Method and device for verifying digital map of road section, machine-readable storage medium

Abstract

The invention relates to a method for validating a digital map of a road section, comprising the steps of: receiving environmental sensor data based on a detection of the road section by an environmental sensor; ascertaining a feature of the road section having a first tolerance on the basis of the environmental sensor data; a feature of the digital map, which corresponds to the ascertained feature and has a second tolerance, is validated on the basis of the ascertained feature of the road section, a set-based method using the two tolerances is used in order to validate the feature of the digital map in order to ascertain the validation result, making one or more hypotheses for the set-based method; evaluating the reliability of the one or more hypotheses made by using a probabilistic approach; the digital map is validated based on the validation result and based on the probabilistic reliability evaluation. The invention also relates to a corresponding device, a computer program and a machine-readable storage medium.

Description

Technical Field

[0001] The invention relates to a method, a device, a computer program and a machine-readable storage medium for verifying a digital map of a road section. Background Art

[0002] High-resolution road maps, so-called "high-definition" (HD) maps, are the basis for automated driving functions. Keeping HD maps reliably up-to-date is crucial in this context. Creating and maintaining such maps can be very complex, as, for example, measurement drives must be performed and the data manually processed. This can hinder the efficient scaling of HD maps and, therefore, the automated driving functions. Summary of the Invention

[0003] The object of the present invention is to provide a concept for the efficient verification of digital maps of road sections.

[0004] This object is achieved by the subject matter of the present invention. Advantageous embodiments of the present invention are each the subject matter of an improved embodiment.

[0005] According to a first aspect, a method for validating a digital map of a road section is provided, comprising the steps of:

[0006] receiving environmental sensor data based on road segment detection by environmental sensors,

[0007] Determine road segment features with a first tolerance based on the environmental sensor data.

[0008] A digital map feature corresponding to the ascertained feature and having a second tolerance is verified based on the ascertained road segment feature, wherein a set-based method using the two tolerances is used for verifying the digital map feature in order to determine a verification result, wherein one or more assumptions are made for the set-based method.

[0009] Use probabilistic methods to evaluate the reliability of one or more assumptions made,

[0010] The digital map is verified based on the verification results and on a probabilistic reliability evaluation.

[0011] According to a second aspect, an apparatus is provided, which is configured to carry out all steps of the method according to the first aspect.

[0012] According to a third aspect, there is provided a computer program comprising instructions which, when the computer program is executed by a computer, for example by an apparatus according to the second aspect, cause the computer to carry out the method according to the first aspect.

[0013] According to a fourth aspect, there is provided a machine-readable storage medium having stored thereon the computer program according to the third aspect.

[0014] The present invention is based on and includes the recognition that the aforementioned object is achieved by using a combination of a set-based approach and a probabilistic approach. In other words, provision is made for, on the one hand, using a set-based method, wherein one or more assumptions are made for the method. On the other hand, a probabilistic method is used to assess the reliability of the one or more assumptions made.

[0015] By performing an additional probabilistic evaluation of the assumptions, the strength of the guarantees derived from applying ensemble-based methods is improved.

[0016] Furthermore, for example, the following advantage arises: for set-based methods, assumptions can also be made that are incorrect with a residual probability that is classified as acceptable, for example by truncating the probability distribution. This advantageously improves the accuracy of the verification result.

[0017] Thus, a corresponding automatically verified digital map can additionally include a guarantee of correctness, so that automated driving functions intended to use the digital map can be implemented based on this guarantee. In other words, it can be provided, for example, that a digital map with such a guarantee allows a higher degree of automated driving functions than a digital map without such a guarantee.

[0018] This results in, inter alia, the technical advantage that digital maps of road sections can be efficiently verified.

[0019] In one embodiment of the method, it is provided that the one or more assumptions made are elements selected from the following group of assumptions: the assumption that the determined feature confirms the position of a feature of the digital map; the assumption that the determined feature actually exists; the assumption that static objects not recorded in the digital map are not systematically ignored; the assumption that there are no systematic errors; the assumption that the digital map has not changed since its last update.

[0020] This results in the technical advantage, for example, that particularly suitable assumptions can be made.

[0021] In one embodiment of the method, it is provided that the probabilistic method is one of the following probabilistic methods: Subjective Logic, Dempster-Shafer theory, classical probability theory, in particular Bayesian inference.

[0022] This results in, for example, the technical advantage that particularly suitable probabilistic methods can be used.

[0023] In one embodiment of the method, it is provided that the probabilistic method is subjective logic, wherein a subjective logic opinion is added to each of the one or more hypotheses made, wherein the individual subjective logic opinions are merged to form an overall result, which is projected onto a residual probability, based on which the digital map is verified.

[0024] In other words, for example, the following provisions may be made: the probabilistic method is subjective logic; a subjective logical opinion is added to each of the at least one hypothesis; the individual subjective logical opinions are combined into an overall result; the overall result is projected onto a residual probability, for example. The digital map is verified based on this residual probability.

[0025] This results in the technical advantage, for example, that digital maps can be efficiently verified.

[0026] In one embodiment of the method, it is provided that if the obtained feature is a temporary obstacle, then the own subjective logical opinion is added to the temporary obstacle respectively, and the own subjective logical opinion is merged with the various subjective logical opinions into an overall result.

[0027] This results in the technical advantage, for example, that temporary obstacles can be efficiently dealt with for the verification of the digital map.

[0028] In one embodiment of the method, the associated tolerances of the individual features each specify a region within which the respective feature lies, wherein it is determined which overlaps the individual regions have, and the digital map is verified based on the determined overlaps.

