Method for planning target trajectory for autonomous vehicle

Abstract

The invention relates to a method for planning a target trajectory of an autonomous vehicle (1), in which a mass value (Q) is determined for each object (2, 3) detected within a planning range (P), said mass value giving a measure for the detection reliability of the object (2, 3). For each object (2, 3), a minimum permissible acceleration (aQ) at which the vehicle (1) brakes to the object (2, 3) is determined as a function of the mass value (Q) of the object. The trajectory candidates and the object costs determined therefor are evaluated as a function of the acceleration (ax (t)) specified by the respective trajectory candidate and as a function of the minimum allowable acceleration (aQ) during braking to the object (2, 3). The target trajectory is selected as a function of the trajectory cost from the set of trajectory candidates, and additionally taking into account the evaluated trajectory candidate and the object cost determined therefor.

Description

Technical Field

[0001] The invention relates to a method for planning a target trajectory for an autonomous vehicle according to the preamble of claim 1. Background Art

[0002] A method for planning a target trajectory is known from DE 10 2020 108 857 A1, which should be driven automatically by a vehicle. The planning is based on determining a discrete set of candidates for the target trajectory and selecting a candidate from the determined candidate set. The selection is based on a specified cost function. When it is found that the framework conditions to be followed and the driving task to be performed change, the cost function for each candidate track segment is matched to the changed framework conditions and driving task, and a pre-control of the selection is performed so that a lower cost is allocated to the track segment that is more suitable for complying with the changed framework conditions and performing the changed driving task than other track segments. The target trajectory, as a data set, includes information about the position route that the vehicle should follow when driving on the target trajectory, and also includes other information about the acceleration and driving speed used for movement when the vehicle is driving on the target trajectory. In addition, a discretization of the trajectory set is performed, and the target trajectory is selected from it, wherein a predetermined set of track base points sorted in time is determined in the prediction range, and a track set extending through different track base points sorted in time is determined. The set of trajectories extending through the trajectory base point in chronological order forms a trajectory cluster, which is taken into consideration as a candidate when selecting a target trajectory. For each trajectory in the trajectory cluster, a cost is determined using a defined cost function, wherein the total cost of a trajectory segment is determined using a weighted sum of the costs of the trajectory segments determined for different framework conditions. The trajectory cost is determined using the sum of the total costs of its trajectory segments. The trajectory with the lowest cost is then selected from the trajectory cluster as the target trajectory.

[0003] A method for controlling a vehicle is known from DE 10 2020 200 183 A1, in which a probabilistic free space map with static and dynamic objects is generated for the vehicle's surroundings, and in which a vehicle trajectory is planned taking into account the probabilistic free space map and optimized using a cost function.

[0004] DE 10 2015 016 544 A1 discloses a method for determining an avoidance trajectory of a vehicle around an obstacle, wherein provision is made for optimizing the avoidance trajectory with respect to defined criteria, wherein these criteria include an upper limit for vehicle acceleration, a lower limit for the distance to the obstacle and a lateral acceleration at the end of the avoidance trajectory.

[0005] DE 10 2016 218 121 A1 discloses a method for trajectory planning, wherein a motion model of an autonomous vehicle, collision-relevant objects, and physical constraints of driving are used for trajectory planning. In this case, a plurality of feasible trajectories are evaluated using a cost function, and the trajectory with the lowest cost is then selected. Summary of the invention

[0006] The purpose of this invention is to propose a new method for planning target trajectories for an autonomous vehicle.

[0007] According to the invention, this object is achieved by a method having the features of claim 1 .

[0008] Advantageous embodiments of the invention are the subject matter of the dependent claims.

[0009] In a method for planning a target trajectory for an automated driving vehicle, in particular for a highly automated driving or autonomous driving vehicle, a set of trajectory candidates is specified for a specified planning range, wherein each trajectory candidate specifies a path that the vehicle should follow when the trajectory candidate is selected as the target trajectory, and specifies the acceleration that the vehicle should have when following the path. Objects are detected within the planning range, and the trajectory candidates are respectively associated with trajectory costs by means of a specified cost function, wherein the cost function includes object costs associated with the detected objects. In this case, the object cost of the object for the trajectory candidate increases as the distance between the object and the trajectory candidate decreases. The target trajectory is selected from the set of trajectory candidates based on the trajectory cost.

