Method and device for operating a motor vehicle based on the recognition of rapidly movable objects

Abstract

The present invention relates to a method for operating a mobile agent (1) based on the recognition of fast-movable environmental objects, in particular for performing a collision avoidance function, having the following steps: detecting (S1) the positions of one or more movable objects (2) in the environment (U) of the mobile agent (1); assigning (S2) the detected environmental objects (2) to a cell grid according to their positions, the cell grid corresponding to a 2D polar coordinate grid having a cell system extending annularly around the mobile agent (1), wherein the number of cells (Z) in the radial direction decreases as the distance from the mobile agent (1) increases; propagating (S2) the environmental objects assigned to the cells (Z) via the cells (Z) in order to obtain radial velocity data, in particular average radial velocity, for each cell (Z); signaling (S5) the collision probability for each cell (Z) according to the radial velocity data (v rad ).

Description

Field of the Invention

[0001] The present invention relates to functions of a vehicle that are based on the recognition of fast-movable objects in the vehicle environment, in particular functions for collision avoidance with these objects. Background Art

[0002] Vehicles or other mobile agents track static and slow-moving objects in the environment and use them, for example, for collision warning functions or the like, while separate environmental monitoring is carried out for fast-approaching objects for running collision avoidance functions, so-called safety protection functions.

[0003] A core requirement for such collision avoidance functions is a rapid reaction to fast-approaching dynamic objects in the environment of the mobile agent. If a possible collision is recognized, the collision avoidance function triggers a corresponding reaction of the mobile agent, in particular automatic deceleration or automatic stopping.

[0004] A method for detecting objects in the vicinity of a motor vehicle is known from document US 9,280,890 B2. Here, the distance to the object is detected and a safety zone around each detected object is defined based on this distance and an insecurity factor regarding the positional insecurity of the object. A grid around the vehicle is established and the potential interaction between the object and the vehicle is determined based on its safety zone, so that a driving intervention can be carried out based on the potential interaction. Summary of the Invention

[0005] According to the invention, there is provided a method for operating a motor vehicle, in particular for performing a collision avoidance function, based on the recognition of fast-movable objects, as well as a device and a vehicle for performing the method.

[0006] Further design options are given in the dependent claims.

[0007] According to a first aspect, there is provided a method for operating a mobile agent based on the recognition of fast-movable environmental objects, in particular for performing a collision avoidance function, the method having the following steps:

[0008] - Detecting the positions of one or more movable environmental objects in the environment of the mobile agent;

[0009] - Assigning the detected environmental objects to a cell grid according to their positions, wherein the cell grid corresponds to a 2D polar coordinate grid having a cell system extending annularly around the mobile agent, and wherein the number of cells in the radial direction decreases as the distance from the mobile agent increases;

[0010] - Propagating the environmental objects assigned to the cells via the cells in order to obtain radial velocity data, in particular average radial velocity, for each cell,

[0011] - Transmit the collision possibility for each cell based on the radial velocity data.

[0012] A mobile agent, such as a self - propelled robot or a motor vehicle, has an environmental sensing device to detect objects in the environment. The aim is to identify static and dynamic objects that may interfere with the movement of the mobile agent or the planned trajectory, so that these objects can be taken into account accordingly. In particular, objects moving rapidly towards the mobile agent must be identified with high recognition accuracy as early as possible and a rapid reaction must be triggered. This function is also referred to as a safety function.

[0013] For this purpose, the above - mentioned method is set up such that the environment of the mobile agent is formed by means of a 2D polar - coordinate cell grid, which has cells that are shown or positioned in polar coordinates. Here, the mobile agent is arranged in the middle, that is, at the origin of the polar - coordinate grid.

[0014] In particular, for radar sensors, it is known to identify objects in the environment with respect to direction and distance, that is, in polar coordinates, so that the detected objects can be assigned to a cell in the 2D polar - coordinate grid in a particularly simple manner.

[0015] By means of a suitable evaluation method, the occupancy of the cells can now be known and evaluated within successive time steps in order to identify possible collisions between the detected objects and the mobile agent.

