Criticality determination for autonomous driving

Abstract

The invention relates to a control unit (22) which is configured to control an object function (O) of a vehicle (10) ego ) and one or more object functions (O) of other static or dynamic objects or other traffic participants in the surrounding environment of the vehicle (10) n ) is used to calculate the criticality of the driving situation (I cri The present invention also relates to an object function (O) for a vehicle (10) ego ) and one or more object functions (O) of other static or dynamic objects or other traffic participants in the surrounding environment of the vehicle (10) n ) is used to calculate the criticality of the driving situation (I cri ) method.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of vehicle sensors and their data evaluation, in particular for automated or semi-automated driving vehicles. Furthermore, the present disclosure also relates to the technical field of risk evaluation of driving situations in the context of so-called "Motion Planning". BACKGROUND

[0002] Automated or semi-automated driving vehicles have sensors, such as cameras, radar and lidar sensors, which sense and identify the surroundings of the vehicle and whose data are evaluated in a control unit by means of suitable software. Based on the information obtained by this data processing, the control unit can automatically trigger and execute brake regulation, speed regulation, distance regulation, compensation and / or evasion regulation by means of corresponding actuators.

[0003] Precise surroundings recognition is of great importance for high-performance driver assistance systems and automated driving vehicles. For this purpose, modern vehicles are equipped with a large number of sensors, such as radar, lidar or camera sensors, which send their measurement values in the form of point clouds. From the point clouds provided by the sensors, reliable information about possible objects in the driving path or in collision with the own vehicle can be obtained. Furthermore, it is also important for automated driving vehicles to be able to assess the accident risk of a driving situation.

[0004] Another known method is to use TTC ("Time to collision") or THW ("Headway") as a measure of the criticality of a driving situation, as described by Saffarzadeh et al. in "A general formulation for time-to-collision safety indicator", Transport, Vol. 166, TRS Issue. TTC is an important time-based safety indicator for identifying rear-end accidents in traffic safety assessment. Saffarzadeh et al. include the linear acceleration of the object in the TTC calculation. However, the use of TTC to determine criticality is disadvantageous because different or changing speeds of the own vehicle lead to false estimates of criticality.

[0005] C. Schmidt in "Fahrstrategien zur Unfallvermeidung im StraBenverkehr fur Einzel und Mehrobjektszenarien (Driving strategies for accident avoidance in road traffic for single and multiple object scenarios)", KIT Scientific Publishing, Karlsruhe, ISBN: 978-3-7315-0198-5 describes a "Motion Planning" approach. Here, a last possible maneuver is first derived for a single object in combination with the achievable standstill circle and is extended to a full evasion maneuver. Next, the scope is extended to an arbitrary number of objects and collision-free trajectories are determined and evaluated in combination with the standstill zone and the kinematic configuration. From an exact description of the lateral vehicle motion possibilities a so-called "passing gate" can be assumed through which the ego vehicle has to pass to avoid any collision with the relevant obstacles.

[0006] In "Motion Planning" obstacles are for example also described as mathematical functions, also called "Potential Fields".

[0007] In D. Reichardt and J. Shick, "Collision avoidance in dynamic environments applied to autonomous vehicle guidance on the motorway", Proceedings of the Intelligent Vehicles'94 Symposium, DOI: 10.1109 / IVS.1994.639475, a method for autonomous vehicle navigation on a motorway to avoid collisions is discussed. Here, the environment information is provided by multiple vision sensor modules and stored in a central dynamic database. Through data fusion and data interpretation a systematic view of the environment is generated based on the data stored in the dynamic database representing the current scene. This systematic view is transformed into a risk map view which integrates information about the relative position and speed of roads, obstacles and traffic signs.

[0008] Michael T. Wolf and Joel W. Burdick in "Artificial potential functions for highway driving with collision avoidance", IEEE International Conference on Robotics and Automation, 2008, DOI: 10.1109 / ROBOT.2008.4543783 propose a series of possible functional components that can provide support for automated or semi-automated vehicles in navigating situations on multi-lane, dense highways. The resulting potential field is constructed as a superposition of different functions for keeping the lane, staying on the road, predefining the speed, and avoiding and overtaking. The construction of the vehicle avoidance potential is the most important one, wherein the structure and protocol of driving on a multi-lane highway are taken into account. In particular, the shape and size of the potential field behind each obstacle vehicle can be adapted to the speed of the vehicle and the surrounding traffic to appropriately facilitate deceleration and / or overtaking of the controlled vehicle. Hard obstacles at the lane edges and soft boundaries between lanes keep the vehicle on the highway, preferably driving in the center of the lane.

[0009] However, the techniques of "motion planning" working with potential fields are often complex to implement. It is the task of the present invention to improve such techniques, in particular to make them realizable in a resource-efficient manner. SUMMARY

[0010] The present invention provides a control unit as defined in claim 1. Furthermore, the present invention provides a vehicle as defined in claim 13, and a method as defined in claim 15.

[0011] The control unit according to the present invention is set up to calculate the criticality of a driving situation in the form of an overlap integral of an objective function of the own vehicle with one or more objective functions of other static or dynamic objects or other traffic participants of the surrounding environment of the own vehicle.

[0012] The driving situation can be predetermined by potential positions of the own vehicle and by objective parameters and surrounding environment parameters of other static or dynamic objects or traffic participants of the surrounding environment of the own vehicle.

[0013] The idea of the present invention is thus not only to construct the environment of the vehicle, but to incorporate the movable objects (cars, buses, trucks, people, etc.) as mathematical "functions" into the mathematical environment.

