Method and apparatus for determining a vehicle's trajectory - Patents.com

By estimating vehicle trajectories with improved accuracy using sensor data from regular vehicles, the method optimizes HD maps, addressing the issue of outdated data and enhancing autonomous driving capabilities.

JP7588138B2Active Publication Date: 2024-11-21BAYERISCHE MOTOREN WERKE AG
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
JP2022513504
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2019-09-03
Filing Date
2020-07-28
Publication Date
2024-11-21
Estimated Expiration
2040-07-28

AI Technical Summary

Technical Problem

The accuracy and resolution of sensor data from regular vehicles are lower than those of environmental mapping vehicles, leading to outdated HD maps that compromise autonomous driving capabilities due to infrequent updates and unrecognized changes in the road network.

Method used

A method and device that estimate the vehicle's actual trajectory using sensor data from regular vehicles, combining measured position and odometry values to optimize HD maps by minimizing measurement errors through graph-based SLAM, incorporating systematic and local error corrections.

Benefits of technology

Accurately determines vehicle trajectories with high precision, enabling efficient optimization of HD maps and enhancing autonomous driving functions by correcting for systematic and local measurement errors.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007588138000007
    Figure 0007588138000007
  • Figure 0007588138000008
    Figure 0007588138000008
  • Figure 0007588138000009
    Figure 0007588138000009
Patent Text Reader

Abstract

An apparatus for estimating an actual trajectory of a vehicle is described. The apparatus is configured to determine a series of measured position values ​​at a corresponding series of time instants using a position sensor. The apparatus is further configured to determine a series of odometry values ​​based on sensor data from one or more vehicle sensors. The apparatus is further configured to determine a series of estimated pose values ​​for the series of time instants and a systematic position offset for the series of measured position values ​​to refine, and in particular, at least locally optimize, an optimization criterion as an estimate of the actual trajectory of the vehicle.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a method and a corresponding device for determining the trajectory traveled by a vehicle, in particular for generating high definition digital (HD) maps of road networks. [Background technology]

[0002] Vehicles can extend and / or complement their vehicle sensors with high definition digital maps (HD maps) for safe, comfortable and efficient automated driving. HD maps provide a priori information about the vehicle's surroundings, which allows for improved vehicle localization and / or detection of anomalies in the vehicle's sensors. Moreover, HD maps can act as a fallback in case of sensor failure and improve the usability of automated driving functions by looking further ahead. Furthermore, HD maps contain information about lane availability and connectivity as well as valid traffic rules, through which automated driving is improved.

[0003] To create the HD map, dedicated environmental mapping vehicles equipped with various high-precision sensors, such as LIDAR (Light detection and ranging) and / or high-precision GPS receivers, can be used. However, using dedicated environmental mapping vehicles entails a relatively high cost. Furthermore, the limited number of environmental mapping vehicles means that only a relatively small part of the road network can be driven regularly and / or individual road segments are driven relatively infrequently. As a result, changes in the road network may not always be recognized in a timely manner and may not be reflected in the HD map. Using an HD map that is not sufficiently up-to-date may impair the quality of the autonomous driving function of the vehicle.

[0004] In order to reduce the effort involved in verifying and / or optimizing the HD map and / or to increase the frequency of verifying and / or optimizing the HD map, (regular) vehicle driving data can be acquired and analyzed. The vehicle driving data can include sensor data of various sensors of the vehicle. Based on the driving data of each individual vehicle, the trajectory of each individual vehicle can be determined, from which the course of the driving path can be derived. Thus, the vehicle driving data can be used to optimize the HD map. Summary of the Invention [Problem to be solved by the invention]

[0005] The accuracy and / or resolution of the sensor data of one or more sensors of the vehicle is usually lower than the accuracy and / or resolution of the sensor data from the sensors of the environmental mapping vehicle. This document is concerned with the technical problem of being able to determine the trajectory of the vehicle with as high an accuracy as possible even when using sensor data of relatively low accuracy, so that in particular the determined trajectory can be used for the creation and / or optimization of a digital map. [Means for solving the problem]

[0006] This problem is solved by each of the independent claims. Advantageous embodiments are especially described in the dependent claims. It is to be pointed out that additional features of claims dependent on an independent patent claim, even without the features of the independent patent claim or only in combination with some of the features of the independent patent claim, can form an independent invention independent of the combination of all the features of the independent patent claim, which can be the subject of an independent patent claim, a divisional application or a subsequent application. This also applies to technical teachings given in the description, which can form an invention independent of the features of the independent patent claims.

[0007] According to one aspect, an apparatus for estimating an actual (driving) trajectory of a vehicle is described. The trajectory can then be estimated based on sensor data of one or more sensors of the vehicle. The apparatus can be located at least partially (or entirely) inside the vehicle and / or at least partially (or entirely) outside the vehicle (e.g., in a back-end server). The apparatus can be configured to create and / or optimize a digital map for a road network based on the estimation of the actual trajectory of the vehicle. The digital map can then be utilized to provide automated driving functionality (especially for highly automated driving). The digital map can describe the course of roads and / or paths and / or lanes of the road network relative to a reference coordinate system (e.g., relative to a global coordinate system).

[0008] The device may be configured to determine a sequence of measured position values ​​for each position of the vehicle at a corresponding sequence of times based on sensor data of the (vehicle) position sensor. The measured position values ​​may here include coordinates, in particular GPS coordinates, for the position of the vehicle in a reference coordinate system for the road network (e.g. in a global coordinate system). The sequence of measured position values ​​may thus describe a measured trajectory of the vehicle in the reference coordinate system. Here, measurement errors and / or measurement inaccuracies may result in stochastic and / or systematic deviations of the individual measured position values ​​from the actual respective positions of the vehicle.

[0009] The device is further configured to determine a series of odometry values ​​for at least a portion of said series of time instants based on sensor data of one or more vehicle sensors (of the vehicle). The odometry values ​​for a time instant now indicate how the pose (i.e. the orientation and / or position of the vehicle) has changed compared to a (vehicle) previous time instant. Exemplary vehicle sensors indicate the vehicle's road speed, steering angle, acceleration and / or yaw rate. The pose values ​​may indicate the pose of the vehicle. To this end, the pose values ​​may include coordinates relating to the position of the vehicle in the vehicle's coordinate system. The odometry values ​​may indicate the change in the position (and / or orientation) of the vehicle relative to the vehicle's coordinate system between two time instants. The vehicle's coordinate system may then be misaligned with respect to the reference coordinate system. The odometry values ​​may thus indicate how the position (and possibly the orientation) of the vehicle has changed between two time instants (relative to the vehicle's coordinate system) in said series of time instants, respectively.