[0016] The cell grid has a 2D polar - coordinate lattice with a cell system that extends annularly around the mobile agent, where the number of cells in the radial direction decreases as the distance from the mobile agent increases. Based on the increasing inaccuracy of the positioning due to the dynamic effects with increasing distance from the mobile agent, the size of the angular range detected by the cells can be larger as the distance increases without significantly affecting the recognition accuracy for possible collisions.

[0017] According to an evaluation method, each of these cells is assigned an occupancy probability with respect to the detected dynamic objects and based on the environmental detection sensing device.

[0018] When an object approaches along the radial direction at a speed faster than a pre - given speed, a collision can already be identified by using a 2D polar - coordinate grid with simplified cells when the spacing is greater than a critical spacing. Considering only the radial component of the speed significantly simplifies the calculation and thus enables very rapid identification of very fast objects approaching the mobile agent.

[0019] Furthermore, for distances greater than a pre - given critical spacing, the cell grid can be not allocated in the tangential direction, so that the cells involved extend annularly around the mobile agent.

[0020] In particular, for distances less than the critical spacing, tangential velocity data, in particular the average tangential velocity, can be taken into account when determining the likelihood of a collision. In the range of spacings less than the critical spacing, the tangential component of the velocity is accordingly important, since a collision can thereby possibly be excluded, even though the object approaches the mobile agent with respect to the radial velocity component.

[0021] It can be arranged that the likelihood of a collision is determined by means of an occupancy grid and a Bayesian occupancy filter in a polar coordinate system in combination with a particle filter.

[0022] Furthermore, a collision warning function can be implemented for each cell of the cell grid on the basis of the occupancy likelihood.

[0023] According to another aspect, there is provided a device for operating a mobile agent based on the recognition of fast-movable environmental objects, in particular for performing a collision avoidance function, wherein the device is configured to:

[0024] - receive the positions of one or more detected movable environmental objects in the environment of the mobile agent;

[0025] - assign the detected environmental objects to a cell grid according to their positions, wherein the cell grid corresponds to a 2D polar coordinate grid having a cell system extending annularly around the mobile agent, and wherein the number of cells in the radial direction decreases as the distance from the mobile agent increases;

[0026] - propagate the environmental objects assigned to the cells via the cells in order to obtain radial velocity data, in particular the average radial velocity, for each cell;

[0027] - signal the likelihood of a collision for each cell based on the radial velocity data.

[0028] According to another aspect, there is provided a mobile agent having: a motion actuator for moving the mobile agent; an environmental detection sensing device for detecting the positions of one or more movable environmental objects in the environment of the mobile agent; and the device as described above. Description of the Drawings

[0029] The embodiments will subsequently be explained in detail with reference to the drawings. Among them:

[0030] Figure 1 shows a mobile agent with dynamic objects in its environment;

[0031] Figure 2 shows a flow chart for explaining a method for performing a collision recognition function in a mobile agent; and

[0032] Figure 3 A schematic diagram showing the division of the environment of a mobile agent in cells, which can be addressed via polar coordinates. Detailed implementation

[0033] Figure 1 A schematic diagram showing a mobile agent 1, with an environment U in which static and dynamic objects 2 can be located.

[0034] The mobile agent 1 has a control unit 11 to control the mobile agent 1. In addition, the mobile agent 1 has an environment detection sensing device 12 to detect the direction and distance of surrounding objects 2 with respect to the agent coordinate system A. In addition, the mobile agent 1 has a motion actuator 13 to move the mobile agent 1 inside the environment U corresponding to the function to be implemented, especially corresponding to the function for autonomous movement to a pre-given target location.

[0035] The environment detection sensing device 12 can have at least one of the following sensor devices: lidar device, radar device, laser scanning device, camera, inertial sensor, odometry sensor, etc. The sensor data detected by the environment detection sensing device 12 is processed in the control unit 11 depending on the sensor device used in the preprocessing, such that the orientation and distance of the object 2 can be known and provided with respect to the agent coordinate system A. The origin of the agent coordinate system A relates to the position of the mobile agent 1 and enables the description of the relative position of the environmental object 2 with respect to the position of the mobile agent 1.