[0014] The control unit according to the application can be, for example, a control device (electronic control unit, ECU, or Electronic Control Module, ECM) or a processor. In principle, any device that can receive electrical signals and process them on the basis of software or hardware can be envisaged as a control unit. The surroundings sensor can be, for example, a radar sensor, a lidar sensor, a camera sensor, an ultrasonic or infrared sensor. The objects in the field of view of the surroundings sensor can be all objects, animals and people that the vehicle can encounter in the driving situation, for example pedestrians, cyclists, dogs, other cars, but also road signs, barriers, walls, buildings, litter bins, etc.

[0015] The criticality here is a quantitative measure that allows the risk of an accident with all other traffic participants at a particular location to be estimated. This location-dependent criticality can be entered into a criticality map and used, for example, for trajectory planning. This allows trajectory planning that determines a trajectory with the lowest accident risk in consideration of the entire detectable surroundings situation.

[0016] The challenge to be solved is to depict a vehicle in a "artificial" environment.

[0017] In an embodiment of the application, the objective function is selected as a function of the function equation:

[0018]

[0019] where i = 1, 2, 3; is a normalization constant; σ = (σ1, σ2, σ3) is the standard deviation of the objective function, x0= (x 1,0 ,x 2,0 ,x 3,0 ) is the location of the object.

[0020] The objective function can also be selected as follows, i.e. different standard deviations σ v or σ h are selected for the driving direction (forward) and the reverse direction (backward). For example, the standard deviation σ v can be selected in dependence on parameters (in dependence on driving parameters, such as speed, etc.), while at the same time the standard deviation σ h is selected (independently of parameters) constantly (a predetermined fixed value, which corresponds, for example, to the dimensions of the vehicle). The standard deviation can also be selected in dependence on the angle between the driving direction vector v and the vector x - x0, so that the standard deviation σ v varies continuously from σ h (backward) in dependence on the angle.

[0021] Here, the normalization constant can be chosen such that the value of the criticality is exactly 1 in the case of a bumper contact with the edge of the object. The function equation chosen here is a three-dimensional Gaussian distribution in the Cartesian coordinate system. However, the application is not limited to the Cartesian coordinate system. The function equation used has the property that its convolution can be determined simply using the same type of function, which greatly reduces the computational effort for determining the criticality. Furthermore, the convolution of two Gaussian functions is known in almost all cases, which makes a basic determination of the criticality possible.

[0022] In an embodiment of the application, the Gaussian distribution is determined by the standard deviation σ, wherein the standard deviation σ is a function of the object parameters, such as speed or acceleration, and / or of the surrounding environment parameters, such as outside temperature, road surface.

[0023] The speed of the object within the field of view can be determined, for example, by evaluating radar or lidar data with the aid of the Doppler effect. The acceleration of the object can be determined, for example, by looking at its speed intermittently. The outside temperature can be determined, for example, with the aid of a thermometer. The road surface can be determined, for example, with the aid of a sound sensor. The visibility and the sun's position, as well as other parameters such as the fog density, can be determined, for example, by evaluating camera data. Taking into account as many factors as possible and thus different environmental data can enable precise and error-free and interference-free trajectory planning.

[0024] The control unit according to the application can be set up to determine the standard deviation σ on the basis of the estimated braking distance s of the object.

[0025] Determining the standard deviation of the object function from the desired braking distance can enable the determination of an object function that contains a danger zone around the object. This facilitates subsequent trajectory planning, which can then plan a trajectory around the braking distance of the recognized object.

[0026] In another variant of the application, the object function of the control unit is determined by the standard deviation σ, wherein the control unit is set up to determine the standard deviation by means of a polynomial model:

[0027]

[0028] where v denotes the object speed, and wherein the individual parameters a j (acceleration) and c ext (environmental parameter) are determined from a known braking distance data set by means of a multiple regression or by means of a neural network.

[0029] The advantage is that the standard deviation of the object function can be determined individually for different driving situations (fog, wet, wet... ) and different objects (static, fast, slow... ). This allows the control unit according to the application to determine the criticality of the driving situation appropriately for very different weather and traffic situations.

[0030] The control unit can also be set up to determine the criticality of the driving situation (I cri )

[0031]

[0032] where n denotes the recognized object, N represents the total number of all recognized objects, O ego represents the object function of the own vehicle, O n represents the object function of the recognized object.

[0033] Instead of the sum of the criticalities with regard to the different objects, alternatively also the maximum of the overlap integral ego • O n dxdydz can be used as the criticality:

[0034] I cri = max n＝1…N (∫∫∫O ego • O n dxdydz))

[0035] It is appropriate to calculate the criticality in the form of the sum of the overlap integrals between the object function of the own vehicle and the object functions, since the criticality thereby describes the overlap of the danger zones and thus describes the collision probability.

[0036] The criticality I cri can be normalized such that, in the case of contact of the own vehicle with another object, the criticality is equal to 1. Checking the criticality for the case of bumper to bumper contact to the value 1 allows a calibration of the criticality. Furthermore, setting the criticality to 1 for the case of bumper to bumper contact allows that a criticality value > 1 can be used to estimate the severity of an accident.

[0037] Determining the location-dependent criticality values and representing them in a criticality map allows to estimate the danger that a location in the surroundings of the vehicle poses to the vehicle. This enables the trajectory planning to minimize the risk of an accident during the planning process.

[0038] The control unit can here be set up to communicate with the vehicle via the Internet, wherein sensor data are received from the vehicle and the criticality is transmitted to the vehicle.

[0039] The control unit can be installed in a vehicle with a surrounding environment sensor. The vehicle can be set up such that the control unit communicates with the surrounding environment sensor, the sensor data of the surrounding environment sensor is transmitted to the control unit and the criticality is received from the control unit.

