g. a steering operation, swabbing or the like) of the ego vehicle can be provided.Furthermore, the method can be carried out repeatedly for a traffic situation at different times. This allows the planning options to be rechecked from time to time. In a practical manner, this takes account of the fact that traffic situations can resolve over time, for example by observing an object for a longer time, by the object moving away, by the ego vehicle coming closer to the object or by the object and / or the infrastructure giving signs (for example braking, accelerating or other speed changes, steering operations, radio messages, traffic signals or the like), which are suitable for evaluating the traffic situation or the object behavior differently or for providing other predictions.Preferably, a first planning option (option 1), in which only predicted trajectories with a probability greater than a first value p1are taken into account, and a second planning option (option 2), in which only predicted trajectories with a probability greater than a second value p2are taken into account, wherein the value p1is greater than the value p2, are provided.Furthermore, for the evaluation of the planning option, whether or not an intervention in the control of the ego vehicle is necessary at the current point in time, it can be taken into account whether a collision with this object is imminent by a predicted trajectory for an object.According to a particular embodiment of the invention, multiple planning options N can be provided.Each planning option N can be planned expediently at least twice, different values being used for the probability each time.If the resulting trajectories of the multiple plannings already differ at the beginning of the trajectory, the planning option can be evaluated in such a way that intervention in the control of the ego vehicle is necessary at the current point in time, with the result that the planning option which also carries out such intervention is then selected.The present invention furthermore comprises an assistance system or AD system for an ego vehicle, that comprises a control device for controlling the ego vehicle, at least one environment sensor for object detection and environment detection, and a prediction planner which can predict future trajectories for detected objects, wherein the assistance system is configured to control the ego vehicle on the basis of a method according to the invention.The present invention furthermore claims a corresponding ego vehicle or vehicle, which comprises an assistance system or AD system according to the invention.DESCRIPTION OF THE INVENTION WITH REFERENCE TO EMBODIMENTSThe invention is explained in more detail below with reference to suitable exemplary embodiments. The following are shown: FIG. 1 shows a simplified illustration of a traffic situation in which the ego vehicle approaches a further vehicle in an adjacent lane; FIG. 2 shows a simplified illustration of a traffic situation in which the ego vehicle approaches a T intersection and a further vehicle approaches the opposite lane; FIG. 3 shows a simplified illustration of an embodiment of the speed profile of an ego vehicle for various planning options for the traffic situation from FIG. 2 ; FIG. 4 shows a simplified illustration of a further configuration of the speed profile of an ego vehicle for various planning options for the traffic situation from FIG. 2 ; FIG. 5 shows a simplified illustration of a first configuration of the speed profile or braking behavior of the ego vehicle in the traffic situation according to FIG. 2 for option 1 and option 2; FIG. 6 shows a simplified illustration of a second configuration of the speed profile or braking behavior of the ego vehicle in the traffic situation according to FIG. 2 for option 1 and option 2; FIG. 7 shows a simplified illustration of a third configuration of the speed profile or braking behavior of the ego vehicle in the traffic situation according to FIG. 2 for option 1 and option 2; FIG. 8 shows a simplified illustration of a traffic situation in which the ego vehicle approaches a T intersection on a road in front, another vehicle is driving in front on the same lane of the road in front and another vehicle wishes to drive onto the road in front from the right; FIG. 9 shows a simplified illustration of a first configuration of the speed profile or braking behavior of the ego vehicle in the traffic situation according to FIG. 8 for option 1 and option 2; FIG. 10 shows a simplified illustration of a first configuration of the steering behavior of the ego vehicle in the traffic situation according to FIG. 8 for option 1 and option 2; FIG. 11 shows a further simplified illustration of the traffic situation from FIG. 8, wherein the ego vehicle has now reached the T intersection; FIG. 12 shows a simplified illustration of a first configuration of the speed profile or braking behavior of the ego vehicle in the traffic situation according to FIG. 11 for option 1 and option 2; FIG. 13 shows a simplified illustration of a first configuration of the steering behavior of the ego vehicle in the traffic situation according to FIG. 11 for option 1 and option 2; FIG. 14 shows a simplified illustration of a first configuration of the speed profile or braking