[0010] As a result, a set of pose values ​​can be determined based on these odometry values, which set of pose values ​​represents a trajectory that is offset (and possibly rotated) relative to the actual trajectory of the vehicle. Furthermore, these odometry values ​​may have errors (due to measurement inaccuracies of one or more vehicle sensors), so that a pose value determined based only on these odometry values ​​(in addition to translation and / or rotation) will typically differ from the actual pose value of the vehicle.

[0011] The apparatus may be configured to combine and / or superimpose said set of measured position values ​​and said set of odometry values ​​to determine an estimate of the actual orbit, whereby a (systematic) position offset may be taken into account to uniformly shift the measured position values, thereby taking into account and offsetting systematic errors (in particular of the measured GPS or satellite coordinates) in measuring the measured position values.

[0012] In particular, the apparatus may be configured to determine a set of estimated pose values ​​for said set of time instants as an estimate of the actual trajectory of the vehicle. The estimated pose value for a particular time instant may then indicate the pose of the vehicle at that particular time instant relative to a reference coordinate system. Furthermore, a (uniform) value of a position offset for said set of measured position values ​​may be determined. The set of estimated pose values ​​and position offset values ​​may then be determined to improve an optimization criterion, in particular to optimize at least locally.

[0013] The optimization criterion may include one position value deviation term for each of the sequence of time instants. The position value deviation term for a time instant may depend on the (e.g. quadratic) deviation of the estimated pose value for the respective time instant from the measured position value for the respective time instant, shifted by a (uniform and / or systematic) position offset. In this way, the sequence of estimated pose values ​​(and thus the orbit estimate) may be determined with particular accuracy.

[0014] Alternatively or additionally, the optimization criterion may include an odometry deviation term for at least some of the time instants in the series (in particular for N-1 time instants when the series includes N time instants). The odometry deviation term for a time instant may then depend on the deviation (possibly to a second order) of the change in pose of the vehicle at that time instant from the series of estimated pose values ​​from the odometry value for that time instant.

[0015] An optimization criterion can therefore be provided for determining said set of estimated pose values ​​such that the (median) deviation of the estimated pose values ​​relative to the measured odometry values ​​and relative to the measured position values ​​(shifted by the position offset) is reduced, in particular minimized. In this way, the actual trajectory of the vehicle can be estimated in a particularly accurate manner.

[0016] The device can be configured to determine an estimate of the actual trajectory of the vehicle by graph-based SLAM, i.e. by simultaneous self-localization and environmental mapping, where the position offset can then be considered in a preferred embodiment as an anchor node that can be shifted as part of the graph-based SLAM method, resulting in a particularly efficient and accurate estimation of the actual trajectory.

[0017] As already mentioned above, the measured position values ​​usually have a certain position measurement uncertainty. As a result, for each measured position value there may be a probability distribution of possible values ​​around the measured position value. The optimization criterion may depend on the position measurement uncertainty (in particular on the probability distribution) of the measured position values. This dependence may then be such that an increase in the position measurement uncertainty (or an increase in the width of the probability distribution) leads to a decrease in the influence of the position value deviation term on the optimization criterion. In this way, the accuracy of the estimation of the actual trajectory can be further improved.

[0018] Correspondingly, the odometry values ​​also usually have an odometry measurement imprecision. As a result, for each measured odometry value, there may be a probability distribution of possible values ​​around the measured odometry value. The optimization criterion may depend on the odometry measurement imprecision (particularly on the probability distribution), in particular such that an increase in the odometry measurement imprecision (or an increase in the width of the probability distribution) leads to a decrease in the influence of the odometry deviation term on the optimization criterion. In this way, the accuracy of the estimation of the actual trajectory can be further improved.

[0019] The device may be configured to determine a position offset for said sequence of measured position values ​​such that a value of the position offset is the same for all measured position values ​​of said sequence and / or is the same for all time instants of said sequence. In other words, a common (systematic) position offset may be taken into account for all measured position values ​​of an orbit. In this way, systematic errors affecting the entire orbit (e.g. systematic errors due to the atmosphere) may be taken into account and offset in an efficient and accurate manner when measuring measured position values.

[0020] The device may be configured to limit the value of the position offset to a certain maximum value. Alternatively or additionally, the device may be configured to determine the value of the position offset taking into account the covariance and / or probability distribution of said position offset. Alternatively or additionally, the optimization criterion may have a term that is limited by increasing magnitude of the value of the position offset. In this way, an accurate localization of the estimate of the actual trajectory in the reference coordinate system (in particular in the global coordinate system) can be performed in an efficient and reliable manner.

[0021] The device may be configured to divide the sequence of times into a number of consecutive partial sequences. The measured position values ​​of one partial sequence may then be shifted by a partial position offset for that partial sequence and by the position offset for all sequences in order to determine an estimate of the actual trajectory. In particular, the device may be configured to determine the sequence of estimated pose values ​​for the sequence of times, the position offsets for the measured position values ​​of all sequences and a number of the partial position offsets for the partial sequences in order to improve an optimization criterion and in particular to optimize at least locally. By taking into account the partial position offsets for different partial regions of the vehicle's trajectory, locally affecting systematic errors (e.g. local multipath effects, etc.) can be taken into account and offset in an efficient way when measuring the measured position values. In this way, the quality of the estimate of the actual trajectory can be further improved.

[0022] As already mentioned above, the device can be configured to determine an estimate of the actual trajectory by a graph-based SLAM method. For this purpose, the device can be configured to create a graph, which has as nodes (referred to in the present document as pose variable nodes) the individual estimated pose values ​​of said sequence of estimated pose values. The estimated pose values ​​can then be initialized based on measured position values ​​and / or based on (measured) odometry values. This graph with the initialized estimated pose values ​​can then be used as a starting point for an (iterative) optimization of the graph, and thus for the estimation of the actual trajectory.

