Method and apparatus for determining a deployment risk of at least one traffic participant in a deployment location

Abstract

The present disclosure relates to a method and an apparatus for determining a deployment risk of at least one traffic participant (12, 14) in a deployment location. The apparatus (100) comprises interface circuitry (110) configured to receive at least one driving data set (112) comprising dynamic information of the at least one traffic participant, the at least one driving data set being associated with the deployment location. The apparatus (100) further comprises processing circuitry (120) configured to determine a number of risk incidents based on the driving data set. The processing circuitry (120) is further configured to determine, for the number of risk incidents, a respective severity of the risk incident. In addition, the processing circuitry (120) is configured to aggregate the severities of the number of risk incidents. Further, the processing circuitry (120) is configured to determine at least one risk score for deployment of the at least one traffic participant (12, 14) at the deployment location based on the aggregated severities.

Description

[0001] Description

[0002] Method and apparatus for determining a deployment risk of at least one traffic participant in a deployment location

[0003] The present disclosure relates to a method and an apparatus for determining a deployment risk of at least one traffic participant in a deployment location. Further, the present disclosure relates to a non-transitory machine-readable medium and a computer program.

[0004] The deployment of autonomous vehicles, i.e. at least partially automated driving vehicles, introduces a new type of risk. Different types of risk may be of interest to different types of stakeholders. For instance, city developers and regulators may be interested in the safety impact of an AV while insurance companies focus more on financial impact. Further, developer and / or deployer of autonomous vehicles may consider the risk of deployment during development and / or deployment of autonomous vehicles.

[0005] Risk may be assessed by frequency and severity of certain events, e.g., accidents. However, for new endeavors, such as the deployment - a combination of operating conditions and location - of autonomous vehicles on public roads, there is not enough historical data that can be used to estimate its risk in this way. Therefore, another method may be useful to estimate the risks caused by the deployment of autonomous vehicles before they hit the road and have an actual impact on safety. Even if autonomous vehicles are already driving on public roads, collisions are so rare that it is 1) very cost-intensive and 2) takes a long time to collect enough data to estimate the risk based on a frequency and severity analysis for specific locations, both in real- world as well as simulation environments - especially when considering their constant software updates.

[0006] Hence, there may be a need for improved risk assessment of deploying autonomous vehicles.

[0007] This need is met by a method for determining a deployment risk of at least one traffic participant in a deployment location, and an apparatus for determining a deployment risk of at least one traffic participant in a deployment location.

[0008] It is noted that in this disclosure, independent of the grammatical term usage, individuals with male, female or other gender identities are included within the term. According to a first aspect, there is provided a method for determining a deployment risk of at least one traffic participant in or at a deployment location. The method comprises receiving at least one driving data set comprising dynamic information of the at least one traffic participant. The at least one driving data set is associated with the deployment location. The method further comprises determining a number of risk incidents based on the driving data set. Further, the method comprises determining, for the number of risk incidents, a respective severity of the risk incident. In addition, the method comprises aggregating the severities of the number of risk incidents. The method further comprises determining at least one risk score for deployment of the at least one traffic participant at the deployment location based on the aggregated severities.

[0009] The proposed method allows for an improved risk assessment of deploying the at least one traffic participant in or at the deployment location. The method may be applicable to any kind of traffic participant, such as an autonomous vehicle, a manually driven car, a motorbike, a bicyclist, a pedestrian, a truck, mining vehicles, or the like. The term vehicle may also include trucks, construction site vehicles, etc. The deployment location is also not limited to a public road, but may also be a construction site, mine, or the like. The method provides a framework for determining, e.g. calculating, etc., the deployment risk score (DRS) and / or location risk index (LRI) for a given actor, i.e. the at least one traffic participant, in a deployment location, such as a specific area, route, or the like, and / or, optionally, under certain conditions, e.g. weather, traffic volume, at day, at night, or the like. The method allows for viewing and determining the risk of deployment as a function, e.g. over time and / or space of likelihood and / or frequency and severity of safety critical interactions between road actors, i.e. traffic participants, such as vehicles, motorbikes, pedestrians, etc. Further, the risk of deployment may be determined over time at a timestep level and / or a event level. In addition, the risk of deployment may be determined over space at various levels of granularities. Further, the risk may be calculated using simulation and / or real-world data. Furthermore, the at least one risk score is explainable and preserves scalar multiplication, which assures that “twice as risky” reflects e.g. twice the risk, e.g. twice the costs, when selecting costs as the severity measure.