[0029] This results in the technical advantage, for example, that digital maps can be efficiently verified.

[0030] If, for example, the first tolerance lies completely within the second tolerance, then it is determined, for example, that the digital map is verified.

[0031] In one embodiment of the method, it is provided that if no overlap exists, it is determined that the digital map is not verified.

[0032] This results in the technical advantage, for example, that it can be efficiently determined that the digital map is not verified.

[0033] In other words, if there is no overlap, it is ensured that the ascertained feature has not been identified at the recorded position of the digital map.

[0034] In one specific embodiment of the method, provision is made for the digital map to be authorized for automated driving functions based on the verification.

[0035] This results in the technical advantage, for example, that automated driving functions can be efficiently enabled.

[0036] Automated driving functions are particularly understood to be driving functions for an at least partially automated motor vehicle. Automated driving functions may, for example, include infrastructure-supported assistance for the motor vehicle.

[0037] Environmental sensor data within the meaning of this specification is determined, for example, using one or more environmental sensors of one or more motor vehicles and / or one or more environmental sensors spatially distributed within the infrastructure that includes the road section. In other words, it is provided, for example, that one or more motor vehicles, each including one or more environmental sensors, are or have been traveling within the road section and, during the travel, detected their respective surroundings using the environmental sensors, so that the corresponding environmental sensor data describes or represents the road section.

[0038] The method explicitly comprises, for example, the step of detecting a road section by means of one or more environmental sensors, in particular one or more environmental sensors of a motor vehicle and / or environmental sensors of an infrastructure.

[0039] Thus, the step of “receiving environmental sensor data based on detection of a road section by an environmental sensor” is expanded, for example, by receiving environmental sensor data based on multiple detections of the road section by one or more environmental sensors.

[0040] For example, it is provided that the motor vehicle drives over the road section several times, so that corresponding surroundings sensor data are ascertained for each drive.

[0041] For example, it can be provided that a plurality of motor vehicles travels along the road section, so that a plurality of environmental sensor data describing the road section is respectively ascertained therefrom.

[0042] Environmental sensor data in the sense of this description describe or represent a road section.

[0043] One or more or all method steps may be performed, for example, by one or more of the following instances: the motor vehicle, a remote server, ie a server remote from the motor vehicle, such as a so-called edge server. An edge server is a server that is part of an edge infrastructure.

[0044] In other words, it can be provided, for example, that a single motor vehicle can in each case ascertain environmental sensor data during a plurality of driving through a road section and can verify a digital map of this road section based on these environmental sensor data.

[0045] For example, it can be stipulated that motor vehicles prepare their own digital maps in advance.

[0046] For example, it may be provided that the motor vehicle transmits environmental sensor data to a remote server, which then uses these environmental sensor data in order to verify the digital map.

[0047] The method is, for example, a computer-implemented method.

[0048] The device is configured, for example, in terms of programming technology, to execute the computer program.

[0049] Explanations made in conjunction with one hypothesis apply similarly to multiple hypotheses, and vice versa.

[0050] The digital map of the road section can be, for example, an HD map. Here, "HD" stands for "High Definition", so that the HD map is a high-resolution map.

[0051] For example, an HD map can be divided into a plurality of route segments, which can each be derived from a so-called SD map, for example. Here, "SD" stands for "Single Definition," so that an SD map stands for a single-resolution map.

[0052] In other words, for example, a motor vehicle can, within the scope of mapping a road section, drive through the road section multiple times, develop SD maps based on the corresponding environmental sensor data, and develop an HD map of the road section based on these SD maps, wherein the HD map is verified, for example, according to the concept described herein.

[0053] The term “agent” is used in the following explanations. Such an “agent” is, in particular, a motor vehicle which comprises one or more surrounding sensors for detecting a road section.

[0054] In the following explanations, the term "Map-Server" refers to "map server".

[0055] An environmental sensor in the sense of this description is, for example, one of the following environmental sensors: a radar sensor, an image sensor, in particular an image sensor of a video camera, a lidar sensor, an ultrasonic sensor, an infrared sensor, and a magnetic field sensor.

[0056] The term "sensor" used in this specification stands for "environmental sensor".

[0057] The following terms are explained below.

[0058] Subjective Logic

[0059] Subjective logic is an extension of classical probability theory for small samples. It was mainly proposed by Audun Yosan. ) and his research team. The core concept of subjective logic is similar to Dempster-Schafer theory, with Dempster-Schafer theory incorporated into subjective logic as a special case. A unique feature of this theory is that, in addition to first-order probability masses, it also incorporates statistical uncertainty equivalent to second-order probabilities. This statistical uncertainty indicates how certain the estimated first-order probabilities are. Accordingly, subjective logic can effectively model probabilistic conclusions about reliability or truth, given limited knowledge.

[0060] The core element of subjective logic is the so-called subjective logical opinion. Subjective logical opinions can be combined using various operators defined within subjective logic theory. Each operator has its own semantics, which are then assigned to the result.

[0061] Because of their kinship with Dempster-Schafer theory and also with classical probability theory (such as Bayesian inference), subjective logic is called an alternative approach.

[0062] Subjective Logic Opinion

[0063] The subjective logical opinion ω(b,u,a) consists of three components: a belief measure vector b, a measure of statistical uncertainty u, and a vector of statistical priors a. These vectors have the same dimensions as the number of elementary events, and u is a scalar where the sum of u and b equals 1.

[0064] Projected Probability

[0065] The projected probability is the probability obtained by projecting subjective logic onto the probability space. In this case, the statistical uncertainty is supplemented by the statistical prior, so that the trust metric sums to 1.