[0023] The graph may have odometry factors as edges between two pose variable nodes, which depend on sensor data from one or more vehicle sensors and / or indicate the vehicle motion between the two nodes. In particular, the odometry factor between a pose variable node for a particular time instant and a pose variable node for a previous time instant may depend on the measured odometry values ​​for that particular time instant. In this case, the odometry factor may indicate a probability distribution for the change in pose value between the two pose variable nodes.

[0024] The graph may further include a position variable node for each measured position value. The graph may further include a connection for each of the pose variable nodes to a corresponding position variable node (or to a corresponding measured position value) via a position factor (i.e., via an edge). The position factor may then indicate how the position variable node is positioned relative to each pose variable node. The position factor may then depend on the measured position value for each position variable node. In particular, the position variable node for a particular time instant may be positioned in a position in the graph that corresponds to the measured position value for that particular time instant. The position factor (i.e., edge) between a pose variable node and a corresponding position variable node may be associated with a probability distribution that depends on the inaccuracy of the position measurement.

[0025] Thus, the (initialized) pose variable nodes of the graph can be prepared based on the measured odometry values ​​and the position variable nodes based on the measured position values. These nodes can be connected to each other via edges that can be associated with probability distributions or covariances that depend on the inaccuracies of the position measurements and / or the inaccuracies of the odometry measurements, respectively. In this way, the available sensor data regarding the position of the vehicle can be represented in the graph in an efficient way.

[0026] A position variable node of a graph is (rigidly) connected to a position anchor (i.e., an anchor node) such that a shift in the position anchor results in a corresponding shift in the position variable node of the graph. In this way, position offsets can be taken into account in a graph in an efficient way.

[0027] The apparatus may be configured to determine the set of estimated pose values ​​(i.e. estimates of the actual trajectory) by optimizing the graph. The optimization of the graph may then include shifting position anchors and / or modifying one or more of the set of pose values ​​and / or shifting pose variable nodes. The optimization of the graph may be performed such that a predefined optimization criterion is improved, in particular at least locally optimized. The optimization criterion may include position deviation terms and / or odometry deviation terms, as described above. By expressing the optimization problem as a graph, the set of estimated pose values ​​may be determined in a particularly efficient and accurate manner.

[0028] The apparatus may be configured to modify the pose of the position anchor as part of the optimization of the graph in response to a probability distribution for the pose of said position anchor, where the probability distribution may be such as to allow translation of the position anchor and to substantially prevent rotation of the position anchor, in this way allowing for position offsets and therefore systematic errors in the measurement of the measured position values ​​to be taken into account and counteracted in a particularly reliable manner.

[0029] As already explained above, the series of times can have several successive partial sequences (partial series of times). The position variable nodes of the graph of one partial sequence can each be connected to an intermediate anchor, so that a shift of the intermediate anchor leads to a corresponding shift of the position variable nodes of the partial sequence. Furthermore, the intermediate anchors of the partial sequences can be connected to a position anchor, so that a shift of the position anchor leads to a corresponding shift of the intermediate anchors of the partial sequences. By taking into account one or several intermediate anchors, systematic errors acting locally in the measurement of the measured position values ​​can also be determined and counteracted in a reliable and accurate manner.

[0030] The device may be configured to determine measured positions of landmarks in the vehicle's surroundings for a particular time in the series of times based on sensor data from one or more external sensors of the vehicle (e.g., cameras and / or lidar sensors). A landmark deviation term may be taken into account in the optimization criterion, which may depend on the deviation of the measured positions of the landmarks from their reference positions. Taking into account one or more landmarks may further improve the quality of the estimation of the actual trajectory.

[0031] The device can be configured to determine a series of measured position values ​​for a number of trips on a section of a road network (and possibly within a limited time interval) at a corresponding series of time instants. Furthermore, the device can be configured to determine a series of odometry values ​​for a number of trips for at least a portion of the series of time instants. In addition, the device can be configured to determine a series of estimated pose values ​​for the number of trips for said series of time instants and a position offset. The series of estimated pose values ​​(i.e. estimates of actual multiple trajectories of different trips) and the position offset can then be determined such that a respective optimization criterion is improved, in particular at least locally optimized. Furthermore, the series of estimated pose values ​​(i.e. estimates of actual multiple trajectories of different trips) and the position offset can be determined such that the position offset for the multiple trips is the same (in particular if the multiple trips are performed within a relatively short time interval). In this way, it can be taken into account in an efficient way that systematic errors in position measurements affect the different trips in the same way. As a result, the quality of the determined estimates of the actual multiple trips can be further improved.

[0032] On the other hand, if necessary, a specialized position offset can be determined at least partially for the different trips (especially if the different trips are relatively far apart in time), in this way changes in the presence of systematic measurement errors can be recognized and taken into account in a reliable manner.

[0033] It should be pointed out that a uniform position offset criterion for several trips on a section of a road network can be applied correspondingly to the partial position offsets (for parts of the section), to the position anchors (for the section) and / or to the intermediate anchors (for parts of the section), in such a way that in particular also local systematic errors in the position determination can be taken into account and offset in an effective and accurate manner.

[0034] The device may be configured to determine alternative data regarding the position of the vehicle and to determine, based on the alternative data, the value of at least one further deviation term for the abovementioned optimization criterion, where the at least one further deviation term may be such that the magnitude of the value of the position offset is increased by the further deviation term. The further deviation term may for example be dependent on the landmarks (as explained above) and / or on measured position values ​​obtained during the further journey. These alternative data may thus comprise, for example, measured positions of the landmarks and / or measured position values ​​of the further journey. By taking into account alternative data regarding the position of the vehicle, systematic errors in the measurement of the position data can be determined and counteracted in a particularly reliable and accurate manner.

[0035] The device may be configured to receive sensor data of a position sensor of the vehicle or vehicle data derived therefrom. Further, the device may be configured to receive sensor data of one or more vehicle sensors of the vehicle or vehicle data derived therefrom. These data may be received, for example, via a (wireless) communication connection. The set of estimated pose values ​​for the set of time instants may then be determined based on the received data by (a device external to the vehicle).

[0036] According to yet another aspect, a (road) motor vehicle (in particular a car or truck or bus or motorcycle) and / or an external vehicle unit (eg a server) is described which comprises the apparatus described herein.