[0010] The method may be computer-implemented and may be carried out by any suitable data processor, computation device, apparatus, or the like. The method may be carried out by a single entity or by multiple entities, a distributed computer system, etc.

[0011] As used herein, the terms “autonomous vehicle” and “at least partially automated driving vehicle” may be understood as mutually interchangeable. The term “vehicle” may include motor vehicles in general such as passenger automobiles including sports utility vehicles (SUV), buses, trucks, various commercial vehicles, and the like, and includes hybrid vehicles, electric vehicles, plug-in hybrid electric vehicles, hydrogen-powered vehicles, and other alternative fuel vehicles (e.g., fuels derived from resources other than petroleum). The terms “autonomous” and / or “at least partially automated driving” may be understood that such vehicle utilizes an at least semi-automated driving system, e.g. according to an SAE level 1 to 5 or the like.

[0012] As used herein, the at least one traffic participant may be any traffic participant involved. The least one other traffic participant may interact with another traffic participant and / or any other object involved. For example, the traffic participant may be any one of a vehicle, such as a car, truck or bus, a bicycle, a motorbike, and / or a pedestrian. Pedestrians, cyclists and / or motorcyclists may also be referred to as vulnerable road users, as these are exposed to a higher risk, in particular the risk of injury.

[0013] Further, as used herein, the risk incident may be understood as any event involving the at least one traffic participant that is critical to safety of the at least one traffic participant. The risk incident may also be referred to as safety incident or safety event. For example, the risk incident may comprise or may be any interaction between the at least one traffic participant and at least one other actor, such as another vehicle, motorbike, bicycle, pedestrian, etc., or other object, to which a security risk may be assigned. The severity of the risk incident may be or may comprise at least one of a bodily injury severity and a property damage severity.

[0014] The risk score(s) as referred to herein, may be understood as a quantitative measure of the deployment risk for a given actor, i.e. the at least one traffic participant, in the deployment location, e.g. a specific road segment or section, a route, or the like. It may also be determined for deployment under certain conditions, such as a certain weather, a certain traffic volume or density, or the like.

[0015] For example, the determination(s) of the number of risk incidents, their severities, and / or the risk score(s) may be performed with a number of parameters that may be configured prior to the determination, e.g. computation. The number of parameters may comprise at least one of the following: frequency of determination, e.g. computation, exposure radius, event and / or safety or risk metric, location discretization, and scaling. The frequency of determination may be understood that depending on a sampling rate of the dynamic driving data, the determination of the risk score etc. may be conducted at every n-th timestep t, and, optionally, adapted to a degree that is resource efficient. The exposure radius may be understood as a radius around a given traffic participant, e.g. an ego vehicle or the like, considered for the determination. The exposure radius may either be static, e.g. x meters or the like, or dynamic, e.g. with an increase or decrease with the traffic participant’s velocity by e.g. a factor of y, or the like. The safety or risk metric may understood that the risk (score) is determined, e.g. computed, based on the exposure and / or severity of certain events at a given time t. The metric used for the risk incident determination may be selected from a list, e.g. based on time-to-collision (TTC), near misses, reachability, etc. Depending on the event type, an appropriate severity metric may be selected. Furthermore, this may be weighted with a closeness factor, if applicable, as e.g. not all near misses are critical to the same extend. The location discretization may be understood that each timestep t may be mapped to a position of the at least one traffic participant which may be mapped to a discrete location on the map. The discretized locations may be used to allow for an easy aggregation of multiple driving data sets (e.g. simulations) and comparison of various scenarios. Possible approaches, among others, may comprise at least one of tracklets based on the traffic participant’s route and squares or hexagons over a 2D map, or the like. Locations can be discretized on different levels of granularity and can be flexibly aggregated. Scaling options may be used to map aggregated risks into certain range of risk values, e.g. [0,1],

[0016] According to an embodiment, the method may further comprise generating output data indicating the at least one risk score. The output data indicating the at least one risk score for the specific location may be output in electronic and / or machine-readable form for further processing, visualization, or the like.