[0066] Bijective Mapping

[0067] Subjective logical opinions can be directly represented as Dirichlet distributions. By means of bijective mapping, subjective logical opinions can be mapped to their corresponding Dirichlet distributions, and vice versa.

[0068] Dirichlet distribution

[0069] The Dirichlet distribution (after Peter Gustav Lejeune Dirichlet) is a family of continuous, multivariate probability distributions.

[0070] Trust Discounting

[0071] An operation in subjective logic theory in which confidence measures in subjective logical opinions are converted into statistical uncertainty. This allows, for example, modeling knowledge loss due to aging of information.

[0072] Cumulative Belief Fusion Operator

[0073] An operator in subjective logic theory that is equivalent to Bayesian inference.

[0074] Weighted Average Belief Fusion Operators

[0075] In subjective logic theory, an operator that combines multiple subjective opinions to make the result more precise rather than more certain. This means that the statistical uncertainty of the result is greater than or equal to the minimum statistical certainty of the input subjective opinions.

[0076] Dempster-Shafer theory

[0077] The theory of evidence proposed by Dempster and Shafer, also known as the "Theory of Belief Functions," is a mathematical theory in the field of probability theory. This theory uses the so-called Dempster rule of combination to combine information from different sources into an overall conclusion, taking the credibility of these sources into account.

[0078] Set-based solutions

[0079] Set-based approaches present measurement uncertainty in the form of a set, within which the measurement can lie. A typical example of a set-based approach is the interval method. Using a set-based approach's worst-case estimate, the conclusion is not only probable but also absolute. Consequently, as long as the assumptions underlying the conclusion hold, the conclusion is guaranteed.

[0080] If the plural forms are used within the scope of this description for “set-based methods” and for “probabilistic methods”, this is always to be understood as including the singular form, and vice versa.

[0081] For example, provision is made for the use of sensor data from networked motor vehicles (fleet data) and / or sensor data from intelligent infrastructure in order to compile or construct digital maps of road sections. Such motor vehicles or such intelligent infrastructure units are referred to below as agents. However, unlike conventional methods, the method described here also directly estimates the reliability of the map, for example. For this purpose, a combination of set-based schemes or methods with probabilistic approaches or methods is used, in particular, a combination of set-based methods with probabilistic methods. In order to obtain the strongest possible guarantees and, on the other hand, to be able to limit interference events as strongly as possible, the HD map is divided into route sections or road sections. These sections can all be derived from single-resolution maps.

[0082] For example, at least one intelligent agent collects sensor data, generates an HD map segment from it, and sends the HD map segment to a map server. Furthermore, at least one intelligent agent verifies existing, unverified map segments based on the sensor data, for example, and sends the verification results to a map server. The map server aggregates the incoming information and, for example, approves map segments that have been verified with sufficient accuracy. For example, the incoming information is aggregated again using a combination of one or more set-based methods and one or more probabilistic methods. For example, the at least one intelligent agent obtains the verified map segment, verifies whether the approval of the current map segment is trustworthy, and uses the corresponding map segment for automated driving functions. Furthermore, the intelligent agent transmits the credibility result as feedback to the map server. The map server updates the HD map based on the feedback. For example, this communication can occur via vehicle-to-everything (V2X) communication, infrastructure-to-everything (I2X) communication, and / or a standard internet connection.

[0083] This results in particular advantages from the implementation as a multi-agent system with multiple, particularly heterogeneous, agents with different capabilities. For example, it is provided that the map server is implemented as an edge server. This minimizes delays and allows the agents to receive map updates in real time. In this case, urgent map updates can be distributed, for example, by broadcast with low latency to all agents involved in the corresponding map segment. All other less time-critical map updates can be transmitted to the agents, for example, via a standard internet connection, as long as there is one.

[0084] A special case is the so-called "commuter mode," in which the multi-agent system consists of, for example, a single motor vehicle. In this case, the map server can also be implemented as a component in the motor vehicle, eliminating the need for V2X communication.

[0085] Building on the concepts described here, the following advantages can be achieved:

[0086] By performing an additional probabilistic evaluation of the assumptions, the strength of the assurances derived from applying the ensemble-based approach is improved.

[0087] For ensemble-based methods, assumptions can also be made that are incorrect with a residual probability that is classified as acceptable (e.g., truncated probability distributions), thereby improving the accuracy of the results.

[0088] Automatically generated HD maps additionally include guarantees regarding their correctness.

[0089] Fully automated development and verification / plausibility validation of HD maps can be performed in near real time.

[0090] Advantage: Map users are quickly informed about map changes and possible inconsistencies / obstructions and can react accordingly.

[0091] This method can handle heterogeneous agents (different capabilities, different trustworthiness) and can utilize data from a large number of different agents, thereby obtaining more resources for mapping.

[0092] The method can use data from agents of different manufacturers and thus achieve a large range of applications (eg networked motor vehicles from different manufacturers).

[0093] This approach can use data from agents with different capability levels, thus reaching a wide range of applications across products of different price classes.

[0094] The following application scenarios can be set:

[0095] Case 1: Route segment has not been mapped yet

[0096] In this case, the agent cannot yet use the HD map to support automated driving functions, but by creating the map, it creates the premise so that other agents or mapping agents can use the map when they come back in the future.

