Method and device for determining an off-road route for a vehicle

Abstract

The invention relates to a method (1) for determining an off-road route for a vehicle (2), wherein driving lines (3) are determined for the vehicle (2). The invention is characterized in that terrain properties (GE) are determined, wherein at a time (i) target points (Z) are marked on a geometric map (GK) at a distance from the vehicle (2), wherein connecting lines (L) from a vehicle position to the target points (Z) are determined on a plane spanned by the geometric map (GK), and wherein the target points (Z) are represented in a matrix and derivatives of the connecting lines with respect to the path are formed, wherein a comparison of determined terrain properties (GE) with driving properties (F) of the vehicle (2) is then carried out, wherein if the driving properties (F) exceed the terrain properties (GE), a driving line (3) is output as drivable to a vehicle user and / or to a system for automated driving operation. The invention further relates to a device for carrying out the method.

Description

[0001] The invention relates to a method for determining an off-road route for a vehicle, wherein driving lines for the vehicle are determined. The invention further relates to a device for carrying out the method.

[0002] Off-road driving away from established roads or paved tracks is generally not feasible. Therefore, it is advantageous to be able to verify the suitability of the terrain.

[0003] For state-of-the-art vehicles, it is therefore essential to continuously check the suitability of the terrain while driving. This is particularly important for off-road vehicles, although the driver must perform this check manually. Perception approaches from the Advanced Driver Assistance Systems (ADAS) for semi-autonomous or autonomous driving are disadvantageously not designed for unstructured terrain, as they assume a generally traversable and structured environment. Furthermore, existing state-of-the-art systems rely on long sensor ranges and therefore sometimes exhibit significant latency in perception.

[0004] A method known from the prior art is image semantic segmentation. This method segments areas around the vehicle with similar object properties. Machine vision methods, such as convolutional neural networks, are used to recognize objects like cars, vans, or pedestrians.

[0005] Furthermore, there are image-based approaches, for example, which have disadvantages such as problems with the accuracy of semantic segmentation. For instance, image data from the surround-view camera is played back with a time delay to allow viewing under the hood, without any evaluation of the image data.

[0006] From DE 10 2012 021 420 B4 a method for determining a drivable off-road track is also known, in which the determination of a driving route is carried out taking into account driving characteristics on the basis of a terrain profile.

[0007] From DE 10 2021 126 925 A1, a navigation system for off-road vehicles is known, whereby it is intended to create a 3D map showing the terrain and to provide the driver with a terrain map, for example via a head-up display, which, in addition to the terrain model, shows several routes and drivability information, in particular drivable off-road routes.

[0008] DE 10 2009 006 409 A1 discloses a method for route optimization which can also be applied when driving off-road or in unpaved terrain and which provides for the recording of at least one influencing factor such as terrain characteristics as map information in addition to the topographical features, which can be used for the drivability or route finding.

[0009] From DE 103 38 241 A1, a terrain navigation system is known, wherein it is designed to receive a terrain target and query several terrain routes for the starting point and the terrain target, wherein at least one terrain route corresponds to reaching the terrain target. The system aims to reach the terrain target from a starting point on a paved road via a known terrain route that is not located on a paved road.

[0010] The object of the present invention is therefore to provide a method and a device that overcome the aforementioned disadvantages. In particular, it should be possible to derive the drivability of terrain quickly, computationally efficiently and accurately.

[0011] According to the invention, this problem is solved by a method with the features of claim 1, and in particular of the characterizing part of claim 1. Advantageous embodiments and further developments are described in the dependent claims. A device for carrying out the method is also described.

[0012] The core of the inventive method involves determining terrain properties. At a given time, target points are marked on a geometric map at a distance from the vehicle. Connecting lines from the vehicle's position to the target points are determined on a plane spanned by the geometric map. The points are represented in a matrix, and derivatives of the connecting lines are calculated based on the path. Subsequently, the determined terrain properties are compared with the vehicle's driving characteristics. If the driving characteristics exceed the terrain properties, a driving line is output as passable to a vehicle user and / or to a system for automated driving operation. A derivative based on the path describes, in particular, the respective path between the vehicle and the respective target point.

[0013] This advantageously creates a way to perform a rapid, accurate, and computationally efficient assessment of vehicle drivability based on geometric sensor data. Sensors already installed in a vehicle, such as depth cameras, radar, lidar, and / or ultrasound, can be used to scan terrain very precisely.