[0017] In an embodiment, the at least one risk score may be provided and / or used for deployment of an autonomous vehicle. The knowledge of the at least one risk score for the deployment location may be used in a variety of ways in connection with the deployment of an autonomous vehicle. For example, the at least one risk score may be used in the planning of a vehicle deployment or a vehicle fleet deployment, in route planning, in driving strategy planning, in driving trajectory planning, in planning an area of deployment of an autonomous vehicle, in the determination of a cost distribution for the vehicle deployment, in the calculation of insurance fees, or the like, wherein this is not limited herein. Further, the at least one risk score may be used by city developers and regulators, which may be interested in the safety impact of an autonomous vehicle.

[0018] According to an embodiment, the deployment location may be at least one of a geographical location, a fictional location, a geographical area, a road section, a road area, a traffic area, a road route, and a travel route. In an embodiment, the severities may be aggregated over time, per driving data set. In other words, the at least one risk score may be a series of aggregated severities of determined, e.g. detected, risk incidents and / or safety events for a single driving data set that can be mapped over time. The severities may be optionally weighted, e.g. by closeness to become a collision when the risk incident and / or safety event is a near miss. As used herein, a near miss event, or more generally a near miss, may be understood as any event that came close to becoming a collision but did not become one, e.g. due to an evasive maneuver by either one or both the vehicles and / or traffic participants involved. Based on both real-world data and simulation data, near miss events occur more frequently than actual collisions. The worst point may also be referred to as a critical point in time, where the near miss is closest to becomes a collision.

[0019] According to an embodiment, the severities may be aggregated over space, per driving data set. In other words, the risk score may be a series of aggregated severities of determined, e.g. detected, risk incidents and / or safety events for a single driving data set that can be mapped over space. For example, a drivable area may be divided into road sections and intersections as defined by standard maps such as OpenStreetMap, OpenDrive maps, Lanelet2 maps or the like. Further, for example, a ground surface may be divided into easily scalable, uniformly discretized surfaces such as H3 discretization, S2 discretization. In addition, for example, the severities may be aggregated over trips or routes taken by something like an autonomous shuttle service where the risk over a block, zip-code, district or any other broader area which needs to be analyzed.

[0020] In an embodiment, each severity of the number of risk incidents may be associated with a corresponding risk incident location. In other words, the respective risk incident determined and / or identified from the driving data set may be indicated by the incident location and / or the severity. This result may be stored in a database for further aggregation.

[0021] According to an embodiment, the method may further comprise receiving location data comprising map data of the deployment location. The location data and / or map data may be provided by a suitable map provider or the like and received from there or from a data storage or the like. Further, the map data may be real or fictional.

[0022] In an embodiment, the method may further comprise performing location discretization based on the map data. For example, discretized locations may be used to allow for an easy aggregation and comparison of multiple driving data sets. Accordingly, the risk score may be determined as a function of location discretization.

[0023] According to an embodiment, the number of risk incidents and its severity may be determined for a number of timesteps with respect to processing of the driving data set, and each of the number risk incidents and its severity is mapped to a location in the map data.. For example, the risk (score) may be computed based on severity, , of certain events at a given time t, further optionally including duration.

[0024] In an embodiment the location discretization may comprise at least one of determining tracklets based on a route of the at least one traffic participant and determining squares or hexagons over map data. This is not limited herein.

[0025] According to an embodiment, the risk score may be determined location based. In other words, the at least one risk score may be determined for e.g. an entire, location, route, intersection, or the like. This allows to filter for different locations.

[0026] In at least some embodiments, the at least one risk score may be normalized with respect to e.g. a length of road segment, a size of an area, a duration spent on a specific location, road segment, area, etc.

[0027] In an embodiment, the location-based determination may be carried out for at least one of a geographical location, a fictional location, a geographical area, a road section, a road area, a traffic area, a road route, and a travel route.

[0028] According to an embodiment, the risk score may be determined condition based. In other words, the risk score may be determined for e.g. a certain weather condition, traffic volume, lighting, or the like.

[0029] In an embodiment, the condition-based determination may be carried out for at least one of a certain weather condition, a certain traffic density or traffic volume, and a certain lighting.

[0030] According to an embodiment, multiple driving data sets may be received. The number of risk incidents and their severities may be determined across the multiple driving data sets for a specific location and / or specific condition. The risk score may be determined based on the severities of the multiple driving data sets. In other words, the, optionally weighted, severities of multiple driving data sets may be aggregated for a given scenario, e.g. for a specific route and / or under certain conditions and turned into a risk score.