[0097] Any automated mapping method can be used for mapping. However, in addition to mapping, provision is also made for a preliminary estimate of the uncertainty during the mapping process. To ensure that the actual road boundary lies within the estimated uncertainty, a set-based approach is used to estimate the uncertainty. For example, a ray tracing approach can be used, which expresses the measurement uncertainty as an interval along the entire measurement chain, resulting in an area within which the lane boundary is guaranteed to lie. Similarly, other map features, such as traffic signs, can also limit the boundaries of the area within which they are guaranteed to lie.

[0098] For example, a distinction can be made between essential and non-essential map features. Non-essential map features do not necessarily need to be identified in the sensor image due to weather conditions, occlusions, etc., so their absence does not result in a loss of assurance. Examples of non-essential features are, for example, landmarks used for positioning.

[0099] For example, multiple measurement sequences are aggregated based on the sensor quality of the agent. A measurement sequence, for example, is a sequence of all measurements recorded by a networked motor vehicle as it drives through a corresponding map segment. Because set-based methods guarantee that the measured object is within the set of possible locations output, the intersection of these result sets leads to increasingly narrow boundaries. For example, such a large number of measurement sequences can be aggregated to achieve the accuracy required for HD maps.

[0100] Case 2: The route section has been mapped but has not yet been approved for automated driving functions.

[0101] In this case, the agent cannot yet use the HD map to support automated driving functions, but by verifying the creation premise, other agents or mapping agents can use the map when they come back in the future.

[0102] The goal of the mapping process is to achieve the best possible guaranteed accuracy, but the main goal here is to confirm the existing map with the greatest possible certainty. In order to be able to confirm a map segment or road segment or route segment, the following properties must generally be met, for example:

[0103] The drivable area, which is called "free space" in English, must cover the entire driving road.

[0104] All essential map features (eg traffic signs, lane markings) must be recognized / identified at the position or location at which they are entered in the digital map, ie the HD map.

[0105] To achieve this, a set-based approach and a probabilistic method are combined. The set-based approach determines a guaranteed drivable area. If this area is large enough, the first request for the map segment is confirmed. Conversely, if a static obstacle is detected on the road, the confirmation is withdrawn.

[0106] To confirm essential map features, a set-based approach is also initially used. Here, a measurement generates an area within which the corresponding essential map feature must lie. In the exceptional case where the resulting area lies completely within the tolerance of the position of the map feature, the map segment can be confirmed with certainty. However, a single measurement sequence is usually not sufficient to guarantee the required accuracy. Therefore, a set-based approach is combined with a probabilistic approach. To this end, a check is first performed to determine whether the area derived from the set-based approach overlaps with the corresponding map feature. If not, the map feature is guaranteed not to have been identified at the recorded location. Accordingly, confirmation is withdrawn. Otherwise, it is at least plausible that the map is correct.

[0107] If the map at least appears to be credible, a probabilistic approach is used to determine the remaining probability that the HD map's reliability is still compromised due to the violation of the assumptions. If the assumptions built into the mapping method are violated, the HD map's reliability is compromised. The advantage of a ensemble-based approach is that the assumptions underlying the results are clearly known. These assumptions can be, for example:

[0108] Assume that the actual measurements confirm the location of map features;

[0109] Assume that the detected map features actually exist (that is, the measurement is not a false positive);

[0110] It is assumed that unrecorded static objects are not systematically ignored (i.e. false negatives hinder the detection of obstacles);

[0111] Assume that there are no systematic errors (e.g., faulty calibration causing systematic shifts in the HD map);

[0112] Assume that the HD map has not changed since the last update.

[0113] Subjective logic is a particularly advantageous method for determining residual probabilities. A subjective logic opinion ω(b,u,a) consists of three components: a belief measure vector b, a measure of statistical uncertainty u, and a vector a of statistical priors. These vectors have the same dimensions as the number of elementary events, with u being a scalar where the sum of u and b equals 1.

[0114] First, a subjective logical opinion is added to each hypothesis. For the subjective logical opinion for the first hypothesis, i.e., the hypothesis that the measurements actually confirm the corresponding map feature, one can count, for example, how many measurements actually fall within the tolerance and how many measurements fall outside the tolerance. If covariance is available, each measurement can be assigned a probability of belonging to the map feature. Proportionally, some are counted as inside and some as outside. The confidence quality vector b and the statistical uncertainty u are derived from the corresponding number of measurements inside and outside the tolerance. The statistical prior can be derived, for example, from the ratio of the area of the map feature with the tolerance to the sum of all areas attributed to these measurements, as determined by applying a set-based approach.

[0115] The subjective logical opinions regarding the second and third hypotheses—i.e., regarding false positives and false negatives—are of an epistemic nature. They are derived from the sensor specifications and the detection method used, as well as the number of measurement sequences for which reliability is determined. In particular, possible common cause errors (CFEs) must be considered. To this end, in addition to the verification result, the verification agent also transmits a hash value that describes the algorithm used. This allows the mapping server to determine whether the same or different algorithms were used, but it cannot track which specific algorithms were used.

[0116] Methods for reliability estimation and self-assessment allow for the determination of subjective logical opinions regarding the assumption that there are no systematic errors in the agent. In this case, redundant information from other agents or through overlapping sensor coverage is used to verify the consistency and credibility of the perception results.

[0117] To evaluate the map's real-time assumptions in the subjective opinion, for example, the timestamp of the last verification of a route segment and the frequency of visits to that route are primarily used. Furthermore, statistical data describing the probability of relevant changes are used, for example. For example, queuing theory can be used to determine the probability that a relevant change has occurred since the last verification. For example, the subjective opinion can be constructed by constructing an initial subjective opinion based on the verification results and then correcting it using a confidence discount operator based on the subjective theory and the probability of change.

