Method for determining control parameters for driving a vehicle

Abstract

A method for determining control parameters for driving a host vehicle is provided. Static and dynamic characteristics detected in an external environment of the host vehicle are determined via a perceptual system of the host vehicle. The following operations are performed via a processing unit of the host vehicle: determining a plurality of driving hypotheses by applying static and dynamic characteristics, each driving hypothesis including a drivable path of the host vehicle and a trajectory distribution matching the drivable path, selecting a most suitable driving hypothesis from the plurality of driving hypotheses, and determining a driving hypothesis of the host vehicle by applying static and dynamic characteristics. A most suitable trajectory is determined for a trajectory distribution associated with the most suitable driving hypothesis, and control parameters for driving the primary vehicle are determined from the most suitable trajectory.

Description

Technical Field

[0001] The present disclosure relates to a computer-implemented method for determining control parameters for driving a main vehicle. Background Art

[0002] For autonomous driving and various advanced driver assistance systems (ADAS), it is important and still a challenging task to provide or support the safe movement of a main vehicle in a reliable manner. This task includes main components such as perception, prediction, planning, and control.

[0003] Recently, machine learning has increasingly dominated in the field of perception systems. That is, machine learning algorithms have been applied to perception sensors and the data provided by these sensors. This is especially true for vision-based perception systems. Similarly, the field of traffic scene and trajectory prediction is also dominated by machine learning-based methods. This is due to the high complexity of real-world traffic scenes, and it is infeasible or impossible to explicitly model real-world traffic scenes. However, in the fields of planning and control, model-based and rule-based methods still dominate.

[0004] On the one hand, different methods for perception and prediction and on the other hand for planning and control may require a large amount of computational effort, and suitable interfaces are needed between the corresponding implementations based on different methods. However, machine learning has shown strong capabilities to implement complex systems or tasks without explicitly and analytically defining and modeling the problems behind them.

[0005] Therefore, there is a need for a method that allows combining perception, prediction, planning, and control tasks to provide a human-like understanding of traffic scenes and ensure safety when driving a vehicle. Summary of the Invention

[0006] The present disclosure provides a computer-implemented method, a computer system, and a non-transitory computer-readable medium according to the independent claims. Embodiments are given in the dependent claims, the detailed description, and the drawings.

[0007] In one aspect, the present disclosure relates to a computer-implemented method for determining control parameters for driving a host vehicle. According to the method, static and dynamic characteristics detected in the external environment of the host vehicle are determined via a sensing system of the host vehicle. Via a processing unit of the host vehicle, the following method steps are performed. A plurality of driving hypotheses are determined by applying the static and dynamic characteristics, wherein each driving hypothesis includes a drivable path of the host vehicle and a trajectory distribution matching the drivable path. The most suitable driving hypothesis is selected from the plurality of driving hypotheses. A most suitable trajectory is determined for the trajectory distribution associated with the most suitable driving hypothesis. Finally, control parameters for driving the host vehicle are determined based on the most suitable trajectory.

[0008] The sensing system of the host vehicle may include a radar system, a lidar system, and / or one or more cameras to monitor the external environment or the surrounding environment of the host vehicle. Thus, the sensing system may be configured to monitor the dynamic context of the vehicle, which includes a plurality of other road users moving in the external environment. Other road users may include, for example, other vehicles and / or pedestrians. In addition, the sensing system may also be able to determine and track the motion state of the host vehicle itself. In addition, the sensing system may be configured to monitor the static context in the external environment, i.e., static entities such as lane items, traffic signs, etc., e.g., regarding their location and semantic attributes.

[0009] The dynamic characteristics determined by the sensing system may thus reflect the current motion states of other road users and the host vehicle. In contrast, the static characteristics may be related to static entities in the environment of the host vehicle, and the static characteristics may provide boundary conditions for the movement of other road users and the host vehicle.

[0010] The drivable path is defined as the spatially distributed space in which the host vehicle can drive. Thus, it only describes the drivable area or free space near the host vehicle without providing any requirements on how the host vehicle can drive within this area. The drivable path may be represented in vector form, e.g., a polygon or a grid map, where the representation may depend on the specific implementation. For example, the prediction module of the processing unit may be implemented in such a way that by applying the static and dynamic characteristics, the drivable path and the trajectory distribution of the host vehicle, i.e., each driving hypothesis, are determined.

[0011] Compared with the drivable path, the trajectory includes spatial and temporal information. Specifically, the trajectory can include a series of positions, which also include the associated time ranges. An intuitive representation of the trajectory can be a series of points. In order to determine reliable control parameters for driving the main vehicle, each driving hypothesis thus includes both the drivable path and the trajectory distribution, because for example, the drivable path may also cover the free space in front of a leading vehicle approaching the main vehicle, while the predicted trajectory or trajectory distribution will not jump in front of such a leading vehicle, i.e., unless a lane change is performed for this purpose. Instead, it may be desirable for the predicted trajectory to maintain a safe distance behind the specific future trajectory of such a leading vehicle.

[0012] In order to determine the control parameters, multiple driving hypotheses need to be considered, which include the corresponding drivable paths and the corresponding trajectory distributions, because in the real world, the external environment of the main vehicle usually involves multiple feasible paths for the future movement of the main vehicle. Such multiple feasible paths may be due to intersections, roundabouts, or even the existence of multiple lanes in the case of a straight road. Each of these feasible paths of the main vehicle may require different driving behaviors and thus may have different possible trajectory distributions.