[0036] To identify an object 2 approaching the mobile agent 1, a method as detailed in the flowchart in conjunction with Figure 2 is implemented in the control unit 11. This method can be executed as a software and / or hardware algorithm in the control unit 11.

[0037] As Figure 3 schematically shown, the environment of the mobile agent 1 is shown in polar coordinates with an origin P for the collision recognition function, such that the position of the dynamic object 2 in the environment of the mobile agent 1 can be located by means of polar coordinates, and the mobile agent 1 is located at the origin P of the polar coordinates. The environment of the mobile agent 1 is assigned to cells Z for this purpose.

[0038] The cell Z is defined as the range between two radial distances of the mobile agent 1 and between two azimuth angles, as long as the larger distance of the cell edge is less than the critical distance R cr . For the range of the radial distance with a larger distance greater than the critical distance R cr , the cell Z is only defined by the distance range between the two distances, but not bounded by the azimuth angle. Thus, the cell extends beyond the range of the critical distance around the mobile agent 1 in a ring shape.

[0039] To identify a collision risk, the BOF method (BOF: "Bayesian Occupancy Filter" in English, "Bayes‘scher Belegungsfilter" in German) can now be implemented in the control unit 11 of the mobile agent 1.

[0040] First, in step S1, dynamic objects 2 in the environment of the mobile agent 1 are detected by means of the environmental detection sensor device 12. These objects are described by their position o in polar coordinates and are subsequently assigned to the cells of a 2D polar coordinate grid as shown, for example, Figure 3 in. The 2D polar coordinate grid represents the environment of the mobile agent 1 by simply mapping the environmental area onto the surface covered by the 2D polar coordinate grid. Below, the area in the environment of the mobile agent 1 represented by the cell Z in question is understood as the cell area.

[0041] The method described below is based on the Bayesian occupancy filter in combination with the particle filter method in order to obtain the collision probability in step S2.

[0042] The method of the Bayesian occupancy filter consists in estimating the map posterior p(v t |z 1：t , c 1：t ) and p(o t |z 1：t , c 1：t ), where, in the cells of the 2D grid, v t corresponds to the velocity at time step t, ot corresponds to the position at time step t, z 1：t corresponds to the measurements from time step 1:t; and c 1：t corresponds to the cell state for time points 1:t.

[0043] During a single time point, the cell states are independent of each other, so that

[0044]

[0045] However, when a dynamic object moves in the environment of the mobile agent such that the object is assigned to multiple cells of the 2D polar coordinate grid in its movement direction, the cell states are no longer independent of each other.

[0046] To calculate p(v t，i |z 1：t , c 1：t ) from the previous estimate, assume:

[0047] p(v t，i |z 1:t , c 1:t ) = ∫p(v t，i|v t-1 ,z t-1 ,c 1:t )p(v t-1 |z 1:t-1 ,c 1:t-1 )dv t-1

[0048] When the above formula is implemented recursively and periodically, it is important for the ongoing estimation to compute only the states of previous time steps.

[0049] Now, the Bayesian occupancy filter method (BOF) can be applied to individual cells of a polar grid, as disclosed, for example, in "Bayesian Occupancy Filtering for Multi-Target Tracking: An Automotive Application" by C. Coué et al., International Journal of Robotics Research, SAGE Publications, 2006, 25, pp. 19 - 30 and "Efficient Formulation of Bayesian Occupancy Filters for Target Tracking in Dynamic Environments" by M.K. Tay et al., International Journal of Vehicle Autonomous Systems, Volume 6, ISSN: 1471 - 0226, in combination with a particle filter, as known, for example, from "Grid-Based Mapping and Tracking in Dynamic Environments Using a Unified Evidence Environment Representation" by G. Tanzmeister et al., IEEE International Conference on Robotics and Automation 2014.

[0050] However, according to the above method, the BOF method is applied in a 2D polar grid. Each cell in the 2D polar grid has a particle and an individual filter is assigned to each cell. If a particle moves between cells of the polar grid, then the particle is assigned to the corresponding individual particle filter.