behavior of the ego vehicle in the traffic situation according to FIG. 11 for option 1 and option 2, wherein option 2 has been calculated with a low probability using (dashed line) and without (dotted line); FIG. 15 shows a simplified illustration of a first configuration of the steering behavior of the ego vehicle in the traffic situation according to FIG. 11 for option 1 and option 2, wherein option 2 has been calculated with a low probability using (dashed line) and without (dotted line), and FIG. 16 shows a simplified schematic illustration of an embodiment of a vehicle with an assistance system according to the invention.FIG. 1 shows a traffic situation in which the ego vehicle 1 approaches a further vehicle 10 on a road, in particular a freeway. The ego vehicle is in the left lane and the further vehicle 10 is in the right lane. In the traffic scenario, ego vehicle 1 has a higher speed than further vehicle 10 or is faster than vehicle 10. In the present traffic situation, there could therefore be two predicted trajectories for the further vehicle 10, one trajectory (solid line) follows, for example, its lane and another trajectory (dashed line) changes the lane to the left to the lane of the ego vehicle 1.A further traffic situation is shown in FIG. 2, in which the ego vehicle 1 approaches a T intersection and has traveled according to traffic rules. Another vehicle 12 now approaches from the opposite direction. Here too, there could be two predicted trajectories for the further vehicle 12, wherein one trajectory (solid line) follows the lane straight ahead and the second trajectory (dashed line) makes the vehicle 12 turn to the left. In both traffic scenarios described, if the predicted trajectories or predictions represented by dashed lines have a very high probability of a meaningful behavior of the ego vehicle 1 reducing or braking the speed in order to avoid a collision with the further vehicle 12. Although the further vehicle 12 in the traffic scenario according to FIG. 2 would violate the traffic regulations when turning, the speed reduction of the ego vehicle 1 would possibly avoid a collision or would at least mitigate its consequences. In contrast, it would be a reasonable behavior for the ego vehicle 1 to continue to drive on without reducing the speed if the probability of the continuous trajectories or predictions is very high or higher.In FIG. 3, various stages of a scheduled conservative driving speed profile are shown, wherein the speed of the vehicle can be reduced from 50 km / h to 0 km / h until standstill, in order to stop in front of an obstacle on the road. Option 1 (solid line) shows a reasonable behavior if the blockage by the obstacle has a high probability. The deceleration begins early and is comfortable for the occupants. Option 2 (dashed line) shows a non-clear traffic situation, as a result of which the delay only begins later. The closer the vehicle comes to the obstacle, the clearer the situation becomes. The delay is thus less comfortable, but only for a short time. Option 3 (dotted line) shows a traffic situation in which blockage by the obstacle is initially unlikely, but then becomes clearer and more likely. The deceleration thus only starts very late and is therefore abrupt and not very comfortable or unpleasant, the unpleasant braking lasting about 40-60 m and thus being longer than in option 2 (50-60 m).In Figure 4, the velocity profile for the three options is shown, in case the probability of blockage by the obstacle is very low and even disappears at some distance from the obstacle. Option 1 (solid line) describes the traffic situation in which the speed has already decreased almost to 30 km / h when the obstacle disappears, and it takes some time for the vehicle to reach 50 km / h again. Option 2 (dotted line) describes the traffic situation where the speed loss is much less when the obstacle disappears and the speed of 50 km / h is reached again much faster. Further, option 3 (dotted line) describes the traffic situation where a disappearing obstacle would have no effects, and the vehicle would easily continue to travel.Options 1- 3 shown in FIGS. 3 and 4 represent different levels of conservative driving style. Option 1 is too conservative in many cases, as the vehicle would most likely deflect others too much. Option 3 represents a lack of predictive driving style and responds very late, which leads to hard braking maneuvers. Option 2 is a compromise of both. Finally, it can be stated that a behavior depending on the prediction probability could contribute to making an ADS drive more man-like. Specifically, conservative driving in high probability predictions and less conservative driving in low probability predictions could be one way to achieve this goal. Modern behavior designers can handle objects differently in terms of safety margins, but optimization of a path or trajectory is often performed by defining all detected objects simultaneously with a set of parameters defining the behavior, e.g., early braking or late braking, the allowable comfort deceleration levels, allowable decelerations, allowable deviations from the set speed, preventing collision or planning collision mitigation, and the like. The presented method describes how the behavior planning can