[0013] In order to select the most suitable driving hypothesis, different criteria can be applied. The criteria include a specific cost function that can be evaluated for hypothesis selection, and include navigation information for the main vehicle. For example, the selected driving hypothesis must conform to a navigation route that can be predetermined for the main vehicle. For example, the specific cost function for evaluating the driving hypothesis can depend on the physical conditions of the driving vehicle, safety requirements, regulatory requirements, and comfort requirements.

[0014] One hypothesis associated with the corresponding minimum value of the specific cost function can be selected from multiple driving hypotheses. Similarly, a specific cost function can be applied to determine the most suitable trajectory for the trajectory distribution associated with the most suitable driving hypothesis that has already been selected.

[0015] The trajectory distribution can be represented by a set of possible points or positions of the main vehicle at different consecutive time steps. Based on the specific cost function, the most suitable point or position can be determined for each time step, and these points or positions can be connected to provide the most suitable trajectory.

[0016] The control parameters for driving the main vehicle can include, for example, the acceleration and yaw rate of the main vehicle. The control parameters can also include the angle of the steering wheel of the main vehicle, the front wheel angle, etc.

[0017] The method according to the present disclosure provides a general extensible and scalable framework for predicting the drivable path and trajectory of a host vehicle and for controlling the host vehicle. In addition, human-like scene understanding and prediction are provided to support further control modules of the host vehicle. Further, the drivable path and trajectory distribution are determined simultaneously, while according to known methods, usually the drivable area of the free space is first determined before considering the trajectory. Thus, the reliability of determining appropriate control parameters for driving the vehicle is improved.

[0018] Predicting or determining multiple driving hypotheses including corresponding drivable paths and corresponding trajectory distributions can be implemented as a machine learning algorithm, which can be trained via known drivable paths and known trajectories for a specific traffic scenario. In addition, the control parameters can also be checked or controlled via the machine learning algorithm. Further, a model-based or rule-based method can be integrated into the method to ensure safety, for example, by performing appropriate checks on the control parameters.

[0019] According to an embodiment, a corresponding set of cost functions can be determined for each of the trajectory distributions of each driving hypothesis. The most suitable driving hypothesis can be selected from the multiple driving hypotheses by utilizing the corresponding set of cost functions and by associating the corresponding drivable path of each driving hypothesis with the navigation information for the host vehicle.

[0020] Thus, in order to select the most suitable driving hypothesis, two different criteria must be met: Based on the cost function, for example, by determining the corresponding minimum value of the cost function, a decision can be made about the most suitable driving hypothesis, and in addition, the navigation requirements of the host vehicle must also be met. For example, the feasibility of the corresponding drivable path of the driving hypothesis can be checked via an external map provided to the control system of the vehicle.

[0021] In addition, by using the static and dynamic characteristics detected in the external environment of the host vehicle, a prediction of the traffic scenario within the environment of the host vehicle can be determined. The predicted traffic scenario can also be used to evaluate the set of cost functions associated with the corresponding driving hypothesis.

[0022] The prediction of the traffic scenario can include a grid-based prediction of the motion states of relevant road users (e.g., for the host vehicle itself, for other vehicles, and for pedestrians). Thus, the actual traffic scenario around the host vehicle can affect the selection of the most suitable drivable path and the most suitable trajectory, since the set of cost functions can depend at least in part on the predicted traffic scenario. That is, the set of cost functions determined for each of the trajectory distributions with respect to each driving hypothesis can be evaluated and modified based on the prediction of the traffic scenario around the host vehicle.

[0023] The cost function can include a physical cost function, which is related to the physical conditions for the driving path and trajectory distribution belonging to the corresponding driving hypothesis, and the cost function can further include a safety cost function, which can be related to the safety conditions for the driving path and trajectory distribution for the corresponding driving hypothesis. The safety cost function can be associated with a lower priority than the physical cost function.

[0024] Therefore, a hierarchy of different types of cost functions can be introduced, where the physical cost function representing the physical conditions has the highest priority. This may be due to the fact that the physical conditions of driving cannot or should not be violated. In other words, even if the safety conditions cannot be met, the selected drivable path and the selected trajectory must in any case conform to the physical conditions. For example, the safety conditions can include a safety distance from other road users.

[0025] In addition, the cost function can further include a rule cost function, which can be related to the rules for the driving path and trajectory distribution for the corresponding driving hypothesis. The rule cost function can be associated with a lower priority than the safety cost function. That is, if the violation of the rule represented by the rule cost function is related to the safety of the host vehicle, then the rule can be violated. The rules can be related to solid lines on the road that should not be crossed, stops required by specific traffic signs, speed limits, etc.

[0026] The cost function can also include a comfort cost function, which can be related to the comfort of the passengers of the host vehicle with respect to the driving path and trajectory distribution for the corresponding driving hypothesis. The comfort cost function can be associated with a lower priority than the rule cost function. That is, if it may be necessary to violate the comfort requirements of the passengers of the host vehicle in order to meet safety, rule, and physical requirements, then the comfort requirements of the passengers of the host vehicle can be violated.

[0027] Multiple trajectories can be generated for the trajectory distribution associated with the selected most appropriate driving hypothesis. The set of cost functions associated with the most appropriate driving hypothesis can be applied to the multiple trajectories, and the most appropriate trajectory can be selected from the multiple trajectories by evaluating the cost functions.

[0028] In other words, the set of cost functions can be used to select the most appropriate driving hypothesis, and within the selected most appropriate driving hypothesis, the set of cost functions can also be used to select or determine the most appropriate trajectory. External conditions such as physical conditions, safety conditions, or rule conditions can be implicitly included in the cost function, and thus the external conditions can be automatically integrated, for example, by applying machine learning algorithms, into the framework for determining the most appropriate drivable path and the most appropriate trajectory.