[0051] As known from the prior art mentioned above, in order to implement a particle filter, the map can be shown as

[0052]

[0053] where, w [k] corresponds to a weighting coefficient and corresponds to the Dirac function. Thereby, particles in cell Z can be generated. The distribution of velocities follows the weighting coefficient.

[0054] With the help of sensor measurements, cell motion likelihoods and cell velocity likelihoods can be generated from the following measurement parameters:

[0055]

[0056] where, x i is assumed to be the midpoint of cell Z (with label i) in polar coordinates, and z m,iIt is assumed that the dynamic object 2 has a positioning with the shortest distance relative to the current cell Z. O corresponds to the occupancy probability, V corresponds to the speed, C corresponds to the position of the current cell, and σ corresponds to the variance.

[0057] If multiple measurement methods are applied, the sensor parameters can be simply merged by combining the individual values p i (z|O, V, C). For example, it can be simply assumed that the maximum value of these values is taken.

[0058] Particles are generated for each cell. These particles have 1D positions and speeds. The number of particles is proportional to p i (z|O, V, C) and is calculated as n i = n max p i (z|O, V, C), where n max corresponds to a pre-given number of the maximum particles per cell. This ensures that the total number of particles is not too large.

[0059] In the case of a radar sensor, a measurement of the radial velocity is additionally obtained. From this, a particle sampling of the velocity can additionally be obtained. The sampling and resampling can be implemented as described in "Grid-based Mapping and Tracking in Dynamic Environments using a Unified Evidential Environment Representation" by G. Tanzmeister et al., IEEE International Conference on Robotics and Automation 2014. The conditional probability is calculated as

[0060]

[0061] where z i，vr corresponds to the measurement of the radial velocity and σ vr corresponds to the velocity variance along the radial direction. Through the above formula, the resampling of the particles can be performed, whereby the particles closer to the center point of cell Z are weighted more strongly.

[0062] In a lidar sensor, there is no explicit measurement of the radial velocity, so the initial velocity distribution is distributed equally.

[0063] in the above formula is related to the radial velocity. The greater the radial velocity, the smaller. Next, the particles are considered individually, where a constant velocity is assumed. That is, in the case of assuming a constant velocity, the propagation of the particles is assumed as follows:

[0064]

[0065] where r corresponds to the radial vector, r corresponds to the radial interval, Corresponding to the unit vector in the radial direction, θ corresponds to the azimuthal value of the polar coordinates, Corresponding to the unit vector perpendicular to the radial direction. The propagation is performed in polar coordinates by the simple equation r k+1 = r k + vdt.

[0066] As long as the spacing is greater than a pre-given critical spacing R cr , then the polar coordinate grid can be assumed to be one-dimensional and the azimuthal angle can be ignored. The particle filter then learns a much smaller number of particles than in the 2D case.

[0067] For each cell, the number of required particles is calculated based on the measurement. When this number is greater than the number of required particles in the cell, the number of particles is increased, and in the opposite case, the number is decreased. The maximum number of particles in each cell is determined, for example, between 10 and 100, preferably 30. The particles corresponding to the true object velocity are retained and given a higher weight.

[0068] For regions with a spacing less than the pre-given critical spacing R rc , the risk of collision is high, and the tangential component of the 2D polar coordinates must be considered. In this case, the particle filter is applied on a 2D polar coordinate grid. This distinction is allowed because in the case of objects at a great distance, only the distance to the movable object is important. However, when the object approaches the mobile agent to a spacing less than the critical spacing R rc , the polar coordinate grid is assumed two-dimensionally so that the tangential component of the velocity is also estimated. In these cases, the average velocity of all particles represents the cell velocity. After estimating the velocities of these individual particles, the following values can be introduced:

[0069]

[0070] where p cell，vel is calculated for the radial component (vrad) and the tangential component (vtan) of the velocity and σ vel illustrates the difference in the radial velocities of all the particles in a cell, and σ max illustrates the maximum value of this difference. Then, the time t col l until a possible collision can be approximately calculated by an equation in step S3.

[0071]

[0072] where d corresponds to the spacing between the cell and the mobile agent 1, and t col l corresponds to the estimated time until the collision.