be adapted to the probabilities of the object prediction with the aim of driving more conservatively, like humans, and in unclear situations without again driving too conservative, while reducing the risk of traffic distraction and the risk of the driving task.For this purpose, a plurality of action options are planned, wherein the "best" option is to be selected and executed in each case. For example, multiple options may be scheduled with different sets of object predictions depending on their probability value and with different parameter sets defining the scheduled behavior. However, all planning options should take into account the accelerating, decelerating and steering capabilities of the actuators. They differ in how early they begin to intervene, how close they are allowed to pass by obstacles or whether they avoid a collision, for example with all means, or only operate a collision sequence mitigation. As an example, two planning options may be described: on the one hand, an ACC-like longitudinal control behavior in which early braking and gentle steering are in the foreground, and on the other hand, a behavior similar to emergency braking and steering, the interventions only beginning very shortly before a potential collision.However, it should be noted that, for example, option 1 (solid lines) shown in FIGS. 3 and 4 also perform emergency braking if necessary if, for example, the situation should change critically, while options 2 and 3 (dashed and dotted lines) will never react meaningfully to an ACC object, since only late and heavy intervention is taken. The first planning option (option 1) is to take into account only predictions with very high probability and to plan a very pleasant driving behavior with, for example, early and gentle deceleration. The next planning options (option 2 to option N) are also taking into account increasingly unlikely predictions, each option resulting in less comfortable and delayed reactions, i.e. if there are N planning options, option 1 is carried out only with a subset of the object predictions or the predicted trajectories with the highest probabilities. Specifically, all predictions with a probability greater than a value p1are taken into account. For option 2, all predictions of option 1 plus all predictions with a probability greater than value p2 (where p1>p2) are taken into account. Moreover, this occurs equally for all planning options. The last planning option N finally takes into account all predictions with a probability greater than pN (with pN>=0). In case of pN = 0, the planning option N takes all predictions into account. In the case of pN>0, the predictions with the probabilities p<=pN are completely ignored. All planning options operate with different parameters. Option 1 is configured to allow for early and comfortable deceleration and Option N to brake as late as possible (i.e., also comparatively hard or heavy). Of course, the method functions not only with differently configured decelerations or brakings, but can also take account of or perform lateral or lateral movements, i.e. steering movements. That is, when iSd is referred to as "early braking", "comfortable braking", or "hard braking", the same applies to early, comfortable, or "hard" steering. It is also possible that a planning option provides a comfortable and early braking or deceleration without steering being effected, and a second, late and hard steering or lateral movement without deceleration. As a next step, the best result of the N-planning results must now be found by selecting the best option.In general, a "best" option iSd of the invention is understood to mean an option which, if possible, avoids or at least weakens a collision or an impact on an object and at the same time is as comfortable as possible. One way to select the best option is the so-called "double test method". The idea behind this method is that the different planning options n|n=1...N take more and more predictions into account. Each option n- 1 considers only a subset of the predictions of option n. Thus, a prediction with the probability q may be considered only by a planning option n with a probability threshold pn<q. In this case, the question arises in particular as to whether or not it is necessary at this point in time to choose this option n in order to avoid an accident or impact.For example, referring to FIG. 2, the solid trajectory or prediction may have a probability of 80% and the dashed trajectory of 20%. Thus, two planning approaches or options are present, wherein option 1 takes into account probabilities >50% and option 2 all probabilities, so that the braking operations or decelerations shown in FIGS. 5-7 can be selected at different times for the different options (respectively, option 1 is shown in full and option 2 is shown in dashed lines). In the illustration according to FIG. 5, the ego vehicle still has a certain distance from the intersection at a first point in time, in the illustration according to FIG. 6, the ego vehicle has already arrived closer to the intersection at a second point in time, and in the illustration according to FIG. 7, the ego vehicle has almost reached the intersection at a third point in time. As long as a certain distance from the intersection is present, option 