[0118] For example, similar to the approach described here, additional hypotheses can be evaluated using subjective logical opinions. For example, the individual subjective logical opinions can be fused into an overall result. For example, a weighted average confidence fusion operator can be used for this purpose. The fused subjective logical opinions can then be projected onto residual probabilities, for example. For this purpose, projected probabilities from subjective logic can be used. For example, a bijective mapping from subjective logic theory can be used to transform the subjective logical opinions into a Dirichlet distribution and then integrate them over pre-specified confidence intervals.

[0119] Once mapping has been completed and the remaining probability from the validation has been reduced to an acceptable metric, the route segment is permitted, for example, for automated driving functionality.

[0120] Case 3: Route section is permitted

[0121] If the route segment has the necessary permissions for automated driving functions, the agent can use the map for automated driving functions. If intelligent infrastructure units are involved, this driving function could, for example, be guiding other connected and automated motor vehicles through intersections.

[0122] The agent can also perform a plausibility check on the HD map, for example, to confirm the map's real-timeness and, on the other hand, to react to changes in the road since the last time it was driven. For plausibility checking of the map, the verification method from scenario 2 can be reused, for example.

[0123] Dealing with temporary obstacles

[0124] Temporary obstacles can include construction sites, accidents, or even parked vehicles at the edge of a lane. To account for these, the agent has a perception system that can detect such temporary obstacles. If temporary obstacles are taken into account, it may no longer be possible to confirm all map features in each measurement sequence due to occlusions. Consequently, the underlying mapping method must be modified accordingly.

[0125] During initial mapping, for example, invisible locations or places are interpolated. For example, the uncertainty range of the interpolated segments is correspondingly large, and more measurement cycles are used, for example, to ensure that an HD map with the required tolerance is produced.

[0126] Map verification can become more complex, for example, because temporary obstacles can cause obstructions, making the entire road no longer recognized as drivable. However, in this case, for example, map validation is not simply revoked. Instead, a more refined division of route segments can be performed, for example, using a grid map. Thus, grid cells where the measurement falls are marked as "occupied," grid cells in the line of sight to the measurement location are marked as "free," and obscured cells are marked as "unknown." For example, subjective logic or the Dempster-Schafer theory can be used to model the grid map. For example, map segments or road segments are then permitted for grid cells identified as definitely unoccupied. For example, obscured cells are neither confirmed nor discarded, but the timestamps of the measurement series, for example, do not apply to these obscured cells, thus still reducing their certainty. If the certainty that a grid cell is drivable decreases to such an extent that the probability of occupancy rises above the accepted residual probability, the grid cell is entered into the map, for example, as a temporary obstruction. If, for example, the road width is too narrow due to an overestimated temporary obstacle, so that the motor vehicle cannot pass safely, then for example the permission for the entire road segment is withdrawn. For example, if an unregistered non-temporary obstacle is detected, then the permission is also withdrawn.

[0127] If a temporary obstacle is identified, an overestimation is performed with the aid of a set-based approach, which ensures that the temporary obstacle is within the overestimation. This overestimation is entered into the map as a temporary obstacle. For example, a subjective logical opinion is also added to each temporary obstacle: the temporary obstacle is actually still there. For example, for verification, such a subjective logical opinion is determined for each temporary obstacle from each measurement sequence and, for example, fused with the existing subjective logical opinion. This fusion is performed, for example, using a cumulative trust fusion operator from subjective logic theory. For example, the more frequently a temporary obstacle is seen, the higher its probability of existence is, for example. However, if a temporary obstacle is, for example, not detected more frequently, the probability of existence decreases. If the probability of existence, for example, drops below an accepted metric, the obstacle is deleted from the map, which is called "pruning" in English.

[0128] Dealing with unreliable agents

[0129] Agents performing mapping are generally not homogeneous but can differ significantly from one another, for example, in their sensor equipment and processing quality. Consequently, the quality of the measurement sequences, for example, can vary. If measurement accuracy is modeled correctly, poorer measurement accuracy leads to larger uncertainty areas when applying ensemble-based methods, but has no impact on the method itself.

[0130] Conversely, for verification purposes, the reliability and accuracy of the results must be considered, for example. This applies in particular to the probability of false detections (false positives and false negatives) and the probability of systematic errors. Therefore, each agent is assigned a subjective logical opinion about its reliability, and its verification results are weighted by the reliability of the agent. One possible way to determine the reliability of an agent is to use trust management and misbehavior detection mechanisms.

[0131] For example, the permissions for automated driving functions are further subdivided into multiple categories. For example, the stronger the permission, the less demanding the automated driving function is. For example, some automated driving functions may not handle parked vehicles on the road. For example, route sections or road sections where these situations could occur are still not permitted for the driving function. Exemplary permission categories may include the following:

[0132] A suitable fallback level for widespread sensor failures requires that the guarantees be very comprehensive and that the map be very up-to-date. Furthermore, the map must not contain significant uncertainties, such as the presence of temporary static objects. The automated vehicle can use this map to continue driving despite widespread sensor failures.

[0133] Parking garages for automated valet parking (AVP): In appropriately equipped parking garages, the road map can be assumed to be up-to-date and sufficiently accurate at all times thanks to the appropriately installed static sensor systems. For example, specialized landmarks are installed within the garage for positioning. Consequently, motor vehicles can fully rely on maps that are guaranteed to be up-to-date and sufficiently accurate.