[0014] The method is designed to determine an off-road route for a vehicle, thereby defining driving lines for the vehicle. In other words, route planning for off-road journeys can be created based on an off-road map generated using environmental sensors, and in particular with the aid of a driver assistance system.

[0015] The sensors can be located on or near the roof of the vehicle. The vehicle is, in particular, an off-road vehicle or an all-terrain vehicle.

[0016] According to the invention, comfort can be improved. In particular, this results in improved safety for the driver during off-road driving, whereby the driver and the vehicle can be better protected by safely calculating drivability.

[0017] According to an advantageous further development of the idea, it can be provided that the target points are marked as a virtual circle around the vehicle, and / or the target points are marked by the system or user, and / or the connecting lines are designed as a function, and / or the terrain characteristics include information on the step height and / or gradient.

[0018] A radius R can be defined as the destination. Target points can then be marked at any desired interval within this radius. These target points, including intermediate points, can be actively set by the driver. This can be done, for example, via a touchscreen on a vehicle display, where a map can be shown. Alternatively, in a highly automated vehicle, the destination can be set externally, for example, via an app.

[0019] Another advantageous embodiment may provide that the driving characteristics include a safety margin, which is chosen to be larger for automated driving operation than for manual driving operation.

[0020] Another advantageous embodiment can provide that, for data reduction, the matrix is ​​converted from a three-dimensional representation to a two-dimensional representation, where one parameter of the two-dimensional representation describes a Euclidean distance. This can save considerable amounts of data.

[0021] According to a highly advantageous further development of this idea, it can be provided that the drivability is recalculated after a predetermined time or distance, where the distance or time depends on the driving speed. In particular, the drivability can also be recalculated continuously. Since the accuracy and resolution and the viewing angle of the sensor, and thus also the accuracy and resolution of the geometric map, change with proximity to the target, the drivability can be calculated more precisely.

[0022] Another advantageous embodiment may provide that, if a route classified as impassable is selected, it is driven on for automated driving until the driving characteristics no longer meet the terrain characteristics, whereby the speed and / or steering of the vehicle is reduced.

[0023] In particular, the vehicle can be driven back along its path towards its starting position as soon as its driving characteristics are no longer sufficient for the terrain. In the event of driving back, it can be checked whether a 180° turn, also known as a G-turn or tank turn, is possible. If so, the vehicle can be turned and then driven forward towards the starting point.

[0024] A further advantageous embodiment provides that the test of drivability for a rotation of the vehicle around its center of gravity, also known as a G-turn or tank turn, is carried out along the driving lines, wherein the driving lines are arranged concentrically around the vehicle's center of gravity as circular lines. The drivability test can therefore be performed along these circular lines. The concentric driving lines can be represented by discrete points that can be mapped in a matrix, whereby drivability can be determined analogously to the determination of the driving line.

[0025] The invention further relates to a device for carrying out the method. According to the invention, a control unit is included which evaluates the target points from the geometric map in the matrix to determine the terrain characteristics. The device has the same advantages and features as described with regard to the method.

[0026] Further advantageous embodiments of the method and the device according to the invention are also evident from the exemplary embodiment, which is described in more detail below with reference to the figures.

[0027] This shows: Fig. 1. One possible embodiment of the method; Fig. 2 a schematic evaluation of the procedure; Fig. 3 remaining vectors and targets after an assessment of traversability according to an execution of the procedure; and Fig. 4. Formation of vectors during a G-turn according to one implementation of the procedure.

[0028] In the presentation of the Fig. Figure 1 shows a possible embodiment of the method 1, wherein a vehicle 2 is depicted in a terrain. Fig. 1. shows the vehicle 2 at a location P(x,y,h)i and a schematic procedure to derive two-dimensional vectors from height information h, a shift by z and h and n points P on a geometric map GK. Fig. Figure 1 further shows a calculation of lines connecting the current location of the ego-vehicle with m target points Z1, Z2, Z3, Z4 at intervals d on a circle 4 at a distance R on the surface of a geometric map GK. From the resulting vectors V, a drivability can then be calculated, as for example in Fig. 2 shown. Advantageously, the vectors V are calculated and formed for both wheels at a distance of one wheelbase.