[0073] By considering the tangential velocity for the known time until a possible collision according to the size vehicleDim of the mobile agent 1, it can be determined in step S4 whether the environmental object 2 hits the mobile agent 1.

[0074] v tan *t col <vehicleDim

[0075] If it is determined in step S4 (option: yes) that the mobile agent 1 is hit by the environmental object 2, then a collision warning is signaled in step S5. Otherwise, it jumps back to step S1.

[0076] Thus, a collision warning can be signaled in step S5 by observing the p cell,vel value, its velocity, its acceleration value, and the acceleration range of the adjacent cells.

[0077] According to the signaled collision warning, the mobile agent 1 can correspondingly activate a collision avoidance function in step S6. The collision avoidance function can include, for example, automatic forced braking, avoidance, or acceleration of the mobile agent 1.

Claims

1. A method for operating a mobile agent (1) based on the recognition of fast-moving environmental objects, the method having the following steps: - Detecting (S1) the positions of one or more moving environmental objects (2) in the environment (U) of the mobile agent (1); - Assigning (S2) the detected environmental objects (2) to a cell grid according to their positions, wherein, the cell grid corresponds to a 2D polar coordinate grid having a cell system extending annularly around the mobile agent (1), wherein the number of cells (Z) in the radial direction decreases as the distance from the mobile agent (1) increases; - Propagating (S3) the environmental objects assigned to the cell (Z) via the cell (Z) in order to obtain radial velocity data for each cell (Z), - Based on the radial velocity data (v rad ), the likelihood of a collision is signaled (S5) for each cell (Z).

2. The method according to claim 1, wherein, the method is used to perform a collision avoidance function.

3. The method according to claim 1, wherein, propagating (S3) the environmental objects assigned to the cell (Z) via the cell (Z) in order to obtain an average radial velocity for each cell (Z).

4. The method according to claim 1, wherein, For distances greater than a pre-given critical spacing (R cr ), no allocation of the unit cells in the tangential direction is performed, such that the cells (Z) involved extend annularly around the mobile agent (1).

5. The method according to claim 4, wherein, Consider the tangential velocity data (v cr ) when determining the likelihood of a collision for distances less than the critical spacing (R tan ).

6. The method according to claim 5, wherein, The average tangential velocity is considered when determining the likelihood of the collision for distances less than the critical spacing (R cr ).

7. The method according to any one of claims 1 to 6, wherein, the collision possibility is determined by combining a occupancy grid in a polar coordinate system and a Bayesian occupancy filter with a particle filter.

8. The method according to any one of claims 1 to 6, wherein, a collision warning function is implemented for each cell of the cell grid according to the occupancy possibility.

9. A device for operating a mobile agent (1) based on the recognition of fast-moving environmental objects, wherein, the device is configured to: - Receive the positions of one or more detected moving environmental objects (2) in the environment (U) of the mobile agent (1); - Assign the detected environmental objects (2) to a cell grid according to their positions, wherein the cell grid corresponds to a 2D polar coordinate grid having a cell system extending annularly around the mobile agent (1), wherein the number of cells (Z) in the radial direction decreases as the distance from the mobile agent (1) increases; - Propagate the environmental objects assigned to the cell (Z) via the cell (Z) in order to obtain radial velocity data for each cell (Z), - Based on the radial velocity data (v rad ), the likelihood of a collision is signaled for each cell (Z).

10. The device according to claim 9, wherein, the device is used to perform a collision avoidance function.

11. The device according to claim 9 or 10, wherein, propagating the environmental objects assigned to the cell (Z) via the cell (Z) in order to obtain an average radial velocity for each cell (Z).

12. A mobile agent (1), comprising: a motion actuator for moving the mobile agent (1); an environmental detection sensing device (12) for detecting the positions of one or more moving environmental objects (2) in the environment (U) of the mobile agent (1); and a device according to any one of claims 9 to 11.

13. A computer program product comprising a computer program which is configured to perform the method according to any one of claims 1 to 8 when the computer program is executed on a computing unit.

14. A machine-readable storage medium having stored thereon a computer program which is configured to perform the method according to any one of claims 1 to 8 when the computer program is executed on a computing unit.

Images