[0037] According to yet another aspect, a computer-implemented method for estimating an actual trajectory of a vehicle is described. The method includes determining a sequence of measured position values ​​for a position of the vehicle at a corresponding sequence of time instants based on sensor data from a position sensor. These measured position values ​​may indicate a position of the vehicle relative to a reference coordinate system. The method further includes determining a sequence of odometry values ​​for at least a portion of the sequence of time instants based on sensor data from one or more vehicle sensors. Here, the odometry values ​​for a time instant may indicate how a pose of the vehicle has changed compared to a previous time instant.

[0038] The method further includes determining a set of estimated pose values ​​for said set of time instants and a (systematic and / or uniform) position offset for said set of measured position values ​​so as to improve, and in particular at least locally optimize, an optimization criterion (as an estimate of the vehicle's actual trajectory), where the optimization criterion may include one position value deviation term for each of said set of time instants. The position value deviation term for a time instant may depend on the deviation of the estimated pose value for that time instant from the measured position value for that time instant shifted by the position offset. Furthermore, the optimization criterion may include an odometry deviation term for at least some of the time instants of said set of time instants. The odometry deviation term for a time instant may depend on the deviation of a change in the pose of the vehicle at that time instant obtained from said set of estimated pose values ​​from the odometry value for that time instant.

[0039] According to yet another aspect, a software (SW) program is described that can be configured to execute on a processor (e.g., on a vehicle controller and / or on a server) and thereby perform the methods described herein.

[0040] According to yet another aspect, a storage medium is described. The storage medium can include a SW program that is configured to execute on a processor and thereby perform the methods described herein.

[0041] It should be noted that the methods, devices, and systems described herein can be used alone or in combination with other methods, devices, and systems described herein. Furthermore, any aspect of the methods, devices, and systems described herein can be combined with each other in various ways. In particular, features of the claims can be combined with each other in various ways.

[0042] The present invention will be described in detail below with reference to examples. [Brief description of the drawings]

[0043] [Figure 1] FIG. 1 illustrates exemplary components of a vehicle. [Figure 2a] FIG. 2 illustrates exemplary sensor data along a vehicle trajectory. [Figure 2b] FIG. 2b illustrates an exemplary vehicle trajectory estimated based on the sensor data of FIG. 2a. [Figure 3a] FIG. 1 illustrates an example location anchor for a position value. [Figure 3b] FIG. 1 illustrates an example vehicle trajectory estimated by shifting position anchors. [Figure 4a] FIG. 13 illustrates an example tree of position anchors for different partial sequence position values. [Figure 4b] FIG. 1 illustrates an example vehicle trajectory estimated by shifting position anchors. [Diagram 5] FIG. 2 illustrates a flowchart of an exemplary method for determining a trajectory of a vehicle. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0044] As explained at the beginning, this document concerns the accurate determination of a vehicle's trajectory based on (possibly error-prone) sensor data of one or more sensors of the vehicle, where in particular it must be possible to accurately localize the trajectory in a (global or world) reference frame.

[0045] FIG. 1 shows exemplary components of a vehicle 100. The vehicle 100 has one or more external sensors 102, which are configured to obtain sensor data (also referred to in this document as surrounding data) related to the surroundings of the vehicle 100. Exemplary external sensors 102 are image cameras, radar sensors, lidar sensors, ultrasonic sensors, etc. A control unit 101 of the vehicle 100 can be configured to detect one or more objects in the surroundings of the vehicle 100 based on the surroundings data (particularly based on image data of the camera). In particular, landmarks (e.g. buildings) whose exact (reference) positions (in a reference coordinate system) are known can be detected in the surroundings of the vehicle 100 based on the surroundings data. Furthermore, measured positions of the landmarks can be determined based on the surroundings data. The measured positions can then be compared with the reference positions of the landmarks in order to locate the vehicle 100 (in a reference coordinate system).

[0046] The vehicle 100 further comprises a position sensor 103 configured to obtain sensor data (also referred to herein as position data) relating to the position of the vehicle 100. The position data may be repeatedly obtained at a series of times. In particular, the position data may include multiple position values ​​for different times, where a position value for a particular time indicates the position (in a reference coordinate system) of the vehicle 100 at that particular time. The (multi-dimensional) position values ​​may for example include GPS coordinates of the vehicle 100. Here, the position values ​​are typically associated with a certain uncertainty (e.g. a given standard deviation).

[0047] Furthermore, the vehicle 100 may have one or more vehicle sensors 104 configured to obtain sensor data (also referred to herein as vehicle data) related to the state of the vehicle 100. The vehicle data may indicate, for example, speed, orientation, yaw rate, acceleration, etc. The control unit 101 may be configured to determine a pose of the vehicle 100, i.e., a location (position) and orientation of the vehicle 100 (determined based on odometry) based on the vehicle data. For example, based on the vehicle data, (multi-dimensional) pose values ​​related to the pose of the vehicle 100 for a series of time instants may be determined, where each individual pose value typically has a certain uncertainty (e.g., a predefined standard deviation).

[0048] Additionally, the vehicle 100 may have a communication unit 105 configured to exchange data with an external unit (e.g. with a back-end server) via a (wireless) communication connection. The vehicle 100 may be configured to transmit sensor data acquired during the travel of the vehicle 100 (in particular surrounding data, position data and / or vehicle data) to the external unit so that the external unit can determine the trajectory (in a reference frame) traveled by the vehicle 100. Alternatively or supplementarily, the control unit 101 of the vehicle 100 may be configured to determine (at least in part) the trajectory traveled by the vehicle 100 based on the acquired sensor data and transmit it to the external unit via the communication unit 105.

[0049] 2a illustrates an example of sensor data acquired while a vehicle 100 is traveling. In particular, FIG. 2a illustrates a series of position values ​​201 at a series of successive times. Each of the position values ​​201 has a certain uncertainty represented by a circle 202 surrounding each of the position values ​​201. As explained above, each of the position values ​​201 may have been acquired based on (or possibly only based on) the position sensor 103 of the vehicle 101. Each of the position values ​​201 may be considered as a position variable node of the graph 215.