[0031] For example, in each individual, discretized location, risk incidents, their severities, and optionally closeness, across all driving datasets of the same scenario, e.g. traffic, weather, lighting or illumination, etc., are aggregated, e.g. averaged, ensuring statistical sufficiency. Further, the aggregation may also be carried out across multiple or even all scenarios. By way of example, costs as a measure of severity for all safety events occurred in a discretized location may be aggregated, e.g. accumulated, divided by the number of drive throughs. Further, this value may then be normalized, e.g. by length or the like, and e.g. linearly, mapped into a [0,1] set.

[0032] In an embodiment, the risk scores of the multiple driving data sets may be aggregated across multiple locations and / or multiple conditions and / or scenarios. In other words, once risk scores of [0,1] are available for all discretized locations and scenarios, aggregations of choice may be conducted in two dimensions, namely location and / or condition based, in an arbitrary order. For example, aggregating across locations may be understood that for a list of discretized locations, the risk scores may be combined, e.g. averaged. Optionally, the risk of each discretized location may be weighted, e.g. by the length of a road segment, number of events or average time spent in this section. This aggregation may be conducted on a subset of scenarios, e.g. when interested in the risk score of a route under a specific condition, e.g. in rain. Alternatively or additionally, to reflect the risk of a deployment involving several scenarios, e.g. weather conditions, traffic conditions etc., the risk scores can be aggregated by weighting them, e.g. according to their likelihood of occurrence.

[0033] According to an embodiment, the method may further comprise determining an overall risk score for deployment of the at least one traffic participant at the deployment locations. In other words, once aggregated over all locations and / or scenarios, a single number between [0,1], wherein the deployment risk score may characterize the driving risk of the deployment locations and operating domain of choice.

[0034] In an embodiment, the number of risk incidents may be determined by applying at least one metric. For example, the at least one metric may be based on at least one of a time-to-collision between traffic participants involved in the driving data set, a near miss event traffic participants involved in the driving data set, and / or a reachability of traffic participants involved in the driving data set. In other words, the risk may be calculated as a function of frequency and severity values of a road safety metric or a combination of multiple road safety metrics with a severity choice and a closeness choice such as: a. time-to-collision (TTC) with cost of potential collision weighted by the maximum collision probability, b. a responsibility-sensitive-safety (RSS) defined minimum distance, c. near-collisions and / or near misses with the associated costs weighted by the closeness to become a collision, d. collisions with associated severity ranking, MAIS level or associated costs, and / or e. reachability with cost of potential collision weighted by the likelihood of a trajectory overlap, etc.

[0035] According to an embodiment the determining of the number of risk incidents may be further based on an exposure radius or distance around the at least one traffic participant. The exposure radius may be understood as a radius around a given traffic participant, e.g. an ego vehicle or the like, considered for the determination. The exposure radius may either be static, e.g. x meters or the like, or dynamic, e.g. with an increase or decrease with the traffic participant’s velocity by e.g. a factor of y, or the like.

[0036] In an embodiment, the at least one traffic participant may comprise or may be at least one of a vehicle, pedestrian, motorbike, and a bicycle.

[0037] According to an embodiment, the at least one driving data set may further comprise static information of the at least one traffic participant.

[0038] According to an embodiment, the static information of the at least one traffic participant may comprise at least one of a dimension and a mass of the at least one traffic participant.

[0039] In an embodiment, the dynamic information may comprise trajectory data of the at least one traffic participant. The trajectory data may be derived from real world data and / or simulation data. The trajectory data may refer to at least one traffic situation. The traffic situation may also refer to at least one maneuver captured in or derivable from the trajectory data.

[0040] In a further aspect, there is provided an apparatus for determining a deployment risk of at least one traffic participant in a deployment location. The apparatus comprises interface circuitry and processing circuitry coupled to each other. The interface circuitry is configured to receive at least one driving data set comprising dynamic information of the at least one traffic participant, the at least one driving data set being associated with the deployment location. The processing circuitry configured to determine a number of risk incidents based on the driving data set. Further, the processing circuitry is configured to determine, for the number of risk incidents, a respective severity of the risk incident. The processing circuitry is further configured to aggregate the severities of the number of risk incidents. Further, the processing circuitry is configured to determine at least one risk score for deployment of the at least one traffic participant at the deployment location based on the aggregated severities.

[0041] The apparatus is configured to carry out the method according to the first aspect. Therefore, it may be modified in accordance with any one of the examples described herein. For the technical effects of the apparatus, reference is made to the above. The apparatus may be implemented as a single entity or may be distributed over multiple entities, such as a distributed computer system.