[0134] The basis for automated driving functions is preferably one in which these guarantees are comprehensive and enable the use of the map as a planning basis. The map assumes responsibility for safety as a fallback or redundant safety path. Simultaneously, the map is still verified for plausibility during driving. If a discrepancy arises, the automated driving function can fall back to a "no-map" scenario—that is, a scenario without a map—and at least traverse the faulty route section with significantly reduced performance.

[0135] Route sections or road sections contain temporary static objects: the guarantees are limited and more complex driving functionality must be provided, ie more certain recognition and avoidance of obstacles.

[0136] Route sections include construction sites: Road markings in construction sites can be confusing for automated vehicles. Therefore, only perception systems that are robust enough for this challenge are permitted. For example, the automated driving function can rely on guaranteed free space, determined based on the trajectory of a vehicle that recently drove through the construction site.

[0137] Route segment with urgent safety warning: The agent reports a specific hazard on this route segment. This hazard warning can include wrong-way drivers, people on the highway, debris, or weather conditions such as icy lanes or fog banks. In this case, the non-HD map-based driving function will revert to an extra cautious driving mode until the warning is resolved.

[0138] Particularly advantageous embodiments for determining the residual probability include subjective logic. However, the application of Dempster-Schafer theory or Bayesian inference is also conceivable. More generally, for example, a score can be used that increases monotonically with increasing estimated certainty and decreases accordingly when the measurement contradicts the map. For example, if the score or scores exceed a corresponding threshold, permission is granted.

[0139] A particularly advantageous embodiment involves using a weighted average confidence fusion operator to determine the overall reliability. However, other combinations are also possible. For example, the minimum of the individual evaluations can be used, or the probabilities derived from the reliability measures can be added together (i.e., the corresponding elementary events can be ORed together). For example, it is also provided that a threshold value is determined for each individual evaluation, wherein the map is only approved, for example, if all threshold values are met.

[0140] For example, a trajectory actually traveled by one or more motor vehicles is used to derive a reference line therefrom. The reference line can then be used, for example, as a pilot control for longitudinal and / or transverse guidance and thus support driving that is as comfortable as possible.

[0141] For example, camera sensors and / or radar sensors and / or lidar sensors and / or ultrasonic sensors can be used for detection. Furthermore, the trajectory of the motor vehicle can also be used as an input for mapping. The trajectory is determined, for example, based on positioning, which typically uses the motor vehicle's odometer, GNSS (e.g., GPS), and landmarks detected in the sensor images of other motor vehicle sensors. Furthermore, it is provided that the trajectories of other dynamic objects can also be used to determine the guaranteed drivable area.

[0142] For example, a commuter mode is provided in which a single automated motor vehicle generates its own HD map so that it can automatically drive a regular route based on this HD map. A measurement sequence is then formed by the motor vehicle repeatedly driving the same route section.

[0143] The expression "at least one" means "one or more". BRIEF DESCRIPTION OF THE DRAWINGS

[0144] The present invention will be explained in more detail below with the aid of a preferred embodiment. Here it is shown:

[0145] Figure 1 Flowchart of the method according to the first aspect,

[0146] Figure 2 According to the device of the second aspect,

[0147] Figure 3 According to the machine-readable storage medium of the fourth aspect,

[0148] Figure 4 infrastructure,

[0149] Figure 5 Flowchart of the method according to the first aspect,

[0150] Figure 6 A block diagram for initially mapping a road segment is explained as an example.

[0151] Figure 7 A block diagram illustrating, by way of example, a process for validating a digital map of a road section, and

[0152] Figure 8 A block diagram for verifying the plausibility of a digital map of a road section is explained by way of example.

[0153] In the following, the same reference numerals may be used for the same features. DETAILED DESCRIPTION

[0154] Figure 1 A flow chart showing a method for validating a digital map of a road section comprises the following steps:

[0155] receiving 101 environmental sensor data based on detection of a road section by an environmental sensor,

[0156] Based on the environmental sensor data, a feature of the road section with a first tolerance is determined 103 ,

[0157] A feature of the digital map corresponding to the determined feature and having a second tolerance is verified 105 based on the determined feature of the road section, wherein a set-based method using the two tolerances is used 107 to verify the feature of the digital map in order to determine a verification result, wherein one or more assumptions are made 109 for the set-based method.

[0158] By using probabilistic methods to evaluate the reliability of one or more assumptions made by 111,

[0159] Based on the verification results and on the probabilistic reliability evaluation, the digital map is verified 113.

[0160] Figure 2 An apparatus 201 is shown, which is configured to carry out all steps of the method according to the first aspect.

[0161] Device 201 includes an entry 203 configured to receive environmental sensor data. Device 201 further includes a computing unit 205 configured to carry out the steps of determining, verifying, assessing reliability, and validating the digital map. Computing unit 205 is further configured to use a set-based method using the two tolerances to verify features of the digital map in order to determine a validation result, wherein computing unit 205 is configured to make one or more assumptions for the set-based method.

[0162] The device 201 may also be referred to as a device for validating a digital map of a road section.

[0163] The entry 203 can be implemented as a communication interface, for example. The apparatus 201 includes a communication interface implemented separately from the entry 203, wherein the communication interface is configured to communicate with a map server. The entry 203 receives environmental sensor data from an environmental sensor of the motor vehicle, for example.

[0164] Method features are derived analogously from corresponding device features, and vice versa. This means, therefore, that the technical functions of the method are derived analogously from the corresponding technical functions of the device, and vice versa.

[0165] The device is implemented, for example, in a cloud infrastructure.