[0029] According to a further embodiment, a planning time range can be determined for a plurality of trajectories generated for a trajectory distribution associated with the most suitable driving hypothesis. By determining the planning time range, the generation of the plurality of trajectories can be facilitated. In addition, by restricting such predictions to the planning time range, the reliability of the predicted trajectories can be improved.

[0030] Furthermore, at least one control point can be determined for each of the plurality of trajectories according to the planning time range. The drivable path of the most suitable hypothesis and the corresponding control points can be used by an algorithm that determines control parameters associated with the most suitable trajectory.

[0031] The control points can be related to the spatial positions that the final or most suitable trajectory of the main vehicle is to reach. However, the control points include not only such spatial positions but can also include and define the speed, acceleration, heading angle, etc. of the main vehicle.

[0032] When the drivable path and the corresponding control points are fed into the algorithm for determining control parameters, the most suitable trajectory can also be determined by evaluating a cost function. The control parameters thus define the route of the most suitable trajectory. The algorithm can be a machine learning algorithm. In addition, the control points and the drivable path can also be checked with respect to predefined conditions provided by rules and / or models for the movement of the main vehicle.

[0033] As a further step, the control parameters can be refined with respect to static characteristics determined for the external environment of the main vehicle, with respect to previously planned trajectories, and / or with respect to predefined rules. Thus, the refinement of the control parameters can be performed based on further information provided by the perception system of the main vehicle. For example, semantic segmentation of the environment of the main vehicle can be provided and used for possible adjustment of the control parameters.

[0034] By using static and dynamic characteristics detected in the external environment of the main vehicle, predictions for traffic scenarios within the environment of the main vehicle and predictions for target trajectories with respect to the most suitable trajectory can be determined. By applying the predictions for traffic scenarios and the predictions for target trajectories, a safety check can be performed on the control parameters used to determine and define the most suitable trajectory.

[0035] The target trajectory can be the trajectory of other road users that is different from the main vehicle and may conflict with the most suitable trajectory. For the target trajectory, different possibilities can be considered, which is generally referred to as multimodality. For each of such multimodal target trajectories, corresponding possibilities can be provided to apply the target trajectory when performing the safety check of the control parameters. Thus, the reliability of the final control parameters can be improved.

[0036] In another aspect, the present disclosure relates to a computer system configured to perform several or all steps of the computer-implemented methods described herein.

[0037] The computer system may include a processing unit, at least one storage unit, and at least one non-transitory data memory. The non-transitory data storage and / or memory unit may include a computer program for instructing the computer to perform several or all steps or aspects of the computer-implemented methods described herein.

[0038] As used herein, terms such as processing unit and module may refer to, be, an application specific integrated circuit (ASIC), an electronic circuit, a combinational logic circuit, a field programmable gate array (FPGA), a processor (shared, dedicated, or group) executing code, other suitable components providing the functions, or a combination of some or all of the above, or including an application specific integrated circuit (ASIC), an electronic circuit, a combinational logic circuit, a field programmable gate array (FPGA), a processor (shared, dedicated, or group) executing code, other suitable components providing the functions, or a combination of some or all of the above, such as in a system on a chip. The processing unit may include a memory (shared, dedicated, or group) storing code executed by the processor.

[0039] In another aspect, the present disclosure relates to a vehicle including a sensing system, a computer system as described herein, and a control system configured to receive control parameters determined by the computer system by performing a method also as described herein.

[0040] According to an embodiment, the control system may further be configured to verify the received control parameters and apply the received control parameters during operation of the vehicle. Thus, the control system may perform additional and independent safety checks on the expected trajectory of the host vehicle provided by the received control parameters.

[0041] In another aspect, the present disclosure relates to a non-transitory computer-readable medium including instructions for performing several or all steps or aspects of the computer-implemented methods described herein. The computer-readable medium may be configured as: an optical medium such as a compact disc (CD) or a digital versatile disc (DVD); a magnetic medium such as a hard disk drive (HDD); a solid state drive (SSD); a read-only memory (ROM); a flash memory; or the like. Additionally, the computer-readable medium may be configured as a data memory accessible via a data connection such as an Internet connection. The computer-readable medium may be, for example, an online data repository or cloud storage.

[0042] The present disclosure also relates to a computer program for instructing a computer to perform some or all steps or aspects of the computer-implemented methods described herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Exemplary embodiments and functions of the present disclosure are described herein in connection with the following drawings, which schematically illustrate:

[0044] Figure 1 a diagram of a vehicle including components for performing the methods according to the present disclosure,

[0045] Figure 2 a diagram of a network architecture for the perception step and prediction step of the method,

[0046] Figure 3 an overview of the method according to the present disclosure,

[0047] Figure 4 a detailed diagram of the method according to the present disclosure,

[0048] Figure 5 a flowchart illustrating a method for determining control parameters for driving a host vehicle according to various embodiments,

[0049] Figure 6 a diagram of a system according to various embodiments, and

[0050] Figure 7 a computer system having a plurality of computer hardware components configured to perform the steps of the computer-implemented methods described herein. DETAILED DESCRIPTION

[0051] Figure 1 FIG. depicts a schematic illustration of a vehicle 100 and objects that may surround the vehicle 100 in a traffic scenario. The vehicle 100 includes a perception system 110 having an instrument field of view indicated by line 115. The vehicle 100 further includes a computer system 120 that includes a processing unit 121 and a data storage system 122, which includes, for example, a memory and a database. The processing unit 121 is configured to receive data from the perception system 110 and store the data in the data storage system 122. The vehicle 100 further includes a control system 124 configured to control the actual movement of the vehicle 100. The control system 124 is connected to the computer system 120 via vehicle interfaces 125, 350 (see also Figure 3 ).