[0042] In at least some examples, the apparatus may be operationally connected to control circuitry for at least one vehicle, the control circuitry being configured to operate the at least one vehicle based on the output data. The apparatus may also be used during development of the vehicle, e.g. to generate driving software, etc. However, it may also be used in traffic planning, traffic control, etc.

[0043] In a further aspect, there is provided a non-transitory machine-readable medium having stored thereon a (computer) program having a program code for performing the method according to the first aspect and / or the method according to the second aspect, when the program is executed on a processor or a programmable hardware. Examples may also cover program storage devices, such as digital data storage media, which are machine-, processor- or computer-readable and encode and / or contain machine-executable, processor-executable or computer-executable programs and instructions. Program storage devices may include or be digital storage devices, magnetic storage media such as magnetic disks and magnetic tapes, hard disk drives, or optically readable digital data storage media, for example. Other examples may also include computers, processors, control units, (field) programmable logic arrays ((F)PLAs), (F)PGA), graphics processor units (GPU), ASICs, integrated circuits (ICs) or system- on-a-chip (SoCs) systems programmed to execute the steps of the methods described above.

[0044] In a further aspect, there is provided a (computer) program having a program code for performing the method according to the first aspect, when the program is executed on a processor or a programmable hardware. Thus, steps, operations or processes of different ones of the methods described above may also be executed by programmed computers, processors or other programmable hardware components. Some embodiments of apparatuses and / or methods will be described in the following by way of example only, and with reference to the accompanying figures, in which:

[0045] Fig. 1 illustrates in a schematic block diagram an exemplary apparatus for determining a deployment risk of at least one traffic participant in a deployment location according to an embodiment.

[0046] Fig. 2 illustrates exemplary functional components of an apparatus for determining a deployment risk of at least one traffic participant in a deployment location according to an embodiment.

[0047] Fig. 3 illustrates an example of aggregating severities over time for determining a deployment risk according to an embodiment.

[0048] Fig. 4 illustrates an example of aggregating severities over space for determining a deployment risk according to an embodiment.

[0049] Fig. 5 illustrates in a flowchart an example of an aggregation process for determining a deployment risk according to an embodiment.

[0050] Fig. 6 illustrates in a flow chart an exemplary method for determining a deployment risk of at least one traffic participant in a deployment location according to an embodiment.

[0051] Embodiments of the present disclosure are described more fully hereinafter with reference to the accompanying drawings. Elements that are identified using the same or similar reference signs refer to the same or similar elements. The various embodiments of the present disclosure may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are pro-vided so that this disclosure will be thorough and complete, and will fully convey the scope of the present disclosure to those skilled in the art.

[0052] Fig. 1 illustrates an exemplary apparatus 100 for determining a deployment risk of at least one traffic participant 12, 14 in a deployment location. The specific location may be at least one of a geographical location, a geographical area, a road section, a road area, a traffic area, a road route, and a travel route. The apparatus 100 comprises at least interface circuitry 110 and processing circuitry 120. The processing circuitry 120 is operatively coupled to the interface circuitry 110.

[0053] The interface circuitry 110 is configured to receive at least one driving data set 112 comprising dynamic information, and optionally static information, of the at least one traffic participant 12, 14, wherein the at least one driving data set is associated with the deployment location. For example, the optional static information of the at least one traffic participant may comprise at least one of a dimension or size, a mass, etc. of the at least one traffic participant 12, 14. Further, by way of example, the dynamic information may comprise trajectory data 10 of the at least one traffic participant 12, 14. The trajectory data may be derived from real world data and / or simulation data. The trajectory data may refer to at least one traffic situation. The traffic situation may also refer to at least one maneuver captured in or derivable from the trajectory data. It is noted that although the traffic participants 12 and 14 are each illustrated as vehicles, the at least one traffic participant 12, 14 may be any one of a pedestrian, cyclist, motorcyclist, or the like. Also the type of vehicle is not limited herein, and includes, for example, a bus, truck, or the like. The vehicle 12 may be an autonomous (driving) vehicle, i.e. an at least partially automated driving vehicle. The driving data set 112 may be at least partially derived from real- life data capturing the traffic situation, such as sensor data, video data, or the like, and / or may be at least partially derived from simulated data comprising a number or plurality of traffic situations simulations involving e.g. one of the vehicles 12, 14, which may also be referred to