1 is selected, since no immediate intervention is required according to option 2. In the traffic situation according to FIG. 7, however, option 2 must be selected, since an intervention is necessary in order to prevent the collision. Of course, in this traffic situation, the option with the greatest delay could also be selected, wherein this comparatively simple approach does not always function correctly.Surprisingly, in this example of the method, a strong deceleration maneuver (according to option 2) would therefore be carried out on an object prediction which has a low probability. However, the strategy assumes that the predictions will in most cases become more stable and clear as the ego vehicle approaches an object and the object can be observed for a longer time. Normally, the above example would thus either end with a trajectory for an object whose prediction probability increases as the ego vehicle approaches it-then the strong deceleration would also make sense-or the probability decreases to a very small value or the prediction completely disappears and no intervention is taken at all. However, it is also possible here that the probability changes only late, even if the intervention is already active, or it remains at the low level. In both cases, however, the intervention is the correct reaction or option, because at the beginning of the intervention "knows" the scheduler or the assistance system does not know how the probability will change. A great advantage of the present invention is that the scheduler does not react early to the low probability predictions, as this would deflect both passengers and subsequent traffic and reduce acceptance.As described above, traffic rule violations as in the traffic situation according to FIG. 2, in which the further vehicle 12 is driving ahead of the ego vehicle 1, can also be taken into account in the present method algorithm, for example by assuming that the drivers do not normally violate traffic rules, and therefore the probability of this prediction would be reduced. The algorithm could also mark such predictions as "violation of traffic rules". In addition to these unambiguous situations, however, there may also be ambiguous traffic scenarios in which the assistance system enters a dilemma in which it must either brake markedly for a less likely hypothesis (dashed trajectory), which may be a potential dysfunction for the occupants of the (partially) autonomously operated ego vehicle 1-or, if not braked, a collision is risked if the less likely hypothesis or prediction becomes a reality.In FIG. 8, a traffic scenario is shown in which there is very high probability prediction (~100%) for the vehicle 13, which is slower than the ego vehicle 1, to follow the current lane straight ahead. For the vehicle 14, there is again a probability of 10% of driving onto the driving road before the ego vehicle 1 has passed. The first planning option 1 (solid line) calculates a pleasing deceleration to adapt the speed to the vehicle 13. Consequently, no avoidance maneuver is calculated and vehicle 14 is ignored because the prediction probability of 10% is low. Planning option 2 (dashed line) calculates a late and strong deceleration to vehicle 13 and a late and hard avoidance maneuver for vehicle 14.FIG. 11 shows the traffic scenario from FIG. 8, wherein the ego vehicle 1 has already reached the intersection. When the intersection is reached, the prediction probability of vehicle 14 may still be small, but the "point-of-no-return" has already been reached. Thus, if option 2 is not selected and vehicle 14 would enter the intersection, an accident could no longer be avoided. The choice between options based on the deceleration would fall on option 1, which would be dangerous because of the possible collision with vehicle 14. Selection between the options based on steering results in option 2 in this scenario, but this is also not safe in this case. If this scenario would take place in a curve, an avoidance maneuver could require a smaller steering angle than if the ego vehicle 1 followed only the lane. Instead, however, the verification of the need to choose option 2 is in the foreground by performing the planning of option 2 twice, first in a "normal" manner, using the parameters of option 2 and taking all predictions into account, and secondly in a modified manner by ignoring all predictions with low probability, as provided in option 1. The speed distribution of the ego vehicle 1 for the two options 1 and 2 for the traffic scenario in FIG. 11 is shown in FIG. 12 and the steering activity in FIG. 13. As a result, ignoring the predictions, Option 2 would be likely to schedule a late, strong deceleration on vehicle 13, but no avoidance maneuver because vehicle 14 would be ignored. Planning option 2 with and without the predictions or predictions with a lower probability results in different results (as shown in FIGS. 14 and 15 using the dotted line). However, since the decisive or hazardous situation (possible collision) is now closer to the ego vehicle 1, it is necessary to start the intervention immediately and consequently to choose option 2.In the situation of Fig. 8, option 1 is still a possible solution option for the traffic scenario, since there is still enough time left