[0040] Such an inventive vehicle can as described assess dangerous situations quantitatively with respect to its current driving and traffic situation. BRIEF DESCRIPTION OF DRAWINGS

[0041] The application is exemplarily explained in the following with embodiments shown in the drawings.

[0042] Figure 1 A block diagram is shown which schematically shows the configuration of a vehicle 10 according to one embodiment of the application;

[0043] Fig. 2 schematically shows an exemplary configuration of a control unit for automated driving;

[0044] Figure 3 An exemplary surrounding environment sensor 26 is shown, here in particular a radar sensor;

[0045] Figure 4a , 4b Fig. 4c shows a two-dimensional grid map obtained from the detection events of the radar sensor;

[0046] Figure 5 A contour plot of an exemplary object function is shown which is defined by a Gaussian function with a given standard deviation;

[0047] Figure 6 An exemplary criticality map is shown which can for example be the result of the inventive method;

[0048] Figure 7 The steps of a method for determining a criticality according to the application are schematically shown in a flow chart, which for example is applied to a control unit according to the application. DETAILED DESCRIPTION

[0049] Embodiments of the application are explained in the following with reference to the drawings.

[0050] Figure 1A block diagram is shown which schematically shows the configuration of the vehicle 10 according to one embodiment of the application. The vehicle 10 comprises a plurality of electronic components which are connected to each other via an in-vehicle communication network 28. The in-vehicle communication network 28 can for example be a standard in-vehicle communication network installed within the vehicle, such as a CAN bus (Controller Area Network), a LIN bus (Local Interconnect Network), a LAN bus (Local Area Network), a MOST bus and / or a FlexRay bus, etc.

[0051] In Figure 1 In the example shown, the vehicle 10 comprises a control unit 12 (ECU1) for a braking system. Here, the braking system relates to components which can brake the vehicle. The vehicle 10 further comprises a control unit 14 (ECU2) which controls a drive train. Here, the drive train relates to driving components of the vehicle. The drive train can comprise an engine, a transmission, a drive / push axle, a differential and axle drives. Furthermore, the vehicle 10 comprises a control unit 16 (ECU3) which controls a steering system. Here, the steering system relates to components which can implement a directional control of the vehicle.

[0052] The control units 12, 14, 16, 18 and 22 can also receive vehicle operating parameters from the above-mentioned vehicle subsystems which detect the vehicle operating parameters by means of one or more vehicle sensors. The vehicle sensors are preferably sensors which detect and identify the state of the vehicle and the state of the vehicle components, in particular their motion state. The sensors can comprise a vehicle speed sensor, a yaw rate sensor, an acceleration sensor, a steering wheel angle sensor, a vehicle load sensor, a temperature sensor, a pressure sensor, etc. Sensors can also be arranged along the brake lines, for example, in order to send signals which show the brake hydraulic pressure at different points along the hydraulic brake lines. Other sensors can be provided in the vicinity of the wheels, which detect the wheel speed and the brake pressure applied to the wheels.

[0053] Furthermore, the vehicle sensors of the vehicle 10 comprise a satellite navigation unit 24 (GNSS unit). It should be noted that GNSS in the context of the present application stands for all global navigation satellite systems (GNSS), such as GPS, AGPS, Galileo, GLONASS (Russia), Compass (China), IRNSS (India), etc.

[0054] Furthermore, the vehicle 10 comprises one or more sensors designed to detect the surroundings of the vehicle, wherein the sensors are mounted on the vehicle and detect images of the surroundings of the vehicle or identify objects or states within the surroundings of the vehicle. The surroundings sensors 26 comprise, inter alia, cameras, radar sensors, lidar sensors, ultrasonic sensors, etc. The surroundings sensors 26 can be arranged in the interior or on the exterior of the vehicle, for example on the outside of the vehicle. For example, a camera can be provided in the front region of the vehicle 10, which is arranged to capture images of the region in front of the vehicle.

[0055] Furthermore, the vehicle 10 also comprises a sensor processing unit 22 (ECU 4), which is able to determine a grid map with occupancy probabilities on the basis of the sensor data in the form of point clouds provided by the environment sensors 20. The sensor processing unit 22 (ECU 4) can also be set up to estimate the criticality of the driving situation on the basis of the processed sensor map (see the description of Figure 5 If this is necessary, the determination of the criticality can also be processed by a separate control unit.

[0056] The vehicle 10 also comprises a control unit for automated driving 18 (ECU 5). The control unit for automated driving 18 is designed to control the vehicle 10 so that it can take action in road traffic completely or partially independently of a human driver. If the operating state of automated driving is activated on the control side or the driver side, the control unit for automated driving 18 determines parameters for the automated operation of the vehicle, for example target speed, target torque, distance to the vehicle in front, distance to the road edge, steering process, etc., on the basis of data provided over a predetermined driving path, environmental data captured by the environment sensors 20 or processing data provided by the sensor processing unit 22 and vehicle operating parameters detected by means of vehicle sensors, which are transmitted from the control units 12, 14 and 16 to the control unit 18.

[0057] The vehicle 10 also comprises a user interface 25 (HMI = Human-Machine- Interface), which enables the vehicle occupants to interact with one or more vehicle systems. The user interface 25, for example a GUI = Graphical User Interface, can comprise an electronic display for outputting graphics, symbols and / or content in the form of text and an input interface for receiving inputs, for example manual inputs, voice inputs and inputs by means of gestures, head or eye movements. The input interface can comprise, for example, a keyboard, switches, a touch-sensitive screen (touch screen), an eye tracker, etc.