[0052] The perception system 110 may include a radar system, a lidar system, and / or one or more cameras to monitor the external environment or the surrounding environment of the vehicle 100. Accordingly, the perception system 110 is configured to monitor the dynamic situation 125 of the vehicle 100, and the dynamic situation 125 includes a plurality of road users 130 capable of moving in the external environment of the vehicle 100. The road users 130 may include, for example, other vehicles 140 and / or pedestrians 150.

[0053] The perception system 110 is further configured to monitor the static situation 160 of the vehicle 100. The static situation 160 may include, for example, traffic signs 170 and lane markings 180. Accordingly, the perception system 110 is configured to determine the static and dynamic characteristics of the road users 130. The perception system 100 determines the dynamic and static characteristics of the road users 130 over a predetermined number of time steps (e.g., 0.5 seconds per step).

[0054] The dynamic characteristics include the current position, current speed, and object category of each road user 130. The current position and current speed are determined by the perception system 110 relative to the vehicle 100, that is, relative to a coordinate system having its origin (e.g., the origin is located at the centroid of the vehicle 100, its x-axis is along the longitudinal direction of the vehicle 100, and its y-axis is along the lateral direction of the vehicle 100). Similarly, the static characteristics include the position, size, and object category of objects such as traffic signs and lane markings 170, 180 that constitute the static situation 160.

[0055] The computer system 120 transmits the results or outputs (i.e., the control parameters 460 (see Figure 3 )) of the method according to the present disclosure to the control system 124 via the vehicle interfaces 125, 350 (also see Figure 4 ) so that the control system 124 can reliably control the future movement of the vehicle 100.

[0056] Figure 2 Details of the processing unit 121 included in the computer system 120 of the vehicle 100 are depicted (see Figure 1 ). The processing unit 121 includes a perception system for the method according to the present disclosure, wherein the system is based on a deep neural network 210. The perception step of the method is performed via the perception system 110 of the vehicle 100 (see Figure 1 ) and provides different inputs related to the movement of the vehicle 100 to the deep neural network 210. Based on these inputs, the deep neural network 210 performs the prediction step 320 of the method (also see Figure 3 and Figure 4 ).

[0057] The input to the deep neural network 210 includes the dynamic situation 125 (i.e., the dynamic characteristics of the road user 130), the static situation 160, and the ego-dynamics 220 of the vehicle 100. The deep neural network 210 is used to generate a prediction output 230. When training the deep neural network 210, the prediction output 230 and the ground truth (GT) 240 are provided to the loss function 250 for optimizing the deep neural network 210.

[0058] The static situation 160 includes static environment data, which includes the respective positions, respective dimensions, and specific categories of static entities in the environment of the vehicle 100, e.g., the positions and dimensions of traffic signs 170 and lane markings 180. As described above, the static situation 160 (i.e., the static environment data of the vehicle 100) is determined via the perception system 110 of the vehicle 100 and additionally or alternatively derived from a pre-defined map available for the surrounding environment of the vehicle 100.

[0059] The static situation 160 is represented by one or more of the following:

[0060] - A rasterized image from an HD (high-definition) map, where the high-definition map covers the accurate positions of, e.g., lane markings, such that when the vehicle can be accurately located in the HD map, accurate information about its surroundings is provided to the vehicle.

[0061] - A drivable area determined via the perception system 110 of the vehicle 100, such as a grid map or an image data structure, where each pixel of such a map or image represents the drivability of a specific area in the instrument field of view of the perception system 110.

[0062] - Lane / road detection via the sensors of the perception system 110, where using the detected lane markings, road boundaries, guardrails, etc. from the sensors, the perception system 110 can be configured to construct a grid map or similar image-like data similar to a rasterized map for describing the static situation 160.

[0063] - A static occupancy grid map.

[0064] The ego-dynamics 220 can also be represented as one of the road users 130 among the road users 130 and can thus be included in the dynamic situation input.

[0065] As will be described in detail below, the deep neural network 210 includes an encoder and different decoder heads that are used to provide a prediction output 230 in the form of a selected or most suitable driving hypothesis, which includes a drivable path or a driving path prediction and a trajectory prediction, i.e., a trajectory distribution associated with the drivable path such that the trajectory distribution matches the drivable path. For the selected driving hypothesis, the most suitable trajectory is determined. For this trajectory, control parameters for driving the vehicle 100 are determined (see, for example, Figure 4 step 446 in

[0066] The ground truth 240 defines the task of the deep neural network 210. It covers information such as about the position as occupancy probability and offset within a grid, as well as other attributes such as speed and acceleration, and / or other regression and classification tasks, such as the future position, speed, maneuvers, etc. of the road users 130 being monitored within the current traffic scene.

[0067] Figure 3 Depicts an overview of the most important steps and items related to the method according to the present disclosure. The method includes the perception step 310 as described above. The data provided by the perception step 310 is used by the step 320 of driving hypothesis generation, which is also described above Figure 2 in the context and Figure 4 is described in more detail below

[0068] The driving hypotheses generated by the step 320 of driving hypothesis generation are evaluated in the next step 330 of driving hypothesis selection to determine or select the most suitable driving hypothesis from multiple driving hypotheses.