for the start of the avoidance maneuver and it is the aim to choose the most comfortable option that avoids an accident.In general, there could be not only two scheduling options (option 1 and option 2), but N-fold thereof. The proposed algorithm or method sequence could be configured as follows:1.) Plan Option N twiceStep 1.1: Planning option N taking all predictions with p>pN into account;Step 1.2: Planning option N taking all predictions with p>pN-1 into account;If the resulting trajectories of both plans differ, a reaction to the pN predictions must be carried out. However, it may also be that this reaction is immediately necessary or in 1 second or in 5 seconds etc., for which reason step 1.3 is to be carried out:Step 1.3: If the resulting trajectories of both plans already differ at the beginning of the trajectory, immediate intervention is required for the predictions with low probability and option N is selected;End of the selection method.If this is not the case, the method continues with step 2.2.) Plan Option N-1 twiceStep 2.1: Planning option N-1 taking all predictions with p>pN-1 into account;Step 2.2: Planning option N-1 taking all predictions with p>pN-2 into account;Step 2.3: If the resulting trajectories of both plans already differ at the beginning, immediate intervention is necessary for the predictions with p>pN-1, and the option N-1 is selected;End of the selection method.If this is not the case, the method continues with step 3.3.) Plan Option 2 twiceStep 3.1: Planning Option 2 Taking all Predictions with p>p2;Step 3.2: Planning Option 2 taking all predictions with p>p1 into account;Step 3.3: If the resulting trajectories of both plans already differ at the beginning, immediate intervention is necessary for the predictions with p>p2and option 2 is selected; end of the selection method.If this is not the case, the method continues with step 4.4.) Selection of Option 1End of the selection method.Furthermore, plans are conceivable which carry out the trajectory planning separately from the speed planning. Such designers usually ignore all dynamic obstacles during trajectory planning and take into account only the static environment for this purpose. Dynamic obstacles are generally only taken into account in the speed planning. In this case, the decision method can be simplified by selecting the speed planning to provide the strongest delay, so that the double planning or "double test method" is then not necessary.Reference numeral 1 in FIG. 16 denotes a vehicle or ego vehicle which has a control device 2 (ECU, electronic control unit or ADCU, assisted and automated driving control unit), various actuators (steering 3, engine 4, brake 5) and sensors for detecting the environment (radar sensor 6, camera 7, lidar sensor 8 and ultrasonic sensors 9 a- 9 d) or environmental sensors. Further, the vehicle includes a display device 10 (e.g., navigation system, infotainment system, onboard computer, and / or the like) to display information to the driver. Further, the driver may input information to the system via a human-machine interface or confirm information / messages (e.g., confirm suggested speeds). The vehicle 1 can be controlled in a (partially) automated manner in that the control device 2 can access the actuators and the sensors or their sensor data. In the area of assisted or (partially) automated driving, the sensor data can thus be used for environment detection and object detection, so that various assistants or assistance functions, such as speed assistants (ISA, intelligent speed assist), adaptive cruise control (ACC), emergency braking assist (EBA, electronic brake assist), lane keeping control or lane keeping assist (LKA), parking assist, traffic jam assist or the like, can be realized via the control device 2 or the algorithm stored there. Furthermore, the control device 2 can access various sensors and actuators of the vehicle 1 in order to determine vehicle parameters, such as the current vehicle speed, the current acceleration or also the current vehicle position (e.g. GPS coordinates).In summary, the invention provides a method for planning behavior in automated vehicles using multimodal probabilistic object predictions, with which a method of the generic type and an associated assistance system are disclosed, with which a human driving behavior can be simulated, so that the acceptance of the occupants is increased without reducing the safety in the respective traffic situation. In the sense of the invention, a probabilistic object prediction or probability announcement about the present traffic situation (such as the occurrence of a specific event in the traffic situation) is understood to mean that this exists at a specific probability. The proposed assistance system or AD system (=ADS, Automated Driving System) generally comprises a module which predicts the movement of other objects into the future (prediction planner or "prediction planner"), usually in the form of a plurality of movement trajectories or a trajectory group which are each provided with a probability in order to map the uncertainty of the prediction. The invention discloses a method for taking this trajectory group into account in the movement planning of the ego vehicle as a function of its probability, in order to