[0010] According to the invention, a mass value is determined for each detected object, which gives a measure of the reliability of the object detection. For each object, a minimum permissible acceleration of the vehicle for braking to the object is determined as a function of the mass value of the object. The trajectory candidates and the object costs determined therefor are evaluated as a function of the acceleration specified by the respective trajectory candidate and as a function of the minimum permissible acceleration when braking to the object. The selection of the target trajectory is additionally performed taking into account the evaluation of the trajectory candidates and the object costs determined therefor.

[0011] The present method makes it possible to take into account the detection quality, i.e. the probability of existence, of an object in a traffic scene (e.g. also with a plurality of objects) and to adapt the maximum braking intervention accordingly during the automatic control of the vehicle. This ensures that an object with a lower detection quality never alone leads to a stronger braking intervention than is permitted. However, the option of increasing the braking force when an object with a higher detection quality requires this is retained. Objects with a lower detection quality can also be avoided.

[0012] In one possible embodiment of the method, the trajectory candidate with the lowest trajectory cost is selected from the set of trajectory candidates as the target trajectory. Thus, the trajectory in which the detected objects have the least impact on the vehicle and / or in which the number of objects is the least is selected as the target trajectory.

[0013] In another possible embodiment of the method, a classification is performed during the evaluation, wherein a distinction is made between unscreened categories and screened categories in the classification. In this case, the unscreened categories are associated with all object costs and related trajectory candidates. That is, trajectory candidates are not screened out. The screened categories are only associated with corresponding object costs and related trajectory candidates, for which the minimum value of the acceleration specified by the corresponding trajectory candidate is greater than the minimum permissible acceleration of the corresponding object. The categories screened in this way can be associated only with corresponding trajectory candidates, whose accelerations are within the permissible range determined by the minimum permissible acceleration. That is, in the screened categories, trajectory candidates that would result in impermissible forced braking are missing. By taking these categories into consideration, the reliability of the method can be further improved.

[0014] In another possible design of the method, when selecting a target trajectory from an unfiltered category and a filtered category, a trajectory candidate with a lower trajectory cost is selected. For two selected trajectory candidates, the minimum value of the acceleration specified by the corresponding trajectory candidate is determined, and the candidate with the larger minimum value is selected as the target trajectory. In this way, for many feasible trajectories, a trajectory that performs weaker braking to the object under the same safety condition can be selected as the target trajectory, thereby improving passenger comfort under the same safety condition. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Embodiments of the present invention are explained in more detail below with reference to the accompanying drawings.

[0016] in:

[0017] Figure 1 Schematically shows the time curve of the vehicle's acceleration and the associated time curve of its longitudinal position as it follows the path of the trajectory,

[0018] Figure 2 Schematically shows the time curve of the vehicle's acceleration and the associated time curve of its longitudinal position when it follows the path of different trajectories,

[0019] Figure 3 The minimum permissible acceleration depending on the reliability of object detection is schematically shown, and

[0020] Figure 4 The method flow for determining the cost of an object is schematically shown.

[0021] Figure 5The method flow of selecting a target trajectory from different categories of trajectories is schematically shown.

[0022] Figure 6 A top view of a traffic scene with vehicles and a plurality of objects is schematically shown.

[0023] In all the figures, parts which correspond to one another are provided with the same reference numerals. DETAILED DESCRIPTION

[0024] exist Figure 1 In the Figure 6 The automated driving, in particular the highly automated driving or autonomous driving, vehicle 1 further shown follows Figure 5 The trajectories T1 to Tn shown in FIG. 1 are the time-dependent curves of their time-dependent acceleration ax(t) and their longitudinal position x(t).

[0025] Each trajectory T1 to Tn is described by its longitudinal position x(t), its transverse position and its derivatives, i.e. its longitudinal and transverse velocities, as well as its longitudinal acceleration ax(t) and transverse acceleration up to the planning horizon P. The longitudinal position x(t) is obtained by integrating a predefined acceleration ax(t). The acceleration ax(t) is defined accordingly so that the integrated longitudinal position x(t) satisfies the following condition:

[0026] For t = [0, P]

[0027] x1(t) <x2(t) (1)

[0028] and

[0029] min(ax1(t)) <min(ax2(t))。 (2)

[0030] Where: x1(t) = longitudinal position x(t) when following trajectory T1,

[0031] x2(t) = longitudinal position x(t) when following trajectory T2,

[0032] ax1(t) = longitudinal acceleration while following trajectory T1,

[0033] ax2(t) = longitudinal acceleration while following trajectory T2.

