Method for creating probabilistic free space maps including static and dynamic objects - Patents.com

Abstract

To provide an improved, stochastically free method of forming a spatial map.SOLUTION: A method for forming a stochastically free spatial map includes the steps of: calling static objects 2a, 2b, 3 and a recognition area polygon WP from an existing peripheral model; collecting predicted orbits T1 and T2 of dynamic objects V1-V7; integrating the static object, the recognition area polygon and the predicted orbits into a single map; setting a longest predicted time; setting a predicted time step; setting a latest predicted time and setting a value of the latest predicted time to 0 in order to set a start time of a set predicted period; setting trust areas around the static and dynamic objects; setting at least one uncertain area around at least one static object or one dynamic object; and forming a first stochastically free spatial map for the latest predicted time.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a method for creating a probabilistic free space map containing static and dynamic objects involved in a reaction based on the establishment of a travel function. [Background technology]

[0002] From the prior art, for example, so-called "jam stop assistants" are known, which are based on a combination of adaptive cruise control (ACC) and lane keeping assistants. Furthermore, lane change assistants have recently been introduced, which must be actively activated or initiated by the driver. Reactions to traffic and lane information (objects) are performed either by measuring the relative distance to the object and controlling the gap (e.g., by ACC or lane keeping assistants) or by creating a grid map with information about the probability that a grid cell is occupied (e.g., a Dempster-Shafer grid map for automated parking).

[0003] However, conventional solutions (such as occupancy grid maps or lists of objects relative to the ego-vehicle) only provide information about the current state of free space. In other words, a particular area (e.g., a grid cell) can be marked as "measured free" / "measured occupied" or "unmeasured." The resulting up-to-date map does not include information about the predicted free / occupied space based on the future behavior of other traffic participants. To plan a driving strategy for a safe, comfortable, and efficient trip, it is necessary to know whether it is wise to move within a certain area in the future. Summary of the Invention [Problem to be solved by the invention]

[0004] It is therefore an object of the present invention to provide a method that overcomes the drawbacks of the prior art, allowing for the creation of improved free space maps. [Means for solving the problem]

[0005] According to the invention, this object is achieved by the subject matter set forth in independent claim 1. Further preferred embodiments are the subject matter of the dependent claims.

[0006] Essentially, the present invention aims to avoid uncertainty in periphery capture and traffic participant prediction. In autonomous driving, periphery capture always leaves behind areas of uncertainty in the periphery model. The problem for the driving functions connected to this function is that the driver will not tolerate the behavior that a system with poor prediction quality would expect, for example, an emergency intervention, such as the sudden recognition of a traffic participant changing lanes in an intrusive manner and the application of emergency braking, instead of a timely and gentle reaction. In other words, the driver would expect the system to gently slow down or change lanes in a timely manner, for example, since the driver would have detected such an intrusive lane change several seconds in advance. The goal of the present invention is to react to these "uncertain areas" early and comfortably, before only an emergency intervention is possible several seconds later. Drivers are more willing to accept a comfortable response that later turns out to be unnecessary (e.g., slowing down slightly without braking to allow for a cut-in lane change, regardless of whether an actual lane change will occur) than a late emergency brake based on a poor lane change prediction.

[0007] However, even if rare, in safety-critical cases an emergency response is justified, so objects with low existence probability cannot be ignored.

[0008] The present invention is based on the basic idea that the above mentioned problems can be avoided by creating a probabilistic free space map that contains the predicted behavior and the uncertainties involved in the prediction.

