Method for planning optimal driving behavior for an at least partially autonomous ego vehicle

Abstract

The present invention discloses a method (100) for planning an optimal driving behavior for an at least partially autonomously driving ego vehicle (50) with the following steps: - Obtaining (102) first data (12) which depict a planned route (20) of the ego vehicle (50); - creating (104) a road graph (30) based on the first data (12) for a section (22) of the planned route (20) of the ego vehicle (50), which section consists of at least one lane section (23) that follows the planned route (20); - determining (106) at least a first route (32) of the road graph (30), which begins at a first position of the ego vehicle (50) and consists of a sequence of lane sections (23, 24) of the road graph (30); - creating (108) at least one first cost volume (40) for the first route (32) of the planned route (20); - determining (110) potential candidate trajectories (42) for the first route (32) of the planned route (20); - Evaluating (112) the number of potential candidate trajectories (42) based on at least the first cost volume (40) for the planned route (20); - selecting (114) at least one optimal candidate trajectory (44) from the number of potential candidate trajectories (42) on the basis of the evaluation carried out; - transmitting (116) the selected optimal candidate trajectory (44) to a control unit (52) of the ego vehicle (50) for planning the optimal driving behavior of the ego vehicle (50).

Description

[0001] The present invention relates to a method for planning an optimal driving behavior for an at least partially autonomously driving ego vehicle. State of the art

[0002] One possible goal of autonomous driving for a vehicle is to control the vehicle based on sensor data in such a way that a defined destination is reached as quickly, comfortably and safely as possible, for example without causing collisions or violating traffic regulations.

[0003] This driving task can be divided into the subtasks of perception, prediction, planning, and control. The task of perception is to extract relevant information from the sensor data, such as the position of objects (other vehicles or road users), identify lane markings, and recognize traffic signs, etc. Since the detected objects are usually dynamic obstacles, their future position must then be predicted (prediction) in order to avoid collisions. Based on this, the task of planning is to generate a trajectory that will be adjusted in the control subtask.

[0004] A special method among the AI-based approaches to trajectory planning is the "End-to-End Interpretable Neural Motion Planner" (NMP) published by Uber ATG. This end-to-end approach spans the aforementioned subtasks of perception, prediction, and planning and, based on so-called lidar point clouds and HD maps containing information about the road layout, etc., generates, on the one hand, a prediction for other road users and, on the other hand, a so-called cost volume. This cost volume assigns a cost value to each possible position in the field of view for each point in time within the planning horizon, which indicates how favorable (or unfavorable) it is to be at that location at that time. Various trajectories that the vehicle is theoretically capable of executing are then evaluated based on this cost volume. The best trajectory is passed on to a controller, which is supposed to steer the vehicle along this plan.

[0005] For NMP, the inputs, such as those based on HD maps and lidar point clouds, are converted into a Cartesian grid and then processed using convolutional neural networks (CNNs). In the area of ​​prediction, state-of-the-art approaches increasingly use vectorized inputs instead of raster representations. This allows for a better representation of the graph structure of the road network. Another known approach involves first aggregating information about other road users along the road graph. Traversals through the road graph are then determined.

[0006] When predicting trajectories, only the information aggregated along the respective traversals is considered. This is intended to achieve a separation between "path" or "maneuver planning" and the determination of the appropriate speed profile.

[0007] However, a weakness of NMP and all grid-based planning algorithms is that, due to the limited interaction radius of information processing in CNNs, interactions between distant agents are often inadequately considered. This poses a problem because, depending on the situation, relevant agents may be at a great distance from the ego vehicle. For example, when overtaking a vehicle traveling at 70 km / h at 100 km / h, oncoming vehicles up to a distance of 250-300 m must be considered to ensure sufficient overtaking distance.

[0008] In the case of graph-based approaches, this is easier because, for example, the interaction radius along the lanes can be larger than across them. Furthermore, due to the representation as a rectangular grid, the cost volume used in the NMP also includes off-road positions. On the one hand, this results in unnecessary computational effort; on the other hand, it cannot guarantee that off-road trajectories are excluded.

[0009] Furthermore, the NMP plan is generated solely based on sensor data and map information. Accordingly, the trajectory is selected without considering the desired route. Thus, while NMP enables collision-free driving, it does not offer the option of considering a given navigation destination.

[0010] Since the PGP model is a predictive approach, collision and rule violation avoidance is only implicitly considered, as the datasets used for training do not include trajectories that demonstrate such behavior. Therefore, high collision and off-road rates are to be expected when using this model as a planner.

[0011] It is therefore the object of the present invention to provide a solution by means of which planning for a driving behavior of an at least partially autonomous vehicle can be improved. Disclosure of the invention

[0012] This task is solved by a method for planning an optimal driving behavior for an at least partially autonomously driving ego vehicle with the features of the independent claim.