[0058] Figure 2aA block diagram is shown which shows the configuration of the autonomous driving control unit 18 (ECU5). The autonomous driving control unit 18 can be, for example, a control unit (electronic control unit ECU or electronic control module ECM). The autonomous driving control unit 18 comprises a processor 41. The processor 41 can be, for example, a computing unit, such as a central processing unit (CPU = Central Processing Unit), which executes program instructions. The processor of the autonomous driving control unit 18 is designed, for example, to calculate an optimal driving position (e.g. following distance to a preceding vehicle or lateral offset, etc.) in consideration of reliable lane region cases on the basis of information of a sensor-based environment model when driving in view of a planned driving operation. The calculated optimal driving position is used to control actuators of the vehicle subsystems 12, 14 and 16, such as brake actuators, drive actuators and / or steering actuators. The autonomous driving control unit 18 further comprises a memory and an input / output interface. The memory can comprise one or more non-volatile computer-readable media and comprise at least one program storage area and a data storage area. The program storage area and the data storage area can comprise a combination of different types of memory, such as a read-only memory 43 (ROM = Read-Only Memory) and a random access memory 42 (RAM = Random Access Memory) (e.g. dynamic RAM ("DRAM"), synchronous DRAM ("SDRAM"), etc.). The autonomous driving control unit 18 can further comprise an external storage drive 44, such as an external hard disk drive (HDD), a flash drive or a non-volatile solid state drive (SSD). The autonomous driving control unit 18 further comprises a communication interface 45 through which the control unit can communicate with an in-vehicle communication network (28 in Fig. 2).

[0059] Figure 2bA sensor processing unit 22 according to the present invention is schematically illustrated. All components of the sensor processing unit 22 are connected via an internal communication network 46. The sensor processing unit 22 includes an application-specific integrated circuit 47 (ASIC, also known as FPGA). The integrated circuit 47 may be, for example, a GPU or a GPU cluster. The integrated circuit 47 is configured to convert sensor data in the form of a point cloud into an occupancy grid map of the sensor's field of view. The sensor processing unit 22 includes a processor 41. The processor 41 may be, for example, a computing unit such as a central processing unit (CPU), which executes program instructions to compile information for processing, for example, via the integrated circuit 47. The sensor processing unit 22 also includes memory and input / output interfaces. The memory may include one or more non-volatile computer-readable media and include at least one program storage area and a data storage area. The program storage area and the data storage area may include a combination of different types of memory, such as read-only memory 43 (ROM) and random access memory 42 (RAM) (e.g., dynamic RAM (DRAM), synchronous DRAM (SDRAM), etc.). In addition, the sensor processing unit 22 may also include an external storage drive 44, such as an external hard disk drive (HDD), a flash drive, or a non-volatile solid state drive (SSD). The sensor processing unit 22 also includes a communication interface 45, through which the control unit can communicate with the vehicle communication network (28 in Figure 2).

[0060] Figure 3 An exemplary surroundings sensor 26 is shown, in particular a radar sensor. The radar sensor 26 is an identification and localization device based on electromagnetic waves in the radio frequency range. The radar sensor transmits a signal as a beam of electromagnetic waves (primary signal) and receives echoes reflected from objects (secondary signal). Information obtained from this, such as the time difference between the two times, is obtained about the detected event ("target point"). Information such as the azimuth indicating the direction towards the target point and elevation angle θ i , the distance r relative to the target point i , explaining the radar sensor 26 and the target point P i The radial velocity v of the relative motion between i and lateral velocity I i The relative motion can be determined, for example, by the Doppler effect caused by the frequency shift of the reflected signal. By stringing together the individual measurements, it is possible to calculate the target point P iThe absolute velocity of the vehicle 1 and the distance and absolute velocity of the target points P i The target points P i can be ascertained as belonging to a distinct object (clustering method), the contour of the object is identified and, if the resolution of the radar sensor 26 is sufficient, an image of the object is obtained.

[0061] Figure 4a , 4b , 4c shows a two-dimensional raster map obtained from the detection events of the radar sensor. In Figure 4a , the field of view 31 of the radar sensor is in front of the vehicle 1 on which the radar sensor is mounted. An object 32 is located within the field of view 31 of the radar sensor. Radar waves reflected on the object 32 generate detection events within the radar sensor, which are transmitted to the radar sensor in the form of target points P i . The totality of the detection events detected in this way is in the form of a point cloud, which can be evaluated in an evaluation unit, either externally or by the sensor. Figure 4b A two-dimensional raster map 33 (shortly: "grid") is shown, which is designed in such a way that it divides the field of view 31 of the radar sensor into individual cells according to a Cartesian coordinate system. Each target point P i is unambiguously assigned to one cell of the raster map by means of a transformation of the position coordinates from the polar coordinate system of the radar sensor into the Cartesian coordinate system of the raster map 33, which is known to the person skilled in the art. Figure 4c The cells in which at least one target point P i is contained are shown hatched, in contrast, the cells which are not assigned to a target point P i are shown hatched.

[0062] In this raster map, all sensor data, primarily the point cloud, are entered. By means of known sensor fusion techniques, the detection events of a plurality of environmental sensors and the information derived therefrom can be sorted into this raster map. Here, too, it can be the detection events of sensors of different sensor types, for example radar sensors, lidar sensors, ultrasound, etc. If the sensors are distributed around the vehicle, such a raster map can depict the environment around the vehicle.

[0063] By means of the distinction between static and dynamic objects, the static environment of the vehicle 1 can be identified. This enables object recognition and the determination of object parameters for the recognized objects, for example the velocity v of the object, the extension R i in three spatial dimensions or the acceleration a of the object. The recognized objects and the object parameters known therefor can be stored in the raster map and / or on an external memory.