[0009] Thus, the method for creating a probabilistic free space map including static and dynamic objects according to the present invention comprises the following steps: - Calling static objects and recognition area polygons from an existing surrounding model; - collecting predicted trajectories of dynamic objects; - Integrating static objects, recognition area polygons and predicted trajectories into a single map; -Setting the maximum prediction time; - Step to set the forecast time step; - setting a latest forecast time and setting the value of the latest forecast time to 0 to set the start time of the set forecast period; - setting confidence regions around static and dynamic objects; - setting at least one uncertainty region around at least one static or dynamic object; - creating a first probabilistic free space map for the latest forecast time; generating at least one further free space map for at least one prediction time step; -Evaluating the created free space map

[0010] The surroundings model used here is generated from sensor data from at least two surroundings sensors, e.g., a radar sensor and a camera sensor, e.g., by sensor data integration methods. However, for static objects, a stored surroundings model, e.g., a semantic grid map, may be used, the corresponding data of which is retrieved based on the vehicle's self-tracking using GPS and / or landmark recognition. If a stored surroundings model is used, it is preferably updated at regular, predefined intervals so that any changes in the static objects can be reliably taken into account.

[0011] In the scope of this invention, the recognition area polygon refers to a polygon track around the vehicle, which means the latest 360° FOV (Field-of-view) at that time. Values ​​that affect this polygon include the sensor's reach and road geometry (curves, mountain passes, etc.).

[0012] When dynamic objects or other traffic participants are detected around the ego-vehicle, one or more trajectories are predicted for each dynamic object, based on, for example, the direction of movement, the speed, and / or the acceleration.

[0013] These above data are entered into a common map on which the subsequent method steps are based.

[0014] Furthermore, a maximum prediction time is set, which indicates the maximum possible trajectory of a dynamic object or other traffic participant that can be predicted. A prediction time step is also set. This is preferable because it allows the user to set how often the prediction is updated. That is, the maximum prediction time can be set to, for example, 10 seconds, and each prediction step can be performed every 0.5 seconds, for example.

[0015] To start the prediction, first, the latest prediction time is set to 0. From this point on, multiple predictions are performed within a predetermined prediction time step until the maximum prediction time is reached. Here, starting from the latest prediction time, one probabilistic free space map is created for each prediction time step. Each prediction time step is added to the latest prediction time until the maximum prediction time is reached. In addition to the latest prediction time, confidence regions are established around static and dynamic objects. These confidence regions describe the area for each grid cell that can be considered to be occupied by each object with a certain probability. Additionally, at least one uncertainty region is established around at least the dynamic object. However, it is also conceivable to establish at least one uncertainty region for multiple dynamic objects.

[0016] The uncertainty region here describes an area that may be occupied by a dynamic object in the future due to movement into this region. A probabilistic free space map is also evaluated, which is preferable because it can be adapted to subsequent navigation functions if necessary.

[0017] For example, a Bayesian network can be used to perform the prediction. The inputs to calculate the probability of a trajectory taken by a given dynamic object or traffic participant include the presence or absence of static or dynamic objects in the vicinity, the speed, acceleration, direction of movement of the traffic participant, and the road conditions. Furthermore, environmental factors such as weather can also be taken into account.

[0018] In one particularly preferred embodiment, a free space map is created for each prediction time step, up to the longest prediction time. This is advantageous because the periphery may change from one prediction time step to the next. In this way, it is possible to ensure that all possible changes are captured.

[0019] In a particularly preferred embodiment, at least one uncertainty region is established based on an existing surroundings model and a trajectory prediction of the dynamic object.

[0020] In another preferred embodiment, at least one uncertainty region is extended along at least one predicted trajectory of a dynamic object, and it is particularly preferred that the uncertainty region is extended in such a way that, at the longest prediction time, all foreseeable trajectories are at least partially covered by the uncertainty region, which is advantageous since predictions always have a certain degree of uncertainty that can be taken into account by the uncertainty region.

[0021] However, it is particularly preferred that the uncertainty region is adapted for each time step. As the prediction time increases, the uncertainty region also becomes larger, since the uncertainty increases along the predicted trajectory. For example, as the vehicle approaches a vehicle traveling ahead, the uncertainty also increases as the time and distance decrease, since the probability of a lane change increases. However, the vehicle may also slow down, which may also lead to uncertainty in the prediction.