[0050] 2a further illustrates pose values ​​211 for the sequence of time instants described above. As explained above, the pose values ​​211 can be determined based on (or possibly only based on) vehicle data from one or more vehicle sensors 104. These pose values ​​211 can be pose variable nodes of a graph 215 that describes the trajectory of the vehicle 100.

[0051] The pose values ​​211 can be determined based on odometry values, which describe the change in pose of the vehicle 100 between two pose values ​​211. These odometry values ​​can be determined based on vehicle data.

[0052] If the pose values ​​211 and the position values ​​201 are measured and / or determined accurately, the positions of the pose values ​​211 and the positions of the position values ​​201 for a particular time should coincide (if the positions of the individual pose values ​​211 are shown relative to a reference coordinate system). However, due to errors and / or inaccuracies in the measurement and / or determination of the pose values ​​211 (especially the odometry values) and the position values ​​201, the positions of the pose values ​​211 and the positions of the position values ​​201 for a particular time will usually differ from each other. Fig. 2a, however, shows (by a dashed straight line) a connection 213 between the pose values ​​211 and the position values ​​201 for the same time. This connection 213 can be considered by an edge in the graph 215.

[0053] Furthermore, landmarks 220 can be detected (based on surrounding data) at independent times during the travel of the vehicle 100. Here, the (reference) positions of the landmarks 220 (e.g., of traffic signs) are usually known (e.g., from a pre-determined digital map). On the other hand, the measured position 221 of the landmarks 220 is accompanied by a certain error and a certain inaccuracy 222. The error of the measured position 221 of the landmarks 220 can then be caused by an error regarding the estimated value of the position of the vehicle 100. In Fig. 2a, a connection 223 between the measured position 221 of the landmarks 220 and the real position is represented (by a dashed straight line). This connection 223 can again be taken into account by an edge in the graph 215.

[0054] The sensor data depicted in Fig. 2a can be used to estimate the location and / or course of the vehicle's actual trajectory 100 in a reference coordinate system. For this purpose, so-called graph-based SLAM (simultaneous localization and mapping) methods can be used. Based on the pose values ​​211, a graph 215 can be prepared describing the pose of the vehicle 100 at a series of time instants. Each pose variable node of the graph 215 then corresponds to each pose value 211. Directly adjacent pose variable nodes are connected to each other via edges, which have one or more boundary conditions or odometry factors 214 for the transition of one pose value 211 to the directly adjacent pose value 211. These one or more boundary conditions or odometry factors 214 then usually depend on the technical characteristics of the vehicle 100 (e.g. maximum steering angle, etc.) and / or on the measurement inaccuracies of the vehicle data. In particular, the odometry factor 214 can depend on the measurement inaccuracies of the odometry values ​​between two pose values ​​211.

[0055] A graph-based SLAM method may be provided for determining a graph 215 such that a given error criterion (e.g., mean square error) related to the deviation of the determined graph 215 from the available sensor data is reduced, and in particular minimized. An estimate of the position of the vehicle 100 at a particular time instant n may be calculated, for example, as:

number

number

number

number

number

number

[0056] At some independent time instants k (e.g., times n=1 and n=N in the example illustrated in FIG. 2a), additional measurements of the position of the vehicle 100 can be determined based on the landmarks 220. The estimate of the position of the vehicle 100 can now be optimized such that the deviation between the measurements 221 of the positions of the landmarks 220 and the reference values ​​of the positions of the landmarks 220 known (e.g., from a digital map) is reduced. Since the reference values ​​of the positions of the landmarks 220 can usually be determined with a relatively high accuracy, the deviation term for the landmarks 220 usually has a relatively large weight in the overall error criterion or optimization criterion.

[0057] A graph 215 can then be determined according to a series of estimates of the position of the vehicle 100 or according to a series of estimated pose values ​​231 of the vehicle 100 at said series of times n=1, ..., N, which reduces, in particular minimizes, the overall error or optimization criterion. Figure 2b shows an example of an optimized graph 215 according to the estimated pose values ​​231 of the vehicle 100 at said series of times n=1, ..., N. This optimized graph 215, in particular said series of estimated pose values ​​231, represents an estimate of the actual trajectory of the vehicle 100.

[0058] As can be seen from Fig. 2b, considering the landmarks 220 that are not related to one another allows the optimized graph 215 to represent a relatively good estimation of the actual trajectory of the vehicle 100 in the immediate vicinity of the landmarks 220. However, due to systematic errors in the position data, in the region 230 between two landmarks 220, the deviation of the estimated trajectory (i.e. the set of estimated pose values ​​231) from the actual trajectory of the vehicle 100 may be relatively large. This deviation arises in particular because in the region 230 between two landmarks 220, no measurements 221 of the positions of the landmarks 220 are available and / or taken into account, and therefore said graph 215 (particularly said set of estimated pose values ​​231) is determined in such a way that the deviation with respect to the available position values ​​201 is reduced.

[0059] Systematic errors in the position data can be seen in Figure 2b by the relatively large deviation between the position estimate 231 and the position values ​​201 in the immediate vicinity of both landmarks 220. The relatively large deviation of the estimated trajectory (i.e., the set of estimated pose values ​​231) from the actual trajectory of the vehicle 100 in the region 230 between the two landmarks 220 results in a "sag" in the set of estimated pose values ​​231, as shown in Figure 2b.

[0060] Systematic errors in the measurement of the position values ​​201 can be taken into account by introducing an additional variable (i.e. an additional node) when determining the graph 215, which allows to shift (shift / move) all considered position values ​​201 in unison. In particular, the individual position values ​​201 (i.e. position variable nodes) can each be anchored via a rigid edge 312 to a common position anchor 321, which can be shifted by a translation 323 to shift each individual measured position value 201 (i.e. position variable node) uniformly. This position anchor 321 can be shifted according to a predefined covariance or probability distribution 322, if necessary.

[0061] Thus, when determining the optimized graph 215 (and thus the set of estimated pose values ​​231), the position anchors 321 can be shifted (particularly by translation 323) as an additional degree of freedom. This shift is performed by the position offset 330. As a result, the systematic errors of the measured position values ​​201 can be implicitly determined and offset (by shifting the position anchors 321). The magnitude of the systematic errors is then described by the position offset 330. This further allows to eliminate deviations of the optimized graph 215 (particularly the set of estimated pose values ​​231) from the actual trajectory of the vehicle 100 that are due to systematic errors of the measured position values ​​201. Figure 3b shows the optimized graph 215 (particularly the optimized set of estimated pose values ​​231) and the shifts of the position anchors 321 caused as part of the optimization.