[0064] In this way, and / or by using information about the own motion of the vehicle 1 (also called "ego motion"), for example the vehicle velocity and the vehicle position (see Fig. 1), the velocity v of the vehicle 1 can be determined, for example, by means of the radar sensor 26 and / or the lidar sensor 27.Figure 1 GNSS 24) can enter the information according to Figure 4a , 4b the grid map 33 of 4c into the accumulated grid map. The vehicle 10 moves on this accumulated grid map and continuously recalculates the new position of the vehicle 1. The sensor data are entered into the accumulated grid map in each measurement cycle containing the compensation of the vehicle position. By means of this method, the sensor data can be accumulated with respect to time and collected and evaluated statically with respect to the "global" coordinate system.

[0065] In the ECU 4 of the sensor processing unit 22 Figure 1 , object recognition is first performed on the detected point cloud according to techniques known to the person skilled in the art in order to detect all movable and immovable objects and their position x i,0 , extent R i and velocity v i within the sensor field of view. Different methods known to the person skilled in the art can be applied for object recognition, such as edge recognition, transformation or size and color recognition. Object recognition can also be performed using artificial neural networks.

[0066] According to the application, such an object recognized by the sensor processing is specified by means of an object function, which defines a danger zone around the object. According to one embodiment, for specifying a movable object, in particular the function equation is chosen for the object function O(x, x0), wherein the object function O(x, x0) is defined by a Gaussian function with a given standard deviation σ = (σ1, σ2, σ3), which specifies the "breadth" of the danger zone around the object:

[0067]

[0068] wherein x = (x1, x2, x3) specifies an arbitrary location as a position vector in three-dimensional space and x0= (x 1,0 ,x 2,0 ,x 3,0 ) specifies the current position as a position vector of the object in three-dimensional space. is also a normalization constant, which is further specified below. The object function O(x, x0) attributes a "potential for accidents" or hazard potential O(x, x0) to the location x.

[0069] Figure 5 A contour plot of an exemplary object function is shown, which is defined by a Gaussian function with a given standard deviation. The object vehicle 20 can be seen, whose danger zone is specified by the object function O(x i ,x i,0 ). The object function O(x i ,x i,0) follows a Gaussian function whose center is located at the current position x of the target vehicle 20 i,0 The ellipse corresponds to the object function O(x i ,x i,0 ) contour lines 51, 52, 53. In the cross-hatched area within the first contour line 51, the object function O(x i ,x i,0 ) has a higher value (ie, the risk from the object is higher). In the dotted area between the first contour line 51 and the second contour line 52, the object function O(x i ,x i,0 ) has a smaller value than in the cross-hatched area. In the dotted area, the distance from the current position of the vehicle is greater, so that the danger from the object is smaller than in the cross-hatched area. In the shaded area between the second contour line 52 and the third contour line 53, the object function O(x i ,x i,0 ) has a smaller value than within the dotted area, meaning the risk from the object is smaller. Outside contour line 53, that is, when the distance from the current position of target vehicle 20 is greater, the risk from the object approaches zero. Second contour line 52 corresponds to a distance from the vehicle position that corresponds to the standard deviation in the respective spatial directions. σ1 corresponds to the standard deviation in the direction of travel. σ2 corresponds to the standard deviation perpendicular to the direction of travel. Target vehicle 20 is moving in the direction of travel at a speed v. Therefore, the standard deviation σ1 in the direction of travel is greater than the standard deviation σ2 perpendicular to the direction of travel.

[0070] Object function O(x i ,x i,0 ) can be used to determine the criticality within the context of trajectory planning, as shown in the following reference Figure 6 Here, the standard deviation σ of the object i This leads to the understanding that a safe distance should be maintained from the object in order to avoid critical situations or even collisions.

[0071] The Gaussian-based object function of the object vehicle 20 is symmetrical along the spatial axis. For example, Figure 5 The object function O(x i ,x i,0 ) is symmetrical in the direction of travel (forward direction) and the backward direction.

[0072] The target function can also be selected so that different standard deviations σ are selected for the driving direction (forward) and the reverse direction (backward) v or σ h When driving forward, for example, the standard deviation σ v can be selected parameter-dependently (depending on driving parameters, such as speed, etc.), and the standard deviation σh Constantly (independent of parameters) selected (predetermined fixed value, which for example corresponds to the size of the vehicle). The standard deviation can also be selected dependent on the angle between the driving direction vector v and the vector x - x0, so that the standard deviation σ is continuously dependent on the angle from σ v (forward) to σ h (backward) changes.

[0073] If the object vehicle 20 is located in front of the own vehicle, which is set up according to the application for determining the criticality of the driving situation by means of the object function, the own vehicle should maintain the same safety distance to the object vehicle 20 as the own vehicle would do when driving in front of the object vehicle 20. When the own vehicle is driving behind the object vehicle 20, the own vehicle should at least keep the distance of the braking distance of the object vehicle 20 apart, so that the own vehicle can still be braked when the object vehicle 20 brakes. When the own vehicle is driving in front of the object vehicle 20, the own vehicle 10 should likewise at least keep the braking distance of the object vehicle apart, so that the object vehicle can still be braked when the own vehicle in front of the object vehicle brakes.

[0074] In general, the standard deviation σ i is selected as a function of one or more object or surrounding parameters. The standard deviation σ i may for example depend on the object speed v i , the object acceleration a i , the object extension R i , or weather conditions such as visibility Sw, sun position Ss, etc.:

[0075] σ i = f(v i , a i , R i , Sw, Ss,...)