[0022] It is also preferable that, after evaluating the probabilistic free space map for the entire time horizon, a trajectory of the ego-vehicle is planned that is optimized with respect to comfort, efficiency, and safety. In this way, in cases of high uncertainty, it is possible to proactively change lanes of the ego-vehicle to avoid sudden braking or steering interventions later. For this purpose, for example, a threshold for uncertainty can be set. If this threshold is exceeded, a preventive driving maneuver is implemented. Alternatively, the probability of occupancy of the uncertain region can be directly used in the cost function for optimization. Instead of a lane change, a preventive measure of deceleration of the ego-vehicle can also be considered.

[0023] It is particularly preferred to optimize at least one driving function and / or the trajectory of the vehicle using this cost function. Using this cost function, the trajectory of the vehicle can be optimized, for example, in terms of the target speed, longitudinal acceleration and jolt, lateral acceleration and jolt, distance to other traffic participants, achieving the required reaction to other traffic participants (e.g., sudden braking), and in terms of the probability of passing through the area occupied by other traffic participants in the future or the probability of performing an unpleasant intervention later. Alternatively or additionally, the optimization can also be performed in terms of the accuracy of lane keeping or the deviation of the vehicle's trajectory angle from the trajectory angle of the lane center.

[0024] In a further particularly preferred embodiment, the recognition area polygon is also taken into account for trajectory optimization. It should be noted that, for example, in left-hand curves on the (right-hand) autobahn, it is also preferable not to drive to the very left edge of the road, where the sensors cannot see far enough ahead. If the road edge suddenly turns out to be a missing tire, for example, a collision can only be avoided by an emergency maneuver. However, the assessment of areas covered by dynamic objects is different: when following a vehicle, the area in front of the vehicle may be covered, but it can be assumed that there is free space there for several seconds, since the other object is moving along the curve.

[0025] It is possible to output the direction, acceleration, speed and predicted trajectory in a global coordinate system, and it is also possible to convert these coordinates to a general coordinate system, i.e. it is also possible to preferably output the coordinates in the road coordinate system.

[0026] In particular, it is preferable that the confidence region be set based on the probability of the object's existence, the variance of the position vector and velocity vector, and the values ​​of the velocity vector and acceleration vector, which takes into account the velocity and acceleration, making it possible to set confidence regions not only for static objects but also for dynamic objects.

[0027] In a further preferred embodiment, the uncertainty region is set based on the trajectory probability, the values ​​of the variance of the position, velocity and acceleration of the last predicted trajectory points of the objects (V1-V7).

[0028] Furthermore, prior to the evaluation of the free space map, it is also preferable to combine grid cells with the same occupancy probability into regions separated from each other by corresponding polygon tracks. The occupancy probability in this case does not necessarily have to be a fixed value, but can be interpreted as a range containing a certain probability value. That is, the range can, for example, include all grid cells with an occupancy probability above 90%. In this way, for example, a reliability region around a traffic participant can be considered. The combination of grid cells into regions is also preferable, since combining them into regions reduces the computational power required for the evaluation.

[0029] Further configurations and embodiments are illustrated by the accompanying drawings. [Brief explanation of the drawings]