[0069] As described above, each driving hypothesis includes a drivable path of the host vehicle 100 and a trajectory distribution. The drivable path refers to an area or free space where the host vehicle 100 can travel in the future without colliding with static items in the external environment of the host vehicle 100, for example. In contrast, each trajectory associated with the trajectory distribution is also defined in time. That is, each trajectory includes an associated planning time horizon. Each trajectory should not have any conflicts with other road users within the planning or future time horizon associated with the corresponding trajectory. A typical representation of a trajectory is a series of points associated with the corresponding time steps. For the trajectory distribution associated with the corresponding driving hypothesis, a corresponding plurality of trajectories are determined.

[0070] When selecting the most suitable driving hypothesis in step 330, the most suitable trajectory is determined for the trajectory distribution associated with the most suitable driving hypothesis. This is performed by generating and evaluating a cost function to evaluate the trajectories against different criteria. Based on the most suitable trajectory, control parameters for driving the main vehicle 100 are determined. The control parameters include, for example, the acceleration and yaw rate of the main vehicle and may further include the steering wheel angle and / or front wheel angle of the main vehicle 100. In turn, the control parameters describe or define the most suitable trajectory for driving the main vehicle 100.

[0071] In a further control step 340, the control parameters for driving the main vehicle 100 are further refined by considering external conditions including the traffic scene and / or static objects around the main vehicle 100. The external conditions may also be provided as a model and / or rules 360. Additionally, a safety check may be performed on the refined control parameters to avoid conflicts with safety conditions also provided by the model and / or rules 360. The output of control step 340 includes the final control parameters, which are transmitted to the vehicle interface 350. That is, the final control parameters are received by further modules of the main vehicle 100 to provide a basis for driving the main vehicle 100.

[0072] Figure 4 Schematically depicts details of the method according to the present disclosure other than Figure 3 the main steps 310 to 340 shown. In addition to the perception 310 of the static and dynamic characteristics detected in the external environment of the main vehicle 100, data 410 related to the self-motion of the main vehicle 100 and data 415 related to the above-mentioned high-definition map (HD map) and the navigation of the main vehicle 100 are also provided as inputs to the deep neural network 210 (see also Figure 2 ), which includes a CASPNet encoder 420 as a first part of the driving hypothesis generation 320. The CASPNet encoder 420 is a machine learning algorithm, such as described in the article "Context-aware scene production network (CASPNet)" by Schaefer, M. et al. on arXiv:2201.06933v1 on January 18, 2022. However, other methods and algorithms may also be used as the encoder for driving hypothesis generation 320.

[0073] The driving hypothesis generation 320 further includes three different decoders, namely, the ego grid decoder 430, the scenario grid decoder 432, and the target trajectory decoder 434. The CASPNet encoder 420 or another system acting as an encoder provides encoded grid maps, which may also be referred to as feature maps, and these grid maps can provide the probability distribution of the positions of the road users 130 (see Figure 1 ) and other dynamic attributes over multiple time steps. These maps also include information about the static environment of the host vehicle 100 in encoded form. After decoding, each cell of such a grid map includes different channels, which include the probability of the position of the road user 130 at a specific time step, as well as additional channels for different dynamic attributes of the road user 130 (including the host vehicle or ego vehicle 100).

[0074] The decoders 430, 432, 434 receive the feature maps provided by the encoder 420 in order to track or decode specific information via the corresponding output heads 431, 433, 435. The ego grid decoder 430 is configured to provide N driving hypotheses for the host vehicle or ego vehicle 100 via N output heads 431 (i.e., output head 1 to output head N). Each driving hypothesis provided by the corresponding output head includes a corresponding driving path prediction for the drivable path and a corresponding trajectory prediction (i.e., trajectory distribution) for the host vehicle or ego vehicle 100. The corresponding trajectory distribution includes a specific distribution of possible positions or points for the corresponding time step.

[0075] By processing the output of the encoder 420 in a different way, the scenario grid decoder 432 is configured to provide predictions based on the main grid, vehicle grid, and pedestrian grid for the traffic scenario in which the host vehicle 100 is currently located. Compared with the driving hypotheses that are only related to the host vehicle 100 (including the corresponding driving path or free space of the host vehicle and information about the possible trajectory), the output 433 of the scenario grid decoder 432 is related to the entire traffic scenario around the host vehicle 100, i.e., the entire dynamic situation 125 schematically depicted in Figure 1 as shown.

[0076] In addition, the target trajectory decoder 434 is configured to provide a target trajectory 435, which may conflict with the trajectory prediction provided by the ego grid decoder 430.

[0077] The entire output of the driving hypothesis generation 320, i.e., the outputs 431, 433, 435 of the three decoders 430, 432, 434, is transmitted to the driving hypothesis selection 330 step. To select the most suitable driving hypothesis, a set of cost functions is generated and evaluated at 440. The cost functions are associated with each of the trajectory distributions provided by the output head 431 of the ego grid controller 430 and include, for example, a physical cost function, a safety cost function, a rule cost function, and a comfort cost function. These cost functions are generated for the trajectory distribution predictions provided by the ego grid decoder 430 by considering different aspects of physical conditions, safety conditions, rules, and comfort conditions. In addition, the output of the scenario grid decoder 432 is considered when estimating the cost functions at 440. Thus, the predictions performed for the traffic scenario, i.e., for the dynamic situation of the host vehicle 100, affect the specific values of the cost functions determined at 440.

[0078] The cost functions determined at 440 provide a first criterion for the driving path selection at 442. That is, by applying the cost functions determined at 440, the corresponding driving path predictions are selected, and thus the entire driving hypothesis including the driving path prediction and the trajectory distribution prediction is selected.