[0062] The systematic errors in obtaining the measured position values ​​201 may also depend on the immediate surroundings of the vehicle 100. For example, relatively tall buildings in the immediate surroundings of the vehicle 100 may modify the systematic errors (e.g. due to multipath effects). In order to be able to take into account different systematic errors in different sections of the trajectory, a tree or hierarchical structure of anchors 321, 421 may be used, as shown in Fig. 4. In particular, one intermediate anchor 421 may be defined for each section of the trajectory, to which the measured position values ​​(or the position variables 301 corresponding to the position values) of each section are rigidly linked. These intermediate anchors 421 may then be shifted relative to the anchor points 424 (illustrated by connections 423) according to the covariance or probability distribution 422 of each intermediate anchor 421. The anchor points 424 of the individual intermediate anchors may then be linked to the (main) position anchor 321 (possibly via a hierarchy of one or more further intermediate anchors 421).

[0063] As can be seen from FIG. 4b, in this way it is possible to determine a graph 215 that is optimized in a particularly accurate manner, taking into account the systematic errors of the position values ​​201 in the individual sections of the trajectory, i.e. an estimation of the actual trajectory.

[0064] This document thus describes a method that can be used to take into account and at least partially offset systematic errors in positioning, especially in GPS-based positioning. Systematic errors in positioning can be caused by locational effects, such as obstruction by house walls or forests, or by atmospheric effects. These systematic errors can result in measurements that are mutually dependent. In the case of systematic errors caused by atmospheric effects, this concerns all measurements in a certain period of time, for example in a given journey or trip of the vehicle 100. There can also be other locational effects, in which case obstruction and multipath effects can occur in certain areas, for example in front of high-rise buildings, which cause systematic errors at that location.

[0065] In the method described herein, the graph-based SLAM method is extended by modeling systematic errors of the position data. In this case, each measured position value 201 (or position variable node 301 corresponding to the measured position value 201) is connected to a position anchor 321 via a rigid edge 312. This position anchor 321 can be defined such that it can only be shifted by translation 323 with a prior distribution (Prior) with covariance 322 set to constant. In this way, systematic shifts of all measured position values ​​201 or position variable nodes 301 are implicitly incorporated into the factor graph 215. That is, the graph-based SLAM optimization determines the most likely systematic shift of the measured position values ​​201 or position variable nodes 301 taking into account systematic errors of the position data when determining the driving trajectory. A hierarchical system with multiple intermediate anchors 421 allows even spatial systematic errors of the position measurements to be modeled and determined.

[0066] The described method can be used in particular for change recognition and for the localization of the driving trajectory required to provide a map patch. First, a factor graph 215 is created. In this case, the individual measured position values ​​201 are not directly used as priors for the vehicle position at each time. Instead, for each position measurement, the corresponding position variables are inserted into the factor graph 215 as position variable nodes 301. These position variable nodes 301 may each be connected via one position factor 313 (i.e. via a corresponding edge) to the associated vehicle pose or pose variable node 211. The position measurement uncertainty originally included in the position prior can be modeled by the position factor 313. In addition, all position variable nodes 301 are connected to a virtual position anchor variable 321 via ideally rigid virtual edges 312. This position anchor variable 321 can represent a pose (position + orientation) and is also usually accompanied by a prior distribution (i.e., a probability distribution 322 on the pose of the position anchor variable 321). To allow systematic shifting of all position variable nodes 301, this prior distribution preferably has a translation covariance 322 that encodes a reasonable assumption on the systematic shift. The orientation of the anchor pose can be fixed with an ideally small orientation covariance to avoid rotation of the position anchor variable 321 (and thereby ensure reliable modeling of systematic errors in the localization).

[0067] The rigid factors 312 between the position variable nodes 301 and the anchor nodes 321 preserve the internal positional relationships between the position variable nodes 301, and allow "arbitrary" systematic translations to be modeled and optimized in the factor graph 215. This improves the localization of the driving trajectory and systematically shifts the localization measurements.

[0068] Furthermore, a hierarchical system with multiple intermediate anchors 421 is described. This allows the systematic errors depending on the location to be modeled. The construction of neighbor relationships can be done in time and / or space. This allows the modeling and consideration of various systematic errors depending on the location. If this variant is used for synchronously determining the location of multiple trajectories (e.g. of multiple vehicles 100), the intermediate anchors 421 are linked to each other based on their temporal or spatial proximity, so that the spatial and / or time dependence of the errors in the underlying position measurement can be modeled for different trajectories. An example of spatial dependence is the systematic error caused by the walls of a house. The time dependence can be caused by the ionospheric conditions at a certain time or by the same satellite constellation for different trips at the same time.

[0069] 5 shows a flow chart of an exemplary method 500 for estimating an actual trajectory of a vehicle 100. The method 500 can be performed, for example, by the control unit 101 of the vehicle 100 and / or by a unit external to the vehicle (e.g., a back-end server).

[0070] The method 500 includes determining 501 a series of measured position values ​​201 (e.g., GPS coordinates) for each position of the vehicle 100 at a corresponding series of times based on sensor data of the position sensor 103 (e.g., based on sensor data of a GPS receiver). In particular, the measured position values ​​201 for a time may indicate the position of the vehicle 100 at that time relative to a reference coordinate system. This position sensor 103 may be part of the vehicle 100. The measured position values ​​201 may reconstruct the measured trajectory of the vehicle 100 (using the sensor data of the position sensor 103).

[0071] The method 500 further includes determining 502 a series of odometry values ​​for at least a portion of the series of times based on sensor data from one or more vehicle sensors 104. The odometry values ​​for a time may now indicate how a pose of the vehicle 100 has changed compared to a previous time. The series of times may include, for example, times n=1,...,N. Then, for times n=2,...,N, a respective odometry value may be provided that indicates how a pose of the vehicle 100 has changed relative to a respective immediately preceding time.