[0030] [Figure 1] 1 is a schematic depiction of a free space map according to an embodiment of the present invention; [Figure 2] 1 is a schematic depiction of an extended free space map according to an embodiment of the present invention; [Figure 3] 1 is a further schematic depiction of an extended free space map according to an embodiment of the present invention; [Figure 4] 1 is a further schematic depiction of an extended free space map according to an embodiment of the present invention; [Figure 5]1 is a further schematic representation of extended free space maps according to a further embodiment of the present invention; [Figure 6] 1 shows a schematic flow chart of an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0031] FIG. 1 shows a schematic representation of a free space map according to an embodiment of the present invention. In this representation, an ego-vehicle 1, together with several other traffic participants V1-V7, is moving on a road consisting of three lanes F1-F3. The ego-vehicle 1 is currently moving along its current trajectory Ta. In addition to the traffic participants V1-V7, lane boundaries 3, e.g., guardrails, and additional static objects 2a and 2b are also shown. At least two trajectories T1 and T2 are predicted for the traffic participants V1-V7. T1 represents the most probable trajectory. Trajectory T2 describes a possible trajectory with a lower probability. T1 is indicated by a solid arrow, while trajectory T2 is indicated by a dashed arrow. For clarity, trajectories T1 and T2 are labeled only once for each of the traffic participants V1-V7, rather than for all trajectories T1 and T2. The recognition area polygon WP is also shown. In the present invention, this recognition area polygon WP indicates a 360° field of view around the vehicle. The static objects 2a, 2b, and 3 and the recognition area polygon WP can be called from, for example, a surroundings model.

[0032] FIG. 2 shows a schematic representation of an extended free space map according to an embodiment of the present invention. In the representation shown, a reliability region K is defined around the static objects 2a, 2b, and 3 and the dynamic objects or traffic participants V1-V7. In this schematic representation, a rectangular reliability region K is assumed for each static object 2a, 2b, and 3 and each traffic participant V1-V7. In this representation, multiple reliability regions overlap. These reliability regions are assumed to be occupied in the free space map. Furthermore, an uncertainty region U is defined along the possible trajectories T1 and T2 of the traffic participant V1. This uncertainty region U may be occupied or remain free depending on the behavior of the traffic participant V1. Alternatively or additionally, a respective uncertainty region U for each of the other traffic participants V2-V7 is also defined. For example, FIG. 2 shows a time point corresponding to the first prediction time step. The uncertainty region U in this case only slightly overlaps the lane F2 of the ego vehicle 1. That is, since there is not yet sufficient need to perform a driving maneuver, the trajectory Te1 that the host vehicle 1 can take in the future corresponds to Ta, where the host vehicle 1 is currently traveling.

[0033] FIG. 3 shows a further schematic illustration of an expanded free space map according to an embodiment of the present invention. FIG. 3 corresponds to the content of FIG. 2. However, since this illustration is for a later prediction time step, the uncertainty region U is larger than that of FIG. 2. This is because, as time passes, V1 approaches V2, for example, and the possibility of a lane change increases, but this is not 100% certain because V1 may brake. Therefore, the uncertainty region U is larger. At the same time, another possible trajectory Te2 is planned from the ego-vehicle 1 along the current traveling direction in addition to the previous trajectory Te1. This further trajectory Te2 corresponds to a possible lane change to lane F3.

[0034] FIG. 4 shows a further schematic representation of an extended free space map according to an embodiment of the present invention. Again, the representation corresponds to that of FIGS. 2 and 3. FIG. 4 describes a later prediction time step than FIG. 3. Therefore, the uncertainty region U is also enlarged compared to FIGS. 2 and 3 and already extends to the full width of the lane F2 of the ego-vehicle 1. The currently planned trajectory Te2 of the ego-vehicle 1 assumes a lane change in order to avoid a possible lane change by V1. This allows for a timely and gentle driving maneuver to be performed, without requiring steering or braking interventions that would be uncomfortable for the driver of the ego-vehicle 1.

[0035] 5 shows a further schematic illustration of an expanded free space map according to a further embodiment of the present invention. In this illustration, grid cells with the same occupancy probability B are merged into corresponding regions 4. The regions 4 are each separated by a polygon track P. As shown in the figure, the regions include a region 4 with an occupancy probability B less than 0.1, a region 4 with an occupancy probability B less than 10%, a region 4 with an occupancy probability B greater than 50%, and a region 4 with an occupancy probability B greater than 90%. The region 4 with an occupancy probability greater than 90% can be, for example, a grid cell in which a traffic participant has been detected and a confidence region K around the traffic participant. Furthermore, an uncertainty region U spanning multiple regions 4 is also shown.