[0079] When selecting the most suitable driving hypothesis, there is a hierarchy or ranking among the different cost functions mentioned above. The highest priority or relevance is associated with the physical cost function because the physical conditions for the movement of the host vehicle 100 cannot be changed. The safety cost function is thus associated with a lower priority than the physical cost function. Similarly, the rule cost function is associated with a lower priority than the safety cost function because safety is more important than following rules. In addition, the comfort cost function is associated with a lower priority than the rule cost function, i.e., with the lowest priority.

[0080] As a second criterion for determining the most suitable driving hypothesis and driving path at 442, the driving path of each driving hypothesis is related to the navigation information provided for the host vehicle 100 at, for example, 415.

[0081] At 444, a planning time range is determined, which is used when generating multiple trajectories for the trajectory distribution associated with the selected or most suitable driving hypothesis. For the planning time range determined at 444, the most suitable trajectory and the control parameters defining the trajectory are determined at 446. Specifically, the above cost functions are applied again at 440 to select the most suitable trajectory from the multiple trajectories generated for the trajectory distribution associated with the selected driving hypothesis. In addition, the data 410 related to self-motion is considered at the step 466 of trajectory and control parameter determination.

[0082] For each of a plurality of trajectories belonging to the trajectory distribution provided by the output head 431, at least one control point is determined or sampled according to the planned time range determined at 444. The drivable path selected at 442 and the corresponding sampled control points are used by a machine learning algorithm that determines control parameters associated with the most suitable trajectory. The control parameters include the acceleration and yaw rate of the main vehicle 100. Additionally, the control parameters may also include the steering wheel angle or front wheel angle of the main vehicle 100. Further details of determining the control parameters by applying the machine learning algorithm are provided, for example, in EP 4 245 629 A1.

[0083] The output of step 330 of driving hypothesis selection is transmitted to control step 340, i.e., to control signal evaluation. Control step 340 includes a refinement step 450 for the control parameters and a safety check 452 applied to the refined control parameters. At 450, the control parameters are refined with respect to the static characteristics or static situation 160 of the main vehicle 100 (see Figure 1 ), with respect to the previously planned trajectory, and with respect to predefined rules. Specifically, the semantic segmentation of the static situation 160 (including information related to, for example, lane markings, static objects, etc.) is used as the input for refinement step 450 of the control parameters.

[0084] Safety check 452 depends at least in part on the target trajectory 435 provided by the target trajectory decoder 434. If the safety check 452 fails, i.e., if the control parameters refined at 450 do not meet the safety conditions related to the main vehicle 100, the method returns to step 330 of driving hypothesis selection to determine a new set of reliable control parameters at 446. If the safety check 452 is passed, the final control parameters are output at 460 according to the method of the present disclosure.

[0085] The final control parameters provided at 460 are transmitted via the vehicle interface 350 (see Figure 3 ) to the control system 124 of the main vehicle 100 (see Figure 1 ). As an optional step, the control system 124 is configured to verify the received control parameters, i.e., to perform additional safety checks in addition to, for example, Figure 4 the safety check 452 shown. After successfully verifying the received control parameters, the control system 124 applies these received control parameters during the operation of the main vehicle 100.

[0086] Figure 5FIG. 500 is shown, which illustrates a method for determining control parameters for driving a host vehicle. At 502, static and dynamic characteristics detected in the external environment of the host vehicle can be determined via a sensing system of the host vehicle. At 504, a plurality of driving hypotheses can be determined via a processing unit of the host vehicle by applying the static and dynamic characteristics, where each driving hypothesis can include a drivable path of the host vehicle and a trajectory distribution matching the drivable path. At 506, the most suitable driving hypothesis can be selected from the plurality of driving hypotheses via the processing unit. At 508, the most suitable trajectory can be determined via the processing unit for the trajectory distribution associated with the most suitable driving hypothesis. At 510, control parameters for driving the host vehicle can be determined based on the most suitable trajectory.

[0087] According to various embodiments, for each driving hypothesis, a corresponding set of cost functions can be determined with respect to each trajectory distribution in the trajectory distribution, and the most suitable driving hypothesis can be selected from the plurality of driving hypotheses by utilizing the corresponding set of cost functions and by associating the drivable path of each driving hypothesis with navigation information for the host vehicle.

[0088] According to various embodiments, a prediction of a traffic scene within the environment of the host vehicle can be determined by using the static and dynamic characteristics detected in the external environment of the host vehicle, and the predicted traffic scene can be used to evaluate the set of cost functions.

[0089] According to various embodiments, the cost function can include a physical cost function, which can be related to physical conditions of the driving path and the trajectory distribution for the corresponding driving hypothesis, the cost function can further include a safety cost function, which can be related to safety conditions of the driving path and the trajectory distribution for the corresponding driving hypothesis, and the safety cost function can be associated with a lower priority than the physical cost function.

[0090] According to various embodiments, the cost function can further include a rule cost function, which can be related to rules of the driving path and the trajectory distribution for the corresponding driving hypothesis, and the rule cost function can be associated with a lower priority than the safety cost function.

[0091] According to various embodiments, the cost function can further include a comfort cost function, which can be related to the comfort of passengers of the host vehicle with respect to the driving path and the trajectory distribution of the corresponding driving hypothesis, and the comfort cost function can be associated with a lower priority than the rule cost function.

[0092] According to various embodiments, a plurality of trajectories can be generated for a trajectory distribution associated with a selected driving hypothesis, a set of cost functions associated with the most suitable driving hypothesis can be applied to the plurality of trajectories, and the most suitable trajectory can be selected from the plurality of trajectories by evaluating the cost functions.