[0013] According to a first aspect, the disclosure relates to a method for planning an optimal driving behavior for an at least partially autonomously driving ego vehicle, which comprises the following steps: The first step involves collecting initial data that depicts a planned route of the ego vehicle. In a second step, a road graph is created based on the first data for a section of the planned route of the ego vehicle, which consists of at least one lane section that follows the planned route. In a third step, at least one first route of the road graph is determined, which begins at a first position of the ego vehicle and consists of a sequence of lane sections of the road graph. In a fourth step, at least an initial cost volume is created for the first route of the planned route. In a fifth step, potential candidate trajectories for the first route of the planned route are determined. In a sixth step, the number of potential candidate trajectories is evaluated based on the minimum initial cost volume for the planned route. In a seventh step, at least one optimal candidate trajectory is selected from the number of potential candidate trajectories based on the evaluation carried out. In an eighth step, the selected optimal candidate trajectory is transmitted to a control unit of the ego vehicle to plan the optimal driving behavior of the ego vehicle.

[0014] A fundamental idea of ​​the present invention is that the present invention is based on an artificial intelligence-based planning algorithm. This algorithm builds on both the NMP and the PGP approach. The information is aggregated analogously to PGP. From the encodings thus generated, a cost volume is generated that contains only positions along the route. By evaluating kinematically feasible trajectories, a plan can then be selected for the ego vehicle that follows the route and does not leave the road. Furthermore, the generated cost volume enables effective collision avoidance.

[0015] A further aspect of the present invention is that the selection of the optimal trajectory takes into account the planned route.

[0016] In addition to collision-free driving, the present invention also offers the possibility of taking into account a predetermined navigation destination in the form of a planned route. The restriction of the planning space to a section of the road graph is used to consider only road sections and trajectories that follow the route.

[0017] A further advantageous aspect of the present invention is that the cost volume used in the NMP only considers positions along the planned route, but not positions off the planned route. This avoids unnecessary computational effort when planning the optimal trajectory.

[0018] One embodiment of the method provides that the first route represents a lane of the ego vehicle. This enables efficient planning of the ego vehicle's driving behavior.

[0019] One embodiment of the method provides that, in the step of creating the at least one cost volume, at least a second cost volume is created for at least a second route. This enables efficient planning of the driving behavior of the ego vehicle.

[0020] One embodiment of the method provides that at least the second route is arranged adjacent to the first route and follows the planned route. This enables more precise planning of the driving behavior of the ego vehicle over a longer planning horizon.

[0021] One embodiment of the method provides for the cost volume to be parameterized according to the arc length and the transverse offset of a reference line along the route. This enables efficient planning of the driving behavior of the ego vehicle.

[0022] One embodiment of the method provides for the reference line to be a center line of the lane. This enables precise planning of the driving behavior of the ego vehicle.

[0023] According to a second aspect, the disclosure relates to a vehicle with a system for carrying out the method according to the invention.

[0024] According to a third aspect, the disclosure relates to a computer program containing machine-readable instructions which, when executed on one or more computers and / or compute instances, cause the computer or computers or compute instances to carry out the method according to the invention.

[0025] According to a fourth aspect, the disclosure relates to a machine-readable data carrier and / or download product comprising the computer program.

[0026] According to a fifth aspect, the disclosure relates to one or more computers and / or compute instances with the computer program, and / or with the machine-readable data carrier and / or the download product.

[0027] Further measures improving the invention are presented in more detail below together with the description of the preferred embodiments of the invention with reference to figures. Example

[0028] It shows: Fig. 1 shows a schematic flow diagram of a method 100 for planning an optimal driving behavior for an at least partially autonomously driving ego vehicle 50 according to an embodiment of the present invention.

[0029] In step 102, first data 12 are retrieved, which map a planned route 20 of the ego vehicle 50. This first data, or more generally, data 12, can include sensor data and / or environmental data of the ego vehicle 50.

[0030] In step 104, a road graph 30 is created 104 on the basis of the first data 12 for a section 22 of the planned route 20 of the ego vehicle 50, which consists of at least one lane section 23 that follows the planned route 20.

[0031] In step 106, at least one first route 32 of the road graph 30 is determined, which begins at a first position of the ego vehicle 50 and consists of a sequence of lane sections 23, 24 of the road graph 30.

[0032] Optionally, the first route 32 can represent a lane of the ego vehicle 50.

[0033] In step 108, at least one first cost volume 40 is created for the first route 32 of the planned route 20.

[0034] Optionally, in step 108 of creating the at least one cost volume 40, at least a second cost volume 46 can be created for at least one second route 34. Thus, a number of routes with corresponding cost volumes can be considered.