[0036] FIG. 6 shows a schematic flowchart of an embodiment of the present invention. In step S1, the static objects 2a, 2b, and 3 and the recognition region polygon WP are retrieved from an existing surroundings model. In step S2, the predicted trajectories T1 and T2 of the dynamic objects V1-V7 are collected. In the subsequent step S3, the static objects 2a, 2b, and 3, the recognition region polygon WP, and the predicted trajectories T1 and T2 are integrated into a first free space map. In step S4, a maximum prediction time is set, and in step S5, which may be performed in parallel with step S4, a corresponding prediction time step is set. Then, in step S6, the latest prediction time is set, and the value of the latest prediction time is set to 0 to set the start of the set prediction period. In step S7, a confidence region K is set around the static objects 2a, 2b, and 3 and the dynamic objects V1-V7. Furthermore, in step S8, at least one uncertainty region U is set around at least one of the static objects 2a, 2b, and 3 and the dynamic objects V1-V7. In step S9, a first probabilistic free space map is created for the latest predicted time. In step S10, at least one further free space map is created for one predicted time step. This step S10 is repeated until the longest predicted time is reached. In a further step S11, the created free space maps, in particular the free space map created at the time when the longest predicted time is reached, are evaluated. This application relates to the invention described in the claims, but the disclosure of this application also includes the following: 1. A method for creating a probabilistic free space map including static (2a, 2b, 3) and dynamic objects (V1-V7), characterized in that it comprises the following steps: - Step (S1) of calling static objects (2a, 2b, 3) and recognition area polygons (WP) from the existing surrounding model; - a step (S2) of collecting predicted trajectories (T1, T2) of dynamic objects (V1-V7); - A step (S3) of integrating the static objects (2a, 2b, 3), the recognition area polygon (WP) and the predicted trajectories (T1, T2) into a single map; - Step (S4) of setting the maximum prediction time; - A step of setting the forecast time step (S5); - setting a latest forecast time and setting the value of the latest forecast time to 0 to set the start point of the set forecast period (S6); - a step (S7) of setting a confidence region (K) around the static (2a, 2b, 3) and dynamic objects (V1-V7); - setting (S8) at least one uncertainty region (U) around at least one static (2a, 2b, 3) or dynamic object (V1-V7); - creating a first probabilistic free space map for the latest prediction time (S9); - generating at least one further free space map for at least one prediction time step (S10); - A step (S11) of evaluating the generated free space map. 2. 2. The method according to claim 1, wherein one free space map is generated for each prediction time step until the maximum prediction time is reached. 3. 2. The method according to claim 1, characterized in that at least one uncertainty region (U) is set based on an existing surrounding model and trajectory predictions of dynamic objects (V1-V7). 4. 4. The method according to any one of claims 1 to 3, characterized in that at least one uncertainty region (U) is extended along at least one predicted trajectory (T1, T2) of a dynamic object (V1-V7). 5. 5. A method according to any one of claims 1 to 4, characterized in that the uncertainty region (U) is adapted for each time step. 6. 6. The method according to any one of claims 1 to 5, characterized in that after evaluation of the probabilistic free space map for the entire period of interest, at least one driving function is adapted and / or a trajectory (Te1, Te2) of the host vehicle (1) optimized with respect to comfort, efficiency and safety is planned. 7. 7. A method according to any one of claims 1 to 6, characterized in that for planning the trajectory (Te1, Te2) of the vehicle (1), the recognition region polygon (WP) is also taken into account. 8. 7. The method according to claim 6, characterized in that at least one driving function and / or the trajectory (Te1, Te2) of the host vehicle (1) is optimized by a cost function. 9. 9. A method according to any one of 1 to 8 above, characterized in that the reliability region (K) is set based on the probability of existence of the objects (2a, 2b, 3, V1-V7), the variance of the position vectors and velocity vectors, and the values ​​of the velocity vectors and acceleration vectors. 10. 10. A method according to any one of claims 1 to 9, characterized in that the uncertainty zone is set based on the probability of the trajectory, the values ​​of the variance of the position, velocity and acceleration of the last predicted trajectory points of the objects (V1-V7). 11. 2. The method according to claim 1, characterized in that, prior to evaluation of the free space map, grid cells having the same occupancy probability are respectively merged into areas separated from each other by polygon tracks. [Explanation of symbols]