[0034] In this case, in addition to the path that the vehicle 1 should follow, the trajectory T1 to Tn also specifies the acceleration profile that the vehicle 1 uses to follow the path. The acceleration ax(t) here refers to the time acceleration profile when following the trajectory T1 to Tn. When the vehicle 1 decelerates, the acceleration ax(t) is negative. The acceleration ax(t) has a minimum value min(ax(t)).

[0035] In Figure 2 shown are examples of different time curves of the acceleration ax1(t) and ax2(t) of vehicle 1 and the associated time curves of the longitudinal position x1(t) and x2(t) of vehicle 1, as well as the minimum values min(ax1(t)) and min(ax2(t)) of the acceleration ax1(t) and ax2(t), when the vehicle follows paths of different trajectories T1, T2.

[0036] In a method for planning a target trajectory for an autonomous vehicle 1, a plurality of trajectories T1 to Tn are specified, for example in the form of a trajectory cluster. The number of trajectories T1 to Tn is, for example, in the order of magnitude of 1000.

[0037] The following conditions apply here:

[0038] x1(t) < x2(t), and min(ax1(t)) < min(ax2(t)). (3)

[0039] This condition means that the trajectories T1 to Tn of the trajectory cluster are determined such that: when min(ax1(t)) < min(ax2(t)), x1(t) < x2(t) applies for any two trajectories T1, T2. This is a boundary condition that applies when specifying the trajectories T1 to Tn.

[0040] That is, for the specified planning range P, a set of trajectory candidates is specified for planning the target trajectory, where each trajectory candidate specifies the path that vehicle 1 should follow when selecting this trajectory candidate as the target trajectory, and specifies the acceleration ax(t) that vehicle 1 should have when following this path.

[0041] Furthermore, objects 2, 3 in the Figure 6 planning range P shown further below are detected, and the trajectory candidates are respectively associated with trajectory costs by means of the specified cost function for planning the target trajectory. Here, the cost function includes object costs related to the detected objects 2, 3, where the trajectory costs of the objects 2, 3 for a trajectory candidate increase as the distance between the objects 2, 3 and the trajectory candidate decreases. Then, according to the trajectory costs, the target trajectory is selected from the set of trajectory candidates.

[0042] Here, a mass value Q is determined for each detected object 2, 3, which gives a measure of the detection reliability of the object 2, 3, that is, the probability of the presence of the object 2, 3 in the surrounding environment of vehicle 1.

[0043] Furthermore, for each object 2, 3, the lowest allowable acceleration aQ for vehicle 1 to brake to the corresponding object 2, 3 is determined according to the mass value Q of the object.

[0044] In Figure 3, which is related to the detection reliability of objects 2 and 3, that is, the minimum permissible acceleration aQ related to the mass value Q of the object. The minimum permissible acceleration aQ is a negative value when the vehicle brakes. Since negative acceleration is a deceleration, the minimum permissible acceleration aQ also determines the maximum permissible deceleration.

[0045] Determine the minimum permissible acceleration aQ when braking against objects 2, 3, taking into account the following conditions:

[0046] min(ax(t))≥aQ. (4)

[0047] In this case, min(ax(t)) is the minimum acceleration on the trajectory T1 to Tn, wherein min(ax(t)) also has a negative value during braking.

[0048] Therefore, formula (4) is used to check whether the acceleration aQ during braking remains within a permissible range, wherein the permissible range is determined by the mass value Q of the respective object 2 , 3 .

[0049] The main point here is that for each value of the minimum permissible acceleration aQ there is at least one defined trajectory T1 to Tn, for which:

[0050] min(ax(t))=aQ. (5)

[0051] exist Figure 3 In the diagram, the minimum permissible acceleration aQ and the mass value are divided into three regions: low, medium and high. It can be seen here that braking with the maximum force is permitted only when the objects 2, 3 have a high mass value Q (indicated by the unshaded area). On the contrary, for objects 2, 3 with medium and low mass values ​​Q, such braking is not permitted (indicated by the shaded area).

[0052] In order to avoid problems when planning a target trajectory, when the vehicle 1 can only avoid a collision with an object 2, 3 with a low mass value Q with a target trajectory having stronger braking, it is additionally provided that the trajectory candidates and the object costs determined therefor are evaluated based on the acceleration ax(t) specified by the corresponding trajectory candidate and based on the minimum permissible acceleration aQ during braking to the object 2, 3, and that the selection of the target trajectory is additionally performed taking into account the evaluated trajectory candidates and the object costs determined therefor.