[0072] The one or more vehicle sensors 104 may indicate, for example, the speed, the distance traveled, the yaw rate, the rotational speed and / or the acceleration of the vehicle 100 to determine odometry values. Alternatively or supplementarily, based on the odometry values, one or more control parameters may be determined, which are used, for example, to control the drive motor, the braking device and / or the steering device of the vehicle 100. Each individual odometry value may indicate a change in position and / or a change in orientation of the vehicle 100. Based on the series of odometry values, the relative vehicle 100 trajectory may be reconstructed and the trajectory may be translated and / or rotated to an estimate of the actual trajectory.

[0073] The method 500 further includes determining 503 a set of estimated pose values ​​231 for said set of time instants (as an estimate of the actual trajectory of the vehicle 100) as well as determining a (systematic) position offset 330 for said set of measured position values ​​201. The estimated pose value 231 for a given time instant may indicate the pose of the vehicle 100 at that time instant relative to a reference coordinate system. The set of estimated pose values ​​231 and position offsets 330 may be determined such that an optimization criterion is improved, in particular at least locally optimized.

[0074] Here, the optimization criterion may include one position value deviation term for each of said series of times. The position value deviation term for a certain time may depend on the deviation 313 of the estimated pose value 231 for that time from the measured position value 201 for that time shifted by the position offset 330. In this way, a series of estimated pose values ​​231 can be determined which leads to a reduction, in particular a minimization, of the (average, possibly squared) deviation of the estimated pose value 231 from the measured position value 201, respectively shifted by the position offset. By taking into account the position offset 330, systematic errors in the measurement of the measured position values ​​201 can be recognized and counteracted in an effective and accurate manner. The accuracy of the estimation of the actual trajectory can be increased in this way.

[0075] Furthermore, the optimization criterion may include an odometry deviation term for at least some of the time instants (particularly, for time instants n=2,...,N) of the series of time instants, where the odometry deviation term for a time instant n may depend on the deviation of the change in the pose of the vehicle 100 at that time instant from the series of estimated pose values ​​231 from the odometry value for that time instant. The optimization criterion may include, for example, a (possibly weighted) sum of the position value deviation term and the odometry deviation term. Taking into account the position value deviation term and the odometry deviation term may provide an accurate estimation of the actual trajectory.

[0076] The methods described herein can improve the quality of locating a trajectory relative to an existing digital map and / or the quality of simultaneously locating multiple trajectories by finding common landmarks (e.g., signs, signals, lane markings, etc.) by modeling systematic, global and local errors in positioning. In particular, the quality of localization of driving trajectories relative to a map or to each other can be improved through implicit modeling of systematic errors. The methods described herein can be used, for example, for change recognition and learning of high-precision digital maps (e.g., for autonomous driving).

[0077] The invention is not limited to the embodiments shown, and it should be noted in particular that the specification and drawings are intended only to illustrate the principles of the proposed methods, devices and systems.

Claims

1. An apparatus (101) for estimating an actual trajectory of a vehicle (100), comprising: determining a series of measured position values ​​(201) at a corresponding series of times based on the sensor data of the position sensor (103), the measured position value (201) for a certain time being indicative of a position of the vehicle (100) at that time relative to a reference coordinate system; determining a series of odometry values ​​based on sensor data from one or more vehicle sensors (104), the odometry values ​​for a time instant indicative of how a pose of the vehicle (100) has changed compared to a previous time instant; A set of estimated pose values ​​(231) for the set of time instants and position offsets (330) for the set of measured position values ​​(201) are determined to improve or at least locally optimize an optimization criterion as an estimate of the actual trajectory of the vehicle (100), the estimated pose value (231) for a given time instant indicating an estimate of a pose consisting of a position and orientation of the vehicle (100) at that time instant within the estimated actual trajectory relative to the reference coordinate system, the optimization criterion including one position value deviation term for each of the set of time instants, The position value deviation term for an instant depends on a deviation (313) of an estimate of a position of the estimated pose value (231) for that instant from the measured position value (201) for that instant shifted by the position offset (330), and the optimization criterion includes an odometry deviation term for at least some instants of the series of instants, the odometry deviation term for a certain instant depends on a deviation of a change in pose of the vehicle (100) at that instant obtained from the series of estimated pose values ​​(231) from the odometry value for that instant. The apparatus (101) is configured as follows.

2. 2. The device (101) according to claim 1, said measured position value (201) having a position measurement uncertainty; the optimization criterion depends on the position measurement inaccuracy of the measured position values ​​(201), in particular such that an increase in the position measurement inaccuracy leads to a decrease in the influence of the position value deviation term on the optimization criterion; and / or The odometry value comprises an odometry measurement uncertainty; The optimization criterion depends on the inaccuracy of the odometry measurement of the odometry, in particular such that as the inaccuracy of the odometry measurement increases, the influence of the odometry deviation term on the optimization criterion decreases. Device.

3. 3. An apparatus (101) according to claim 1 or 2, configured to determine the position offset (330) for the series of measured position values ​​(201) such that the value of the position offset (330) is the same for all measured position values ​​(201) of the series and / or is the same for all time instants of the series; and / or The position offset (330) is a systematic position offset. Device (101).

4. An apparatus (101) according to any one of claims 1 to 3, comprising: configured to limit the value of the position offset (330) to a certain maximum value; and / or configured to determine the value of the position offset (330) taking into account the covariance of the position offset (330); and / or The optimization criterion has a term that is limited by increasing magnitude of the value of the position offset (330). Device (101).

5. An apparatus (101) according to any one of claims 1 to 4, comprising: Dividing the sequence of times into a plurality of consecutive partial sequences; Shifting the measured position values ​​(201) of one partial sequence by a systematic partial position offset for that partial sequence and by the position offset (330) for all sequences, respectively; determining the sequence of estimated pose values ​​(231) for the sequence of time instants, the position offsets (330) for the measured position values ​​(201) of the entire sequence, and the partial position offsets (330) for the partial sequence, so as to improve, and in particular at least locally optimize, the optimization criterion; The apparatus (101) is configured as follows.