achieve the most human possible behaviour at the end. For this purpose, the algorithm aims at planning the reaction to these early and comfortably in the case of safe predictions, while planning late and with strong maneuvers in the case of rather unlikely predictions is to be performed in order to prevent false braking on the one hand, but nevertheless still to be able to prevent an accident.List of reference characters1 (Ego) vehicle 2 control device 3 steering 4 engine control 5 brake 6 radar sensor 7 lidar sensor 8 camera 9 a- 9 dtrans sonic sensor 10 vehicle 11 vehicle 12 vehicle 13 vehicle 14 vehicleReferences included in the specificationThis list of documents cited by the applicant has been produced in an automated manner and is only included for the better information of the reader. The list is not part of the German patent application or utility model application. The DPMA does not take any adhesion for any faults or omissions.Patent Literature citedWO 2016 119 952 A1

[0006] DE 10 2019,003 557 A1

[0007] EP 2 681 085 B1

[0008] DE 10 2020 200 183 A1

[0009]

Claims

Method for controlling an ego vehicle (1) in a traffic situation, in which the ego vehicle (1) comprises a control device (2) for controlling the ego vehicle (1), at least one environment sensor for object detection and environment detection, and a prediction planner which can predict future trajectories for at least one detected object, wherein a plurality of possible future trajectories are predicted for the detected object, a probability is determined for each predicted trajectory whether the object will follow this trajectory or not, and the ego vehicle (1) is controlled on the basis of a trajectory planning which takes into account at least one of the predicted trajectories of the object, and a plurality of planning options are determined for the trajectory planning of the ego vehicle (1), wherein the planning options differ thereby, considering only predicted trajectories selected based on their probability, and evaluating the planning option to determine whether intervention in the control of the ego vehicle (1) is necessary at the current time, wherein a planning option for controlling the ego vehicle (1) is selected from the determined planning options based on the evaluation by - selecting a planning option that performs such intervention at the current time, and - selecting a planning option that changes the current control of the ego vehicle (1) as little as possible at the current time.Method according to claim 1, characterized in that the planning options differ in that different planning strategies and / or planning parameters are taken into account in each case, insofar as intervention time and intervention comfort are varied, wherein the planning options, which use essentially only probable predictions, provide an early and comfortable intervention, and planning options, which also use unlikely predictions, cause a late and therefore rather uncomfortable intervention.Method according to Claim 1 or 2, characterized in that a cruise control and / or a lateral movement of the ego vehicle (1) is provided as an intervention in the control of the ego vehicle (1).Method according to one of the preceding claims, characterized in that the method is carried out repeatedly at different times for a traffic situation.Method according to one of the preceding claims, characterized in that a first planning option (option 1) is provided in which only predicted trajectories with a probability greater than a first value p1 are taken into account, and a second planning option (option 2) is provided in which only predicted trajectories with a probability greater than a second value p2 are taken into account, wherein the value p1 is greater than the value p2.Method according to one of the preceding claims, characterized in that for the evaluation of the planning option, whether or not an intervention in the control of the ego vehicle (1) is necessary at the current point in time, it is taken into account whether a collision with this object is imminent by a predicted trajectory for an object.Method according to one of the preceding claims, characterized in that multiple planning options N are provided.Method according to claim 7, characterised in that each planning option N is planned at least twice, different values for the probability being used each time.Method according to one of the preceding claims, characterized in that if the resulting trajectories of the multiple plannings of a planning option already differ at the start of the trajectories, the planning option is evaluated in such a way that intervention in the control of the ego vehicle (1) is necessary at the current point in time and the planning option which carries out such an intervention is thus selected.Assistance system for an ego vehicle (1), having - a control device (2) for controlling the ego vehicle (1), - at least one environment sensor for object detection and environment detection, - a prediction planner which can predict future trajectories for detected objects, wherein the assistance system is configured to control the ego vehicle (1) on the basis of a method according to one of the preceding claims.A vehicle (1) comprising an assist system according to claim 10.

Images