[0093] According to various embodiments, a planning time horizon can be determined for a plurality of trajectories generated for a trajectory distribution associated with the most suitable driving hypothesis.

[0094] According to various embodiments, at least one control point can be determined for each of the plurality of trajectories according to the planning time horizon, and the drivable path and the corresponding control points can be used by an algorithm that can determine control parameters associated with the most suitable trajectory.

[0095] According to various embodiments, the control parameters can be refined relative to static characteristics determined for the external environment of the host vehicle, relative to a previously planned trajectory, and / or relative to predefined rules.

[0096] According to various embodiments, predictions for traffic scenarios within the environment of the host vehicle and predictions for target trajectories relative to the most suitable trajectory can be determined by using static and dynamic characteristics detected in the external environment of the host vehicle, and the control parameters can be checked for safety by applying the predictions for the traffic scenarios and the predictions for the target trajectories.

[0097] Each of steps 502, 504, 506, 508, and 510 and the further steps described above can be performed by computer hardware components.

[0098] Figure 6 A control parameter determination system 600 according to various embodiments is shown. The control parameter determination system 600 can include a characteristic determination circuit 602, a driving hypothesis determination circuit 604, a driving hypothesis selection circuit 606, a trajectory determination circuit 608, and a control parameter determination circuit 610.

[0099] The feature determination circuit 602 can be configured to: determine static and dynamic features detected in the external environment of the host vehicle via the sensing system of the host vehicle. The driving hypothesis determination circuit 604 can be configured to: determine a plurality of driving hypotheses by applying the static and dynamic features, each driving hypothesis including a drivable path and a trajectory distribution of the host vehicle. The driving hypothesis selection circuit 606 can be configured to: select the most suitable driving hypothesis from the plurality of driving hypotheses. The trajectory determination circuit 608 can be configured to: determine the most suitable trajectory for the trajectory distribution associated with the most suitable driving hypothesis. The control parameter determination circuit 610 can be configured to: determine control parameters for driving the host vehicle based on the most suitable trajectory.

[0100] The feature determination circuit 602, the driving hypothesis determination circuit 604, the driving hypothesis selection circuit 606, the trajectory determination circuit 608, and the control parameter determination circuit 610 can be coupled to each other, for example, via an electrical connection 612 (such as a cable or a computer bus) or via any other suitable electrical connection to exchange electrical signals.

[0101] A "circuit" can be understood as any type of logic implementation entity, which can be a dedicated circuit or a processor that executes a program stored in a memory, firmware, or any combination thereof.

[0102] Figure 7 A computer system 700 having a plurality of computer hardware components is shown, the plurality of computer hardware components being configured to perform the steps of a computer-implemented method for determining control parameters for driving a host vehicle according to various embodiments. The computer system 700 can include a processor 702, a memory 704, and a non-transitory data storage 706.

[0103] The processor 702 can execute instructions provided in the memory 704. The non-transitory data storage 706 can store a computer program that includes instructions that can be transferred to the memory 704 and then executed by the processor 702.

[0104] The processor 702, the memory 704, and the non-transitory data storage 706 can be coupled to each other, for example, via an electrical connection 708 (such as a cable or a computer bus) or via any other suitable electrical connection to exchange electrical signals.

[0105] In this way, the processor 702, the memory 704, and the non-transitory data storage 706 can represent the feature determination circuit 602, the driving hypothesis determination circuit 604, the driving hypothesis selection circuit 606, the trajectory determination circuit 608, and the control parameter determination circuit 610 described above.

[0106] The terms "coupled" or "connected" are respectively intended to include "directly coupled" (e.g., via a physical link) or "directly connected" and "indirectly coupled" or "indirectly connected" (e.g., via a logical link).

[0107] It should be understood that what is described for one of the above methods can be similarly applied to the control parameter determination system 600 and / or the computer system 700.

[0108] List of reference numerals

[0109] 100 Vehicle

[0110] 110 Sensing system

[0111] 115 Field of view

[0112] 120 Computer system

[0113] 121 Processing unit

[0114] 122 Memory, database

[0115] 124 Control system

[0116] 125 Dynamic situation

[0117] 130 Road user

[0118] 140 Vehicle

[0119] 150 Pedestrian

[0120] 160 Static situation

[0121] 170 Traffic sign

[0122] 180 Lane marking

[0123] 210 Prediction system, deep neural network

[0124] 220 Self-dynamics of the host vehicle

[0125] 230 Prediction output

[0126] 240 Ground truth

[0127] 250 Loss function

[0128] 310 Sensing

[0129] 320 Driving hypothesis generation

[0130] 330 Driving hypothesis selection

[0131] 340 Control

[0132] 350 Vehicle Interface

[0133] 360 Models and Rules

[0134] 410 Data Related to the Self-Motion of the Vehicle

[0135] 415 High-Definition Map and Navigation Information

[0136] 420 Encoder

[0137] 430 Self-Grid Decoder

[0138] 431 Output Head for Providing Driving Hypotheses

[0139] 432 Scene Grid Decoder

[0140] 433 Grid-Based Prediction for Traffic Scenes

[0141] 434 Target Trajectory Decoder

[0142] 435 Target Trajectory

[0143] 440 Cost Function Evaluation and Trajectory Selection

[0144] 442 Driving Path Selection

[0145] 444 Planning Range Determination

[0146] 446 Trajectory and Control Parameter Determination

[0147] 450 Control Parameter Refinement

[0148] 452 Safety Check

[0149] 460 Final Control Parameters

[0150] 500 Flowchart Illustrating a Method for Determining Control Parameters for Driving a Host Vehicle