[0037] 1 EGO vehicle 2a,2b Static Objects 3 Lane Boundaries 4 areas B. Occupancy probability F1-F3 lanes K confidence region P Polygon Track S1-S11 method steps Ta Latest trajectory of the vehicle Te1, Te2 Possible alternative trajectories for the vehicle T1,T2 Possible trajectories of dynamic objects U Uncertain Area V1-V7 Dynamic Objects / Traffic Participants WP Recognition Area Polygon

Claims

1. A method for creating a probabilistic free space map including static (2a, 2b, 3) and dynamic objects (V1-V7), characterized in that it comprises the following steps: - A step (S1) of calling static objects (2a, 2b, 3) and a recognition region polygon (WP) from an existing surrounding model; - a step (S2) of collecting predicted trajectories (T1, T2) of dynamic objects (V1-V7); - A step (S3) of integrating the static objects (2a, 2b, 3), the recognition area polygon (WP) and the predicted trajectories (T1, T2) into a single map; - setting a maximum predicted time (S4); - setting the prediction time step (S5); - setting a latest forecast time and setting the value of the latest forecast time to 0 to set the start point of the set forecast period (S6); - setting (S7) confidence regions (K) around the static (2a, 2b, 3) and dynamic objects (V1-V7) based on the probability of presence of the objects (2a, 2b, 3, V1-V7), the variance of their position and velocity vectors, and the values ​​of their velocity and acceleration vectors; - setting (S8) at least one uncertainty region (U) around at least one static (2a, 2b, 3) and / or dynamic object (V1-V7); - creating a first probabilistic free space map for the latest prediction time (S9); - creating at least one further free space map for at least one prediction time step (S10); - A step of evaluating the created free space map (S11) The method according to claim 1, characterized in that after evaluation of the probabilistic free space map for the entire period of interest, at least one driving function is adapted and / or a trajectory (Te1, Te2) of the host vehicle (1) optimized with respect to comfort, efficiency and safety is planned.

2. 2. The method of claim 1, wherein one free space map is generated for each prediction time step until a maximum prediction time is reached.

3. 2. The method according to claim 1, characterized in that at least one uncertainty region (U) is set based on existing surrounding models and trajectory predictions of dynamic objects (V1-V7).

4. 4. A method according to claim 1, characterized in that at least one uncertainty region (U) is extended along at least one predicted trajectory (T1, T2) of a dynamic object (V1-V7).

5. 5. Method according to any one of claims 1 to 4, characterized in that the uncertainty region (U) is adapted for each time step.

6. 6. The method according to claim 1, further comprising the step of: planning the trajectory (Tel, Te2) of the ego-vehicle (1) by taking into account the recognition region polygon (WP).

7. 6. The method according to claim 1, further comprising optimizing at least one driving function and / or a trajectory (Te1, Te2) of the host vehicle (1) by means of a cost function.

8. The method according to any one of claims 1 to 7, characterized in that the uncertainty zone is set based on the probability of the trajectory, the values ​​of the variance of the position, velocity and acceleration of the last predicted trajectory points of the objects (V1-V7).

9. 2. The method of claim 1, further comprising the steps of: prior to evaluation of the free space map, merging grid cells having the same occupancy probability into regions bounded by polygon tracks from each other.

Images