[0053] Here, the object cost of the object is evaluated for each trajectory of the trajectory cluster, wherein, as described above, for a trajectory candidate, the object cost of the objects 2 , 3 increases as the longitudinal and lateral distances between the objects 2 , 3 and the trajectory candidate decrease.

[0054] If there are multiple objects 2, 3, each of the objects 2, 3 may contribute to the object cost of the trajectory candidate. In such a case, it is checked which of the objects 2, 3 is most relevant in terms of the object cost, i.e. contributes the most, and the object cost determination for the trajectory candidate is done only based on the most relevant object 2, 3. In other words: less relevant objects 2, 3 remain ignored when determining the object cost.

[0055] When evaluating trajectory candidates and determining object costs for them, a classification is performed in which a distinction is made between unselected class A and selected class F, both of which are Figure 5 Here, the unfiltered class A is associated with all object costs and associated trajectory candidates. In contrast, the filtered class F is only associated with the corresponding object costs and associated trajectory candidates for which the minimum value min(ax(t)) of the acceleration specified by the corresponding trajectory candidate is greater than the lowest allowed acceleration aQ of the corresponding object 2, 3.

[0056] If the mass values ​​of objects 2 and 3 are such that the maximum acceleration aQ is -5 m / s 2 Braking to objects 2 and 3, and the corresponding trajectories T1 to Tn require braking, during which the acceleration min(ax(t)) drops to -2 m / s 2 , then the condition min(ax(t))>aQ is satisfied, i.e. -2m / s 2 >-5m / s 2 The corresponding trajectories T1 to Tn and the object costs are associated with the categories A and F.

[0057] If the mass values ​​of objects 2 and 3 are such that the maximum acceleration allowed is aQ = -5 m / s 2 Braking to objects 2 and 3, and the corresponding trajectories T1 to Tn require braking, and the acceleration min(ax(t)) during this braking is reduced to -6m / s 2 , then the condition min(ax(t))>aQ is not satisfied. Then the corresponding trajectories T1 to Tn and the object costs are associated with the unfiltered category A and not with the filtered category F.

[0058] exist Figure 4 A possible exemplary embodiment of the method sequence for determining the cost of an object is shown in FIG.

[0059] In this case, in a first branch V1 , a check is made as to whether all trajectory candidates of a trajectory cluster have been evaluated. If so, as indicated by the yes branch J1 , the method ends.

[0060] If not, as indicated by the No branch N1, a check is made in a second branch V2 whether all objects 2, 3 of the trajectory candidate have been evaluated. If so, as indicated by the Yes branch J2, the next trajectory candidate is selected in method step S1 and the method is restarted for this candidate.

[0061] If this is not the case, as indicated by the No branch N2 , in a further method step S2 the object costs are determined for the respective object 2 , 3 of the respective trajectory candidate.

[0062] Then, in the third branch V3 , the condition is checked according to formula (4) and it is confirmed whether the minimum acceleration on the corresponding trajectory T1 to Tn is greater than the minimum permissible acceleration aQ.

[0063] If yes, this is indicated by the yes branch J3 , then in a further method step S3 the object costs of the trajectories T1 to Tn are updated in the unfiltered category A and in the filtered category F for the respective trajectory.

[0064] If this is not the case, as indicated by the No branch N3 , in a further method step S4 the object costs of the trajectories T1 to Tn are updated for the respective trajectories only in the unfiltered category A.

[0065] After method step S3 or method step S4 has been executed, in a further step S5 the next object 2 , 3 of the corresponding trajectory candidate is selected and the checking in branch V2 is continued for this object 2 , 3 .

[0066] That is to say: for each trajectory T1 to Tn of the trajectory cluster and each object 2, 3 in the set of objects 2, 3, the object cost for the corresponding trajectory T1 to Tn is calculated. If the condition min(ax(t))>aQ is satisfied for the corresponding trajectory T1 to Tn and the corresponding object 2, 3, the object cost will be associated with both category A and category F, otherwise it will only be associated with the unfiltered category A.

[0067] When all object costs have been calculated for all possible combinations of trajectories T1 to Tn, two optimal trajectories T1 to Tn are determined from category A and category F, and one of the two trajectories is selected as the target trajectory.

[0068] Figure 5 A possible embodiment of a method flow for such determination and selection is shown.

[0069] In this case, in method step S6 , from all trajectories T1 to Tn of category A that have not been filtered out, the trajectory T1 to Tn having the lowest cost is determined.

[0070] In a further method step S7 , from all the trajectories T1 to Tn of the filtered category F, the trajectory T1 to Tn having the lowest cost is determined.