[0151] 502 Step: Determine the static and dynamic characteristics detected in the external environment of the host vehicle via the perception system of the host vehicle

[0152] 504 Step: Determine multiple driving hypotheses by applying the static and dynamic characteristics, each driving hypothesis including a drivable path of the host vehicle and a trajectory distribution matching the drivable path

[0153] 506 Step: Select the most suitable driving hypothesis from the multiple driving hypotheses

[0154] 508 Step: Determine the most suitable trajectory for the trajectory distribution associated with the most suitable driving hypothesis

[0155] 510 Step: Determine the control parameters for driving the main vehicle according to the most suitable trajectory

[0156] 600 Control parameter determination system

[0157] 602 Characteristic determination circuit

[0158] 604 Driving hypothesis determination circuit

[0159] 606 Driving hypothesis selection circuit

[0160] 608 Trajectory determination circuit

[0161] 610 Control parameter determination circuit

[0162] 612 Connection

[0163] 700 Computer system

[0164] 702 Processor

[0165] 704 Memory

[0166] 706 Non-transitory data storage

[0167] 708 Connection

Claims

1. A computer-implemented method for determining control parameters (460) for driving a host vehicle (100), the method comprising: determining, via a perception system (110) of the host vehicle (100), static and dynamic characteristics detected in an external environment of the host vehicle (100), The following operations are performed via the processing unit (121) of the host vehicle: Determining a plurality of driving hypotheses by applying the static characteristics and the dynamic characteristics, each driving hypothesis comprising a drivable path of the host vehicle (100) and a trajectory distribution matching the drivable path, selecting the most appropriate driving hypothesis from the plurality of driving hypotheses, determining a most appropriate trajectory for the trajectory distribution associated with the most appropriate driving hypothesis, The control parameters (460) for driving the host vehicle (100) are determined based on the most suitable trajectory.

2. The method according to claim 1, characterized in that: For each driving hypothesis, a corresponding set of cost functions is determined with respect to each of the trajectory distributions, and The most appropriate driving hypothesis is selected from the plurality of driving hypotheses by utilizing the corresponding set of cost functions and by associating the drivable path of each driving hypothesis with navigation information (415) for the host vehicle (100).

3. The method according to claim 2, characterized in that: determining a prediction (433) of a traffic scene within the environment of the host vehicle (100) by using the static characteristics and the dynamic characteristics detected in the external environment of the host vehicle (100), The predicted traffic scenarios (433) are used to evaluate the set of cost functions.

4. The method according to claim 2, characterized in that: The cost function comprises a physical cost function related to physical conditions of the driving path and the trajectory distribution for a corresponding driving hypothesis, The cost function includes a safety cost function associated with a safety condition of the driving path and the trajectory distribution for the corresponding driving hypothesis, and The safety cost function is associated with a lower priority than the physical cost function.

5. The method according to claim 4, characterized in that: The cost function further includes a rule cost function associated with a rule of the driving path and the trajectory distribution for the corresponding driving hypothesis, and The rule cost function is associated with a lower priority than the security cost function.

6. The method according to claim 5, characterized in that: The cost function further comprises a comfort cost function related to the comfort level of the passengers of the host vehicle (100) relative to the driving path and the trajectory distribution of the corresponding driving hypothesis, and The comfort cost function is associated with a lower priority than the rule cost function.

7. The method according to any one of claims 2 to 6, characterized in that: generating a plurality of trajectories for the trajectory distribution associated with the selected driving hypothesis, The set of cost functions associated with the most appropriate driving hypothesis is applied to the plurality of trajectories, and The most suitable trajectory is selected from the plurality of trajectories by evaluating the cost function.

8. The method according to claim 7, characterized in that: A planning time horizon is determined for the plurality of trajectories generated for the trajectory distribution associated with the most appropriate driving hypothesis.

9. The method according to claim 8, characterized in that: Determining at least one control point for each of the plurality of trajectories according to the planning time horizon, and The drivable paths and corresponding control points are used by an algorithm that determines the control parameters associated with the most appropriate trajectory (460).

10. The method according to any one of claims 1 to 6, characterized in that: The control parameters (460) are refined relative to the static characteristics determined for the external environment of the host vehicle (100), relative to a previously planned trajectory and / or relative to predefined rules.

11. The method according to any one of claims 1 to 6, characterized in that: determining a prediction (433) for a traffic scene within the environment of the host vehicle and a prediction for a target trajectory (435) relative to the most suitable trajectory by using the static characteristics and the dynamic characteristics detected in the external environment of the host vehicle (100), and The control parameters (460) are checked for safety by applying the prediction (433) for the traffic scenario and the prediction for the target trajectory (435).

12. A computer system (120, 700), the computer system (120, 700) being configured to: receiving, via a perception system (110) of a host vehicle (100), static and dynamic characteristics detected in an external environment of the host vehicle (100), and Perform the computer-implemented method of at least one of claims 1 to 11.

13. A vehicle (100), comprising: Perception system (110), The computer system (120, 700) of claim 12, and A control system (124) configured to receive a control parameter (460) determined by the computer system (120, 700) by executing the method of at least one of claims 1 to 11.

14. The vehicle (100) according to claim 13, characterized in that: The control system (124) is also configured to validate the received control parameters (460) and to apply the received control parameters (460) during operation of the vehicle (100).

15. A non-transitory computer-readable medium comprising instructions for executing the computer-implemented method of at least one of claims 1 to 11.

Images