[0076] As mentioned above, the standard deviation σ i of the object is of the meaning that a safety distance should be maintained to the object in order to avoid a critical situation or even a collision. According to one embodiment of the application, the standard deviation σ i is thus set exemplarily equal to the estimated braking distance s. If the object is within the critical region, it cannot be ensured that a damage can be avoided even if the object brakes. According to this embodiment, the standard deviation σ is calculated in the following form:

[0077] σ = σ v = s(v) = a • v 2 + b • v

[0078] where v denotes the speed of the object. Here, a quadratic relationship a · v 2 is assumed between the braking distance s and the speed v, which is known to the person skilled in the art. For generalization, the model of this embodiment also comprises a linear term b · v. The constants a and b are model parameters which must be determined first.

[0079] In one embodiment of the application, the model parameters are read from a look-up table which has been stored previously on the control unit (22) according to the application on the basis of the currently called object parameters or the surrounding parameters. In another embodiment of the application, the model parameters are determined by means of a multiple regression on the basis of data sets which have been stored previously on the control unit (22) according to the application. These data sets contain values for the model parameters at different object parameters or surrounding parameters. In another embodiment of the application, these data sets are used to train an artificial neural network. In this case, the control unit (22) according to the application also comprises an artificial neural network which determines the model parameters on the basis of the object parameters and / or the surrounding parameters. In this way, the parameters a and b can be determined, for example, in dependence on the surrounding parameters (for example humidity, temperature, fog intensity, etc.) or the object parameters (for example object contour or acceleration). Figure 1 Figure 1 Figure 1

[0080] According to another embodiment, in addition to the estimation of the braking distance, a simulation of a lengthened braking distance on the basis of a poor visibility and thus a poor reaction of the driver is also possible:

[0081] σ = σ v + σ 雾 = a · v 2 + b · v + c 雾

[0082] where the parameter c 雾 simulates the visibility conditions Sw.

[0083] The parameter c 雾 may be chosen, for example, as follows:

[0084]

[0085] where, here, three fog densities are distinguished for the visibility Sw: no fog, thin fog, thick fog. In the case of thick fog, the standard deviation is increased significantly, since the braking distance of the object and thus the required safety distance to the object is higher. In the case of no fog, the value c 雾 corresponds in the order of magnitude to the size of the object.

[0086] In addition, the size R i of the vehicle can also be included in the determination of the standard deviation, here, for example, the size of the vehicle in the direction of travel ​​​

[0087] σ = σ v + σ 雾 + σ R = a · v 2 + b · v + c 雾 + c R

[0088] wherein the contribution term c R may exemplarily be assumed to be c R = 0.5 m.

[0089] It is noted here that the braking distance can also use a higher polynomial of the object velocity v, so that it is applicable for three spatial dimensions:

[0090]

[0091] wherein In this case, the individual parameters a j are determined for the current case by means of a multiple regression or a neural network from a set of known braking distance data, as already described. The constant value c ext may model external influences, such as fog or darkness, as in the example described above. The constant value c ext may likewise be taken from known braking value data or databases or set as a fixed value by the manufacturer, using a multiple regression or a neural network.

[0092] Here, the standard deviation σ i is a measure of the extension of the danger zone around the object, for example a vehicle. An object that is not moving (object velocity v i = 0) poses a lower danger, since the danger zone is approximately given by the extension R i of the object. The danger zone can thus be classified with a small standard deviation σ R , which is given by the contribution term c v of the vehicle size. An object that is moving quickly has a larger danger zone based on the speed-dependent braking distance and includes a larger standard deviation σ i .

[0093] Dangerous weather conditions, such as humidity (which can be measured, for example, using a humidity meter), road icing (temperature T < 4°C, which can be measured using a thermometer), different road surfaces (which can be measured by means of acoustic sensors) or fog (fog density can be determined from camera images), which require a higher safety distance, can be taken into account in the criticality determination and thus in the trajectory planning, in particular, using a model that takes into account the standard deviation σ iThe function of all these factors is plotted. In this case, trajectory planning can be performed in such a way that the planned trajectory of the host vehicle extends as far as possible outside the danger zone of the mobile object, as described below.

[0094] Based on the above object function, not only the danger zone around the vehicle itself can be determined, but also the danger zone around all other traffic participants identified by the sensor can be determined. In order to determine the degree of criticality during autonomous driving, that is, to evaluate how "critical" the traffic situation is for the vehicle itself, the above object function of the vehicle can be used to determine the degree of criticality during autonomous driving. ego and the object functions O of other traffic participants n Import location-related criticality level I cri In the method of the present invention, the object function O of the vehicle is used. ego and the respective object functions O of other traffic participants n The criticality level I is calculated in the form of the overlap integral of the sum cri :

[0095]

[0096] or

[0097]

[0098] Here, x ego,0 is the position of the vehicle, x n,0 , n=1…N are the corresponding positions of other identified traffic participants, and N is the number of traffic participants.

[0099] Here, I cri It can be viewed as a "convolution" of the accident potential of the vehicle and other traffic participants, describing the probability of an accident in each direction and time. The criticality of each time is the maximum probability of an accident at a given point in time, and the criticality of each scenario is the maximum value of the criticality over time.

[0100] Criticality Level I cri The following standardization is applied to the case where the bumper of the own vehicle contacts the bumper of the other vehicle. cri = 1. If no accident occurs (the host vehicle and the other vehicle are farther apart than if the bumpers were touching), the criticality level I cri The value range is For a range of values There is a collision / accident. In this case, the severity level I cri Values ​​>1 are used to evaluate the severity of a "crash" or for risk analysis.

[0101] Assuming x ego,0indicates the current position of the host vehicle, criticality I cri (x ego,0 ,x 1,0 ,…,x N,0 ) evaluates the current driving situation with regard to the accident risk resulting from the hazard potential of the individual traffic participants. If instead of the current host vehicle position x ego,0 an arbitrary position x is used as potential host vehicle position, a criticality map can be created which attributes to each arbitrary position x (potential host vehicle position) a criticality I cri (x,x 1,0 ,…,x N,0 ) of the accident risk. This criticality determination has the advantage that it takes into account the own hazard potential or accident potential by means of the individual objective function O ego or O n of each traffic participant.

[0102] The criticality map I cri (x,x 1,0 ,…,x N,0 ) can be used in the context of trajectory planning (“Motion Planning”) for criticality determination, in particular for defining a minimum-risk maneuver. By means of the standard deviation, the criticality I cri depends on the object parameters (host vehicle and other traffic participants’ positions x i , speeds v i and accelerations a i ) as well as on the surrounding environmental parameters (temperature, humidity, road conditions, etc.).

[0103] While in the above-described embodiments only the objective functions of the other recognized traffic participants are taken into account, in alternative embodiments the objective functions of other objects such as static objects, e.g. buildings, trees, traffic signs, etc. can also be taken into account.

[0104] Figure 6 An exemplary two-dimensional criticality map is shown, e.g. obtained by means of the above-described “convolution” of the hazard potential. The shown two-dimensional criticality map shows a density map or contour plot of the criticality I cri (x,y) of the potential host vehicle positions x = (x1,x2) = (x,y). The criticality I cri(x, y) is displayed here in four levels: <0.3 (white), 0.3-0.6% (dash), 0.6-0.9 (dots), and >0.9 (cross lines). The density or contour map shows two dynamic objects 61 and 62, which are vehicles of other road users. In addition, the density or contour map also shows static objects, in this case in particular two buildings 63 and 64. The first road meets the second road at an intersection and then continues. The criticality level I of the potential position (x, y) of the own vehicle, which maintains a sufficient distance from vehicles 61 and 62, is cri (x, y) is small, while the criticality level I for the position within the danger zone for vehicles 61 and 62 is cri (x, y) is increased. The criticality level I of the potential vehicle position (x, y) located "within" the extension of one of the vehicles 61 and 62 cri (x, y) exceeds the value 1, because the vehicle's stay at these positions means an unavoidable accident. Similarly, the criticality level I of the potential vehicle position (x, y) located "inside" the extension of buildings 63 and 64 is cri (x,y) exceeds the value 1 because the host vehicle's stay at these positions also means an unavoidable accident.

[0105] The acquired criticality map can now be transmitted to the autonomous driving control unit ( Figure 1 18) is used for trajectory planning in order to determine a reliable route for the vehicle 10. Combined with the criticality level I cri With a static description of the environment, trajectory planning can now be performed using control technology. The trajectory planning should always avoid locations that are classified as accident risks by the criticality map. The criticality map can be adapted to the changing object and environment parameters over time.

[0106] In an embodiment of the present invention, the sensor data is stored in a grid map (see Figure 4a 、 4b and 4c), the criticality level I is determined for each grid cell of the grid map. cri And it is stored as an additional value in the grid cell.

[0107] Figure 7 The flowchart shows the steps of the method according to the present invention for determining the degree of criticality, ie how it is used on a control unit according to the present invention in the vehicle 10 according to the present invention. For simplicity, only one spatial dimension is considered in the method shown.

[0108] In the first step S701, the control unit ( Figure 1 22) receives environmental parameters, such as temperature, road conditions, air humidity, etc. In the second step S721, the control unit (Figure 1 the object parameters, such as the position, velocity, acceleration, extension, etc. of the recognized object. In a third step S703, the control unit (22) according to the application determines the model parameters a, b and c based on the environmental parameters and the object parameters for the recognized object. This can be done by the existence of measured data of a multiple regression in a database or by an artificial neural network. In a fourth step S704, the control unit (22) according to the application determines the standard deviation Figure 1 Figure 1

[0109] σ = a · v + b · v + c 2

[0110] wherein v denotes the velocity of the recognized object. In a fifth step S705, the control unit (22) according to the application determines a location-dependent object function Figure 1

[0111]

[0112] wherein is a normalization constant and x0denotes the position of the recognized object. Also the ego object function is determined for the ego vehicle 10. In a sixth step S706, the control unit (22) according to the application determines a criticality Figure 1

[0113]

[0114] the criticality map or the location-dependent criticality values can be transmitted to an external server, which performs a trajectory planning based on them. The trajectory planning is performed, for example, by planning a trajectory that is suitable for all times and all positions with cri <0.8.

[0115] It should be noted here that the seventh step S706 can also be performed on the control unit according to the application, however it is meaningful to perform the trajectory planning in the own control unit based on the improved modularity.

[0116] In order to save the computing effort in the control unit according to the application, which is installed in the vehicle 10 itself, the steps S702 to S705 can also be performed on an online server connected to the control unit according to the application via a network.

[0117] ​​​​​Today, the described criticality determination is already able to handle free space and confinement cases (by adding collision possibilities to the boundaries). Future needs are: a response per scene criticality regarding speed and direction; use of responses in order to obtain critical (to be thresholded) scenes by micro-simulation or field tests; in addition application of responses in fully parametrized nano-scale simulations in order to correct responses in full parameter set; and filtering of random scenes by SOTIF [Safety Of The Intended Functionality] (known but uncertain) range with selection of responses.

[0118] Figure 4a 、 4b It has been shown in Figs. 4a, 4b and 4c how a grid map illustrating the sensor environment of the present application can be established based on e.g. radar or lidar data. The establishment of the criticality map can be on a grid map, however, is not limited thereto. In general, the method according to the present application for establishing a criticality map or determining a criticality is not limited to sensor data in a grid map or point cloud, let alone radar or lidar data. On the contrary, the method according to the present application can be applied to all types of sensor data allowing a reliable object recognition, distance recognition and speed recognition.

[0119] List of reference signs

[0120] 10 vehicle (own vehicle)

[0121] 20 object vehicle (vehicle of other road user)

[0122] 12 ECU1 brake system

[0123] 14 ECU2 drive train

[0124] 16 ECU3 steering system

[0125] 18 ECU5 autonomous driving

[0126] 22 ECU4 sensor processing unit

[0127] 24 GNSS

[0128] 25 HMI

[0129] 26 surroundings sensor

[0130] 31 field of view range of sensor

[0131] 32 object (vehicle)

[0132] 33 grid map

[0133] 34 single cell of a raster map with high occupancy probability

[0134] 35 single cell of a raster map with low occupancy probability

[0135] 41 CPU

[0136] 42 RAM

[0137] 43 ROM

[0138] 45 CAN-IF

[0139] 44 memory unit

[0140] 46 communication system

[0141] 47 application-specific circuitry

[0142] 61 dynamic object 1 (vehicle)

[0143] 62 dynamic object 2 (vehicle)

[0144] 63 static object 1 (building)

[0145] 64 static object 2 (building)

[0146] P i point (probe)

Claims

1. A control unit (22) which is set up to calculate a criticality (I cri ) of a driving situation in the form of an overlap integral of an object function (O ego ) of the own vehicle (10) with one or more object functions (O n ) of other static or dynamic objects or other road users of the surroundings of the own vehicle (10), wherein An object function (O n ) is described with a Gaussian distribution giving the potential of a hazard originating from an object, wherein the Gaussian distribution is determined by a standard deviation (σ) parameterizing a danger zone around the object, wherein different standard deviations are chosen for the driving direction and the backing direction, wherein the object function is determined by a standard deviation (σ), and wherein the control unit is set up for determining a danger zone around the object by means of a polynomial model determining the standard deviation, wherein v denotes the object speed, and wherein the individual parameters a j and c ext are determined by means of a multiple regression or by means of a neural network from a known braking distance data set.

2. The control unit (22) according to claim 1, wherein The driving situation is predetermined by potential positions of the own vehicle (10) and by object parameters and surroundings parameters of other static or dynamic objects or traffic participants of the surroundings of the own vehicle.

3. The control unit (22) according to claim 1, which is further set up to perform an object recognition based on sensor data received from surroundings sensors (26) in order to recognize objects (61, 62) in the surroundings of the own vehicle and to gain object parameters about the objects (61, 62).

4. The control unit (22) according to one of claims 1 to 3, wherein The object function selects a function equation: where i = 1, 2, 3; is a normalization constant; σ = (σ1, σ2, σ3) is the standard deviation of the objective function; and x0= (x 1,0 ,x 2,0 ,x 3,0 ) is the location of the object.

5. The control unit (22) according to one of claims 1 to 3, wherein The Gaussian distribution is determined by a standard deviation (σ) and wherein the standard deviation (σ) is a function of the object parameters and / or the surroundings parameters.

6. The control unit (22) according to claim 5, wherein The object parameters include a velocity of the object and / or an acceleration of the object.

7. The control unit (22) according to claim 5, wherein The surroundings parameters include an outside temperature, a road surface condition, a visibility, a sun orientation or other outside parameters.

8. The control unit (22) according to one of claims 1 to 3, wherein The object function is determined by a standard deviation (σ) and wherein the control unit is set up to gain the standard deviation (σ) based on an estimated braking distance (s) of the object.

9. The control unit (22) according to one of claims 1 to 3, wherein The control unit (22) is set up to gain the standard deviation (σ) by means of the following equation The object function selects a function equation: The Gaussian distribution is determined by a standard deviation (σ) and wherein the standard deviation (σ) is a function of the object parameters and / or the surroundings parameters. The object parameters include a velocity of the object and / or an acceleration of the object. The surroundings parameters include an outside temperature, a road surface condition, a visibility, a sun orientation or other outside parameters. The object function is determined by a standard deviation (σ) and wherein the control unit is set up to gain the standard deviation (σ) based on an estimated braking distance (s) of the object. The control unit (22) is set up to gain the standard deviation (σ) by means of the following equation determining a criticality of the driving situation (I cri ), wherein i indicates the recognized object, N represents the total number of all recognized objects, O ego represents the object function of the own vehicle, and O n represents the object function of the recognized object.

10. The control unit (22) according to one of the claims 1 to 3, wherein The criticality (I cri ) is normalized so that in the case of contact of the vehicle with another object, the criticality equals 1.

11. The control unit (22) according to one of the claims 1 to 3, wherein The control unit (22) is set up to learn a plurality of criticality levels (I cri (x,y)) for each driving situation with different potential positions (x, y) of the own vehicle in order to obtain a criticality map therefrom.

12. Method for calculating a criticality (I ego ) of a driving situation in the form of an overlap integral of an object function (O n ) of the own vehicle (10) with one or more object functions (O cri ) of other static or dynamic objects or other road users of the surroundings of the own vehicle (10), wherein An object function (O n ) is described with a Gaussian distribution giving the potential of a hazard originating from an object, wherein the Gaussian distribution is determined by a standard deviation (σ) parameterizing a danger zone around the object, wherein different standard deviations are chosen for the driving direction and the backing direction, wherein the object function is determined by a standard deviation (σ), and wherein the control unit is set up for determining a danger zone around the object by means of a polynomial model determining the standard deviation, wherein v denotes the object speed, and wherein the individual parameters a j and c ext are determined by means of a multiple regression or by means of a neural network from a known braking distance data set.

Images