[0071] Subsequently, in branch V it is checked whether the minimum acceleration min(ax(t)) on the trajectory T1 to Tn of the unfiltered category A is greater than or equal to the minimum acceleration min(ax(t)) on the trajectory T1 to Tn of the filtered category F.

[0072] If yes, indicated by the yes branch J, then in method step S8 the corresponding trajectory T1 to Tn of the unfiltered category A is selected as the target trajectory. If no, indicated by the no branch N, then in method step S9 the corresponding trajectory T1 to Tn of the unfiltered category A is selected as the target trajectory.

[0073] Figure 6 A top view of a traffic scene with a vehicle 1 and a plurality of objects 2 , 3 is shown, wherein object 2 has a low quality value Q and object 3 has a high quality value Q.

[0074] If, for example, trajectory T5 is the lowest cost trajectory candidate in the screened category F, and trajectory T5 specifies that the acceleration ax(t) decreases to, for example, -5 m / s 2 If the value of is less than 5m / s, then min(ax5(t))=-5m / s 2 .

[0075] In the unfiltered category A, for example, trajectory T1 is the lowest cost trajectory candidate. For example, trajectory T1 specifies that the acceleration ax(t) should be reduced to, for example, -3 m / s 2 Therefore, min(ax1(t))=-3m / s applies. 2 .

[0076] Pass -3m / s 2 ≥-5m / s 2 , the condition min(ax1(t))≥min(ax5(t)) is satisfied, so trajectory T1 is selected as the target trajectory. On the contrary, if the condition is not satisfied, trajectory T5 is selected as the target trajectory.

[0077] If for example min(ax1(t)) = -5 m / s 2 , then both trajectories T1 and T5 brake to the object 3 traveling ahead with the same acceleration ax(t). In this case, trajectory T1 also takes into account the object 2 with a low mass value Q. In contrast, trajectory T5 does not. Since the condition min(ax1(t))≥min(ax5(t)) is also satisfied here, trajectory T1 is selected as the target trajectory and thus avoids the object 2 with a low mass value.

Claims

1. A method for planning a target trajectory for an autonomous vehicle (1), wherein - specifying a set of trajectory candidates for a specified planning horizon (P), wherein each trajectory candidate specifies a path that the vehicle (1) should follow when the trajectory candidate is selected as a target trajectory, and specifies an acceleration (ax(t)) that the vehicle should have when following the path, - detecting objects (2, 3) within the planning range (P), - in each case the trajectory candidates are associated with a trajectory cost by means of a predetermined cost function, wherein: The cost function comprises an object cost associated with the detected object (2, 3), wherein the trajectory cost of the object (2, 3) for the trajectory candidate increases as the distance between the object (2, 3) and the trajectory candidate decreases, and - the target trajectory is selected from the set of trajectory candidates based on the trajectory cost, characterized in that - determining a quality value (Q) for each detected object (2, 3), said quality value providing a measure of the reliability of the detection of said object (2, 3), - for each object (2, 3), determining, as a function of the mass value (Q) of said object, a minimum permissible acceleration (aQ) for braking said vehicle (1) against said object (2, 3), - the trajectory candidates and the object costs determined therefor are evaluated according to the acceleration (ax(t)) specified by the respective trajectory candidate and according to the minimum permissible acceleration (aQ) for braking to the object (2, 3), and The target trajectory is additionally selected taking into account the evaluation of the trajectory candidates and the object costs determined therefor.

2. The method according to claim 1, It is characterized in that A trajectory candidate with a lowest trajectory cost is selected from the set of trajectory candidates as the target trajectory.

3. The method according to claim 1 or 2, It is characterized in that - classification at the time of said assessment, - distinguish between unfiltered categories (A) and filtered categories (F) in the classification, - Associate all object costs and associated trajectory candidates with the unfiltered class (A) - The filtered category (F) is only associated with the corresponding object costs and the associated trajectory candidates for which the minimum value (min(ax(t)) of the acceleration (ax(t)) specified by the corresponding trajectory candidate is greater than the minimum allowed acceleration (aQ) of the corresponding object (2, 3).

4. The method according to claim 3, It is characterized in that When selecting the target trajectory - selecting the trajectory candidate with the lowest trajectory cost from the unfiltered category (A) and the filtered category (F), respectively, - for the two selected trajectory candidates, respectively determining the minimum value (min(ax(t)) of the acceleration (ax(t)) specified by the corresponding trajectory candidate, and - Select the trajectory candidate with the larger minimum value (min(ax(t)) as the target trajectory.

Images