[0029] According to the procedure, a circumcircle, circle 4, with a radius R of P(x,y,h)i can be marked as the target. Target points Zm can be defined on the circumcircle at intervals d = [d1, d2 .. dm-1] m. Fig. 1. Z1, Z2, Z3, and Z4 can be marked. For example, a driver can actively set destination points Z or possible intermediate points Z. This can be done, for example, using a touchscreen in a vehicle on a map displayed there. Alternatively, this can be done automatically in the vehicle.

[0030] Furthermore, a connection from P(x,y,h)i to Zj with j = [1 .. m] can be sought by drawing a connecting line L. The connecting line L can be, for example, a polynomial / spline, a circular arc, or the like, particularly under the condition that the orientation of the vehicle's longitudinal axis at P(x,y,h)i is equal to the current orientation of vehicle 2, i.e., when vehicle 2 has not yet turned. Furthermore, a maximum curvature can be chosen such that vehicle 2, with capabilities F, can travel along line L. In particular, line L is chosen such that it lies on a plane spanned by the geometric map GK. Simultaneously, a discretization can be performed so that line L is represented by n points P at intervals e.

[0031] In one embodiment, m vectors of dimension n times 3 can result, where the number is 3 because each point consists of x, y, and h. In another embodiment, a matrix of size m times n times 3 can result. For example, m = 15 and n = 20 can be chosen, resulting in one dimension = 15 * 20 * 3 = 900. For comparison, the original dimension of the map with 1 megapixel has a GK dimension of 1,000,000. Consequently, an information reduction by a factor of 1000 can advantageously be achieved.

[0032] A further reduction of the data is possible if the shift V from P(x,y,h)i to P(x,y,h)i+1 is mapped in 2D, as in particular as in V=V(z,h) in Fig. denoted by 2, where z can be understood as the Euclidean distance from P(x,y)i to P(x,y)i+1. This allows for a dimension of m times n times 2, or, in the example, 15*20*2 = 600. Consequently, an information reduction by a factor of 1600 can advantageously be achieved.

[0033] Terrain properties GE can be derived from the m vectors, especially using computationally efficient methods, as shown in Fig. Figure 2 shows that the step height can be directly derived from the displacement V(z,h), and the slope at P(x,y,h) can be the derivative of L at P(x,y,h). For crests, in particular, the derivative of L = 0 applies, so that the ground clearance when driving over a crest can be derived from the values ​​h of the displacements. The slope at points A and B can be determined from the slope triangle or from the derivative of L at these points.

[0034] The terrain property GE can then be compared with driving characteristics F. If GE > F + S exists at a location P(x,y,h)j, this location can be marked as impassable, and the point and all subsequent points of a vector can be deleted, as shown in Fig. Figure 3 shows that S represents a safety distance, which can be used, for example, to protect components. If an automatic off-road driving function (AOF) with capability F is activated, GE>F+S2 is checked and appropriate action is taken.

[0035] According to Fig. 3. The destinations Z1 and Z4 remain, which the driver can either drive to, or the driver can select a route Z1 or Z4. The corresponding vector can be sent as a trajectory to the automatic off-road driving (AOF) function and then followed. The aforementioned procedural steps can be repeated to recalculate the drivability, particularly after the vehicle has moved a distance x, preferably after a fixed time and / or depending on a speed.

[0036] In another embodiment, it is possible not to drive directly to targets Z1 or Z4, but instead to check whether Z2 or Z3 can be reached. For example, an observation drive can be initiated. The AOF can select a target from the drivable targets Z2 and Z3, in particular the target furthest away, and can traverse the points along the trajectory, especially by controlling the actuators steering, engine, individual wheel and / or all-wheel drive motors, reduction gears, active suspensions, and / or the like. The speed of the drive and / or steering can be further reduced as the driving characteristics F approach the terrain characteristics GE. In particular, the drivability can be continuously recalculated.Since the accuracy, resolution, and viewing angle of the sensor, and thus also the accuracy and resolution of the geometric map (GK), change with proximity to the target, the drivability can be calculated more precisely. If the AOF approaches an impassable point at a distance < threshold 1, where threshold 1 can be defined as a threshold value, the vehicle 2 can be slowed down and, in particular, can reverse its trajectory or drive back to the starting point P(x,y,h)i, either backwards or forwards.

[0037] In another embodiment, an environmental sensor can additionally or alternatively be used to check whether, for example, a G-turn is possible. If this is possible, a 180° G-turn can be performed and the vehicle 2 can be driven forwards back to the starting point.