[0166] Figure 3 A machine-readable storage medium 301 is shown on which a computer program 303 is stored. The computer program 303 comprises instructions which, when executed by a computer, for example by Figure 2 When the apparatus 201 executes the computer program 303, the instructions cause the computer to implement the method according to the first aspect.

[0167] Figure 4An infrastructure 400 is shown, comprising a first road 401 on which a first motor vehicle 403 travels. The infrastructure 400 also comprises a second road 405 on which a second motor vehicle 407 travels. The first road 401 runs perpendicular to the second road 405.

[0168] Both motor vehicles 403 , 407 include one or more environmental sensors for detecting the surroundings of the respective motor vehicle 403 , 407 .

[0169] The area of the surroundings of motor vehicle 403 that is detected by means of the surroundings sensor system of first motor vehicle 403 is identified with reference numeral 409 and thus represents a section of first road 401 .

[0170] Based on this detection, the first motor vehicle 403 can perform a preliminary mapping of the road section or of the first road 401. This preliminary mapping provides an SD map as a result, which is Figure 4 4. The first motor vehicle 403 sends the SD map 411 to a map server 413, which is implemented in an edge infrastructure, for example. Therefore, the map server 413 can also be implemented as an edge server.

[0171] Similarly, reference numeral 415 identifies the detection area of the environmental sensors of the second motor vehicle 407, wherein the second motor vehicle 407 does not have to perform an initial mapping of the second road 405, because this second road has already been mapped. Therefore, there are already mapped road sections here. However, based on the environmental detection of the environmental sensors of the second motor vehicle 407, the second motor vehicle 407 can verify the already mapped road sections. Here, for example, temporary obstacles are detected, which is Figure 4 For example, the third parked vehicle 417 is shown. The result of the verification is displayed together with the information about the temporary obstacle 417. Figure 4 419 and is sent by the second motor vehicle 407 to the map server 413 .

[0172] The infrastructure 400 further comprises a parking structure 421 , wherein a plurality of motor vehicles 425 are parked, for example, on a roof 423 of the parking structure 421 .

[0173] The parking garage 421 can be realized or configured as a smart parking garage, for example. This means that the parking garage 421 is configured, for example, for AVP operation of motor vehicles.

[0174] The abbreviation "AVP" stands for "Automated Valet Parking" and can be translated as "automatic valet parking." Within the scope of the AVP process, a motor vehicle is guided at least highly automatically from a delivery location to a parking space in the parking building with the support of the parking building's infrastructure. With the support of the parking building's infrastructure 421, the motor vehicle is guided at least highly automatically from the parking location to the retrieval location. This occurs particularly when the parking time has expired or when the driver requests retrieval.

[0175] Therefore, the infrastructure of the parking building 421 can detect the parking spaces and driving paths of the parking building 421 by using environmental sensors arranged inside the parking building 421, and send information about this to the map server 413. Figure 4 The reference numeral 427 is provided in the figure. Such information includes, for example, the occupancy status of the individual parking spaces.

[0176] The map server 413 can create or verify a digital map of the road or parking structure 421 based on the received information, as described above and / or below. Figure 4 The digital map is marked with reference numeral 429. The guarantee seal with reference numeral 431 symbolically indicates that the digital map has been verified by the map server 413.

[0177] Figure 5 A flow chart 501 of a method according to the first aspect is shown. For example, automated cartography can be provided, and for example, verification of digital maps can be provided.

[0178] Environmental sensor data representing a road segment is received 503. In step 505, it is determined whether an HD map for the road segment already exists. If not, initial mapping of the road segment is performed using the environmental sensor data in step 507. An SD map of the road segment is generated 509 as a result of initial mapping 507.

[0179] If it is determined in step 505 that an HD map for the road section already exists, then in step 511 it is determined whether the residual risk that the HD map is incorrect is still acceptable. If not, in step 513 the digital map is verified using the environmental sensor data 503. A combination of set-based and probabilistic methods is used. This ensures not only that the assumptions made are correct, but also that these assumptions are evaluated probabilistically. Only when the residual probability that the assumptions made have been violated falls below an acceptance threshold is the HD map permitted for use in automated driving functions, for example.

[0180] If it is determined in step 511 that the residual risk is acceptable, then in step 515 the digital map, ie the HD map, is released for the automated driving function, wherein it can also be provided here, for example, that a plausibility check is performed beforehand.

[0181] Figure 6 A block diagram 601 is shown, which exemplarily illustrates the initial mapping of a road segment.

[0182] Reference numeral 603 points to a dotted rectangle representing an agent, ie, a motor vehicle, for example. This means that the functional blocks within the block 603 are executed by the agent, ie, on the motor vehicle side.

[0183] Reference numeral 605 points to a dotted rectangle symbolically representing a map server, which means that the functional blocks inside the block 605 are implemented by the map server.

[0184] According to function block 605, the agent's environmental sensing device, i.e., one or more environmental sensors, detects the agent's surroundings. For example, the agent may include a video camera 607 and a radar sensor 609. This environmental detection is used, for example, to roughly localize the agent so that a model matching the current road segment is selected from a database of model templates (also referred to as ModelTemplates) 609 based on the SD map 607. The environmental detection data, i.e., environmental sensor data, can be used, for example, to adapt the parameters of the selected model.

[0185] Furthermore, a set-based approach can be used to describe the uncertainty in the environment detection in the form of a set. This is performed according to function block 611 .

[0186] As described above, adaptation of the parameters of the selected model is performed according to functional block 613 .

[0187] Combining these uncertainties with the adapted model results in an HD map segment with tolerances.

[0188] Specifically, a plurality of measurement sequences from one or more agents are aggregated 615 on the map server 605. In other words, the agent 603 sends the uncertainty, i.e., the tolerance, and the result of combining the adapted model to the map server 605. The map server thus obtains corresponding information from the plurality of agents, so that aggregation can be performed over a plurality of such measurements in order to achieve, for example, the required accuracy.

[0189] Therefore, based on the aggregation 615 , an HD map segment 617 of the road segment may be prepared, wherein the map segment has a tolerance.

[0190] exist Figure 6In the embodiment, the map server 605 is implemented separately from the agent 603. The agent 603 notifies the map server 605 of its map model including tolerances, for example, by means of I2X (Infrastructure to Everything) communication or V2X (Vehicle to Everything) communication.

[0191] Figure 7 A block diagram 701 is shown, which exemplarily illustrates the process of verifying an HD map segment 617 .

[0192] The HD map segment 617 is an example of a digital map representing a road segment.

[0193] With the help of environment detection 605 , features such as landmarks can be obtained in the surroundings of agent 603 according to function block 703 , so that agent 603 can be positioned according to function block 705 based on the associated landmarks or features from digital map 617 .

[0194] Based on the location and recognition of the features, the map features contained in the digital map 617 can be verified. This is done according to function block 707. For this purpose, a set-based approach is used so that the verification result is provided with corresponding guarantees. Subsequently, according to function block 709, the assumptions based on which the set-based approach produces guaranteed results are evaluated probabilistically, i.e., by using probabilistic methods. For example, a probabilistic evaluation is performed using subjective logic. In addition, the agent 603 and the map server 605 are Figure 7 For example, they are separated so that the intelligent agent notifies the map server 605 of its verification results including probabilistic reliability evaluation through V2X communication or I2X communication.

[0195] and Figure 6 Similarly, the map server may verify the aggregation 615 with sufficient certainty over multiple measurement sequences of multiple agents and / or one agent (if the agent has driven the corresponding road segment multiple times).

[0196] The digital map 617 can then be verified, for example, as a result of the aggregation 615 and the corresponding verification. This verification is therefore performed according to the functional block 711 .

[0197] Map server 605 may then, for example, license digital map 617 for automated driving functions.

[0198] Figure 8 A block diagram 801 is shown which illustrates a plausibility check. If a map 617 which has already been authorized exists, then the plausibility check is continued for this map, for example. Figure 7The explanation is essentially the same. However, lower accuracy is permitted here, for example. If the license for digital map 617 is sufficiently reliable, the agent can initiate the automated driving function. Otherwise, a feedback is output to the agent's driver, for example, or a fallback to map-free driving can be initiated. For example, a feedback is sent to a map server so that the digital map can be updated each time the road section is driven.

[0199] In block diagram 801 , the passing of block 803 shows the authenticity verification performed on the motor vehicle side. According to block 805 , the authorization of the automated driving function is performed.

[0200] The updating of the digital map is indicated by function block 809 .

[0201] Here, the map server 605 and the agent 603 are also separated, so that the agent 603 notifies the map server 605 of its feedback through V2X communication or I2X communication.

Claims

1. A method for verifying a digital map (617) of a road section, comprising the steps of: receiving (101) environmental sensor data based on detection of the road section by environmental sensors, determining (103) a feature of the road section with a first tolerance based on the environmental sensor data, Based on the determined features of the road section, features of the digital map (617) corresponding to the determined features and having a second tolerance are verified (105), wherein: In order to verify the feature of the digital map (617), a set-based method using the two tolerances is used (107) to determine a verification result, wherein one or more assumptions are made (109) for the set-based method, By using probabilistic methods to evaluate the reliability of one or more assumptions made by (111), The digital map (617) is validated (113) based on the validation result and based on a probabilistic reliability evaluation.

2. The method according to claim 1, wherein The one or more hypotheses made are elements selected from the following group of hypotheses: assuming that the feature found confirms the location of a feature of the digital map (617); It is assumed that the features found do exist; it is assumed that static objects not recorded in the digital map (617) are not systematically ignored; it is assumed that there are no systematic errors; it is assumed that the digital map (617) has not changed since its last update.

3. The method according to claim 1 or 2, wherein: The probabilistic method is one of the following probabilistic methods: subjective logic, Dempster-Schafer theory, classical probability theory, in particular Bayesian inference.

4. The method according to claim 3, wherein: The probabilistic method is subjective logic, wherein a subjective logical opinion is added to each of the one or more hypotheses made, wherein the individual subjective logical opinions are merged into an overall result, which is projected onto a residual probability, and the digital map (617) is verified in view of the residual probability.

5. The method according to claim 4, wherein If the feature obtained is a temporary obstacle, the subjective logical opinion of the user is added to the temporary obstacle respectively, and the subjective logical opinion of the user is integrated with each subjective logical opinion into a total result.

6. A method according to any one of the preceding claims, wherein: The tolerances assigned to the individual features each specify an area within which the corresponding feature lies, wherein it is determined which overlaps the individual areas have, and the digital map (617) is verified based on the determined overlaps.

7. The method according to claim 6, wherein: If there is no overlap, it is determined that the digital map (617) is not verified.

8. A method according to any one of the preceding claims, wherein Based on the verification, the digital map (617) is authorized for use in automated driving functions.

9. An apparatus (201) configured to carry out all the steps of the method according to any one of the preceding claims.

10. A computer program (303) comprising instructions which, when the computer program (303) is executed by a computer, cause the computer to carry out the method according to any one of claims 1 to 8.

11. A machine-readable storage medium (301) on which the computer program (303) according to claim 10 is stored.

Images