6. An apparatus (101) according to any one of claims 1 to 5, comprising: An apparatus (101) configured to generate a graph (215), comprising: the graph (215) having each estimated pose value (231) of the series of estimated pose values ​​(231) as a pose variable node; the graph (215) having odometry factors as edges between a pose variable node for a time instant and a pose variable node for a previous time instant, the factors depending on sensor data of one or more vehicle sensors (105) and / or indicative of a motion of the vehicle (100) between two of the pose variable nodes and / or depending on the odometry values ​​for that time instant; the graph (215) having one position variable node for each of the position values ​​(201) in the series of measured position values ​​(201); the graph (215) having, for each of the pose variable nodes, a connection to a corresponding position variable node via an edge and / or via a position factor; The position variable node of the graph (215) is connected to one position anchor (321) such that a shift in the position anchor (321) results in a corresponding shift in the position variable node; The set of estimated pose values ​​(231) is determined by optimizing the graph (215). The apparatus (101) is configured as follows.

7. 7. The apparatus (101) of claim 6, wherein the optimization of the graph (215) comprises: Shifting the position anchor (321), and / or modifying one or more estimated pose values ​​(231) of said series of estimated pose values ​​(231); and / or Shifting one or more pose variable nodes of the graph (215). An apparatus comprising:

8. 8. An apparatus (101) according to claim 6 or 7, comprising: as part of optimizing the graph (215), the pose of the position anchor (321) is adapted to be modified in response to a probability distribution of the pose of the position anchor (321); The probability distribution is such that it allows translation of the position anchor (321) and substantially prevents rotation of the position anchor (321). Device (101).

9. Apparatus (101) according to any one of claims 6 to 8, comprising: the sequence of times includes a plurality of consecutive partial sequences, the position variable nodes of the graph (215) of one partial sequence are each connected to an intermediate anchor (421) such that a shift in the intermediate anchor (421) results in a corresponding shift in the position variable nodes of the partial sequence; The intermediate anchors (421) of the plurality of partial sequences are connected to the position anchor (321) such that a shift in the position anchor (321) results in a corresponding shift in the intermediate anchors (421) of the plurality of partial sequences. Device.

10. Apparatus (101) according to any one of claims 1 to 9, comprising: configured to determine an estimate of the actual trajectory of the vehicle (100) by a graph-based SLAM method, i.e., by simultaneous self-localization and environmental mapping; The position offsets (330) are considered as shiftable anchor nodes as part of the graph-based SLAM method. Device (101).

11. An apparatus (101) according to any one of claims 1 to 10, comprising: determining measured positions (221) of landmarks (220) in the vicinity of the vehicle (100) for a particular time in the series of times based on sensor data from one or more external sensors (102) of the vehicle (100); The optimization criterion takes into account a landmark deviation term, which depends on the deviation of the measured position (221) of the landmark (220) from a reference position of the landmark (220). The apparatus (101) is configured as follows.

12. Apparatus (101) according to any one of claims 1 to 11, comprising: the optimization criterion includes at least one further deviation term, which increases the magnitude of the value of the position offset (330); and / or determining alternative data regarding the position of the vehicle (100) and determining values ​​of further deviation terms for the optimization criterion based on the alternative data; The apparatus (101) is configured as follows.

13. Apparatus (101) according to any one of claims 1 to 12, comprising: one measured position value (201) comprises coordinates, in particular GPS coordinates, relating to the position of the vehicle (100) in a reference coordinate system for the road network, and / or an odometry value comprising a coordinate relating to a change in position of the vehicle (100) in a coordinate system of the vehicle (100); and / or The coordinate system of the vehicle (100) is offset with respect to the reference coordinate system Device.

14. 14. The device (101) according to any one of claims 1 to 13, configured to create and / or optimise a digital map for a road network based on an estimate of the actual trajectory of the vehicle (100).

15. Apparatus (101) according to any one of claims 1 to 14, comprising: receiving sensor data of a position sensor (103) of the vehicle (100) or data of the vehicle (100) derived therefrom; receiving sensor data from one or more vehicle sensors (104) of the vehicle (100) or data derived therefrom for the vehicle (100); Based on these received data, determining the sequence of estimated pose values ​​(231) for the sequence of time instants. The apparatus (101) is configured as follows.

16. Apparatus (101) according to any one of claims 1 to 15, comprising: determining, for a number of journeys along a section of the road network, a series of measured position values ​​(201) at a corresponding series of times; determining a respective sequence of odometry values ​​for at least a portion of the sequence of times for the plurality of journeys; The apparatus (101) is configured to determine, for the plurality of runs, a respective set of estimated pose values ​​(231) for the set of times and a respective position offset (330) such that a respective optimization criterion is improved, in particular at least locally optimized.

17. A method (500) for estimating an actual trajectory of a vehicle (100), comprising: determining (501) a series of measured position values ​​(201) at a corresponding series of times based on sensor data of the position sensor (103), the measured position value (201) for a certain time being indicative of a position of the vehicle (100) at that time relative to a reference coordinate system; determining (502) a series of odometry values ​​for at least a portion of the series of times based on sensor data from one or more vehicle sensors (104), the odometry values ​​for a time being indicative of how a pose of the vehicle (100) has changed compared to a previous time; determining (503) a set of estimated pose values ​​(231) for the set of time instants and position offsets (330) for the set of measured position values ​​(201) so as to improve or at least locally optimize an optimization criterion as an estimate of the actual trajectory of the vehicle (100), the estimated pose value (231) for a given time instant indicating an estimate of a pose consisting of a position and orientation of the vehicle (100) at that time instant within the estimated actual trajectory relative to the reference coordinate system, the optimization criterion including one position value deviation term for each of the set of time instants; or The position value deviation term for a given time depends on a deviation (313) of an estimate of a position of the estimated pose value (231) for that time from the measured position value (201) for that time shifted by the position offset (330), and the optimization criterion includes an odometry deviation term for at least some of the time instants of the series of time instants, the odometry deviation term for a given time instant depending on a deviation of a change in pose of the vehicle (100) at that time instant obtained from the series of estimated pose values ​​(231) from the odometry value for that time instant. The method (500),

Citation Information

Patent Citations

  • Map generating device, map generation method, movement method of mobile, and robot device

    JP2012064131A

  • Self-driving vehicle-purpose digital map preparation method

    JP2019002925A

  • Fusion of Position Data by Means of Pose Graph

    US20180306584A1

  • Method and System for Accurate Long Term Simultaneous Localization and Mapping with Absolute Orientation Sensing

    US20190323845A1

  • Information processing device, information processing method, program, and moving body

    WO2019098002A1