[0038] According to Fig.4. Method 1 can be applied such that a rotation, in particular a rotation of 180°, around a vehicle center of gravity SP is enabled at a target Z. Vectors can be created from a start S, here S2 and S3, to a target Z, here Z2 and Z3, whereby the vehicle 2 can perform a circular movement of 180° or more during the G-turn. The vectors, in particular, run along concentric circular paths, specifically referred to as circles 4, with a radius r, here r1, r2, and r3. In particular, the vectors are represented as concentric circles on the geometric map GK around the vehicle center of gravity SP with a radius r. The circles can be discretized by points P(x,y,h) at a distance e from each other and converted into a displacement vector V(z,h). Advantageously, this allows for a very efficient representation of the environment of the vehicle 2 relevant for the G-turn.In particular, skills such as FG for a G-turn can be defined, whereby if GE > FG+S3 for a point on the vectors, that point can be removed. A 180° or 360° G-turn is only unlocked in the vehicle if no points have been removed.

[0039] Advantageously, vehicle 2 is equipped with active sensors. In particular, by adjusting the sensor pitch, zooming in, switching the transmitting antenna, illuminating the area, and / or increasing the transmission power of active sensors such as lidar, radar, and / or ultrasound, a first, previously inaccessible point can be measured more precisely, especially in a second step. If the opening angle of the forward-facing sensors is limited, the field of view can be advantageously extended by performing a G-turn clockwise or counterclockwise to the right or left. In other words, the surroundings can be scanned at various angles by rotating vehicle 2 around its center of gravity (CG).Additionally or alternatively, a more complex procedure can be applied which, in particular, enables higher accuracy with greater latency based on a fusion of the geometric map GK with a map with semantic segmentation for terrain classification.

[0040] In a further embodiment, additional sensors or, alternatively, a drone can be used, which in particular has or possesses active sensors. Method 1 can therefore enable even faster and more energy-efficient preprocessing of the drivability assessment.

[0041] Naturally, the implementation examples can also be combined, resulting in various possibilities. For example, the number of target points Z can be chosen arbitrarily.

Claims

[1] Method (1) for determining an off-road route for a vehicle (2), wherein driving lines (3) for the vehicle (2) are determined, characterized by , that terrain properties (GE) are determined, wherein at a time (i) target points (Z) are marked on a geometric map (GK) at a distance from the vehicle (2), wherein connecting lines (L) from a vehicle position to the target points (Z) are determined on a plane spanned by the geometric map (GK), and wherein the target points (Z) are mapped in a matrix and derivatives of the connecting lines (L) with respect to the path are formed, wherein a comparison of determined terrain properties (GE) with driving properties (F) of the vehicle (2) is then carried out, wherein if the driving properties (F) exceed the terrain properties (GE) an output of a driving line (3) as drivable is given to a vehicle user and / or to a system for automated driving operation. [2] Method (1) according to claim 1, characterized by , that the target points (Z) are marked as a virtual circle (4) around the vehicle (2), and / or the target points (Z) are marked by the system or user, and / or the connecting lines (L) are designed as a function, and / or the terrain features (GE) include information on step height and / or gradient. [3] Method (1) according to claim 1 or 2, characterized by , that the driving characteristics (F) include a safety margin (S), which is chosen to be larger for automated driving operation than for manual driving operation. [4] Method (1) according to claim 1, 2 or 3, characterized by , that for data reduction the matrix is ​​converted from a three-dimensional representation (x,y,z) into a two-dimensional representation (z,h), where z describes a Euclidean distance. [5] Method (1) according to any one of claims 1 to 4, characterized by, that the drivability is recalculated after a predetermined time or distance, whereby the distance or time depends on a driving speed. [6] Method (1) according to any one of claims 1 to 5, characterized by , that when a route classified as not drivable is selected (3), it will be driven on for automated driving until the driving characteristics (F) no longer meet the terrain characteristics (GE), whereby a speed and / or steering of the vehicle (2) is reduced. [7] Method (1) according to any one of claims 1 to 6, characterized by , that the test of drivability for a rotation of the vehicle (2) about its center of gravity is carried out along the driving lines (3), wherein the driving lines (3) are arranged concentrically around the center of gravity of the vehicle as circular lines. [8] Device for carrying out the method (1) according to any one of claims 1 to 7, characterized by, that includes a control unit which evaluates the target points from the geometric map (GK) in the matrix to determine the terrain properties (GE).

Images