[0035] Optionally, at least the second route 34 can be arranged adjacent to the first route 32 and follow the planned route 20.

[0036] Optionally, the cost volume 40, 46 or each of the created cost volumes for the corresponding route sections can be parameterized according to the arc length and transverse offset of a reference line of the respective route section 32, 34. The reference line can preferably be designed as a center line of the lane 22.

[0037] In step 110, potential candidate trajectories 42 for the first route 32 of the planned route 20 are determined.

[0038] In step 112, the number of potential candidate trajectories 42 is evaluated based on the at least first cost volume 40 for the planned route 20.

[0039] In step 114, at least one optimal candidate trajectory 44 is selected 114 from the number of potential candidate trajectories 42 on the basis of the evaluation carried out.

[0040] In step 116, the selected optimal candidate trajectory 44 is transmitted to a control unit 52 of the ego vehicle 50 for planning the optimal driving behavior of the ego vehicle 50.

[0041] The method according to the invention can be mapped with an architecture which is described in more detail below with further aspects of the present invention.

[0042] The inventive architecture comprises an encoder, an aggregator, a decoder and a planner, which are connected to each other sequentially and signal-wise.

[0043] The architecture of the encoder and aggregator is based on PGP. However, unlike PGP, when determining traversals through the road graph, additional assurance is given that the traversals follow the planned route.

[0044] Instead of directly determining trajectories from these traversals, a cost volume is then generated along this section of the road network (decoder). This is intended to achieve a stronger focus on collision avoidance.

[0045] This cost volume is represented in the curvilinear coordinates of the respective road section, meaning it does not include any positions off the road or the planned route. This representation also offers the advantage that trajectories that run along the lane center are vertical straight lines. This creates a strong "inductive bias" for machine learning.

[0046] The final selection of an optimal trajectory 44 for the ego vehicle along the planned route 20 is then performed by evaluating a large number of kinematically feasible candidate trajectories 42 within this cost volume. Since only trajectories within this cost volume are evaluated, it can be guaranteed that no trajectory is selected that deviates from the roadway or the planned route 20.

[0047] Furthermore, it should be noted that it may often be necessary to include adjacent lanes in the planning of the driving behavior for the ego vehicle 50: lane changes, overtaking maneuvers, bottlenecks, etc. The proposed model of the present invention can then be easily extended by estimating cost volumes for multiple lanes that do not overlap and are therefore trivially combinable. Lanes that are not on the target route (e.g., traveling in the opposite direction) are identifiable as such in the map encoding and can therefore be treated differently in the aggregator and encoder.

Claims

[1] Method (100) for planning an optimal driving behavior for an at least partially autonomously driving ego vehicle (50) with the following steps: - Obtaining (102) first data (12) which depict a planned route (20) of the ego vehicle (50); - creating (104) a road graph (30) based on the first data (12) for a section (22) of the planned route (20) of the ego vehicle (50), which section consists of at least one lane section (23) that follows the planned route (20); - determining (106) at least a first route (32) of the road graph (30), which begins at a first position of the ego vehicle (50) and consists of a sequence of lane sections (23, 24) of the road graph (30); - creating (108) at least one first cost volume (40) for the first route (32) of the planned route (20); - determining (110) potential candidate trajectories (42) for the first route (32) of the planned route (20); - Evaluating (112) the number of potential candidate trajectories (42) based on at least the first cost volume (40) for the planned route (20); - selecting (114) at least one optimal candidate trajectory (44) from the number of potential candidate trajectories (42) on the basis of the evaluation carried out; - transmitting (116) the selected optimal candidate trajectory (44) to a control unit (52) of the ego vehicle (50) for planning the optimal driving behavior of the ego vehicle (50). [2] Method (100) according to claim 1, wherein the first route (32) represents a lane of the ego vehicle (50). [3] Method (100) according to one of the preceding claims, wherein in the step (108) of creating the at least one cost volume (40), at least a second cost volume (46) is created for at least one second route (34). [4] Method (100) according to claim 3, wherein the at least second route (34) is arranged adjacent to the first route (32) and follows the planned route (20). [5] Method (100) according to one of the preceding claims, wherein the cost volume (40, 46) is parameterized according to arc length and transverse offset of a reference line of the route (32, 34). [6] Method (100) according to claim 5, wherein the reference line is formed as a center line of the lane (22). [7] Vehicle with a system for carrying out the method according to one of the preceding claims. [8] A computer program comprising machine-readable instructions which, when executed on one or more computers and / or compute instances, cause the computer or computers or compute instances to carry out the method according to any one of claims 1 to 6. [9] Machine-readable data carrier and / or download product with the computer program according to claim 8. [10] One or more computers and / or compute instances with the computer program according to claim 8 and / or with the machine-readable data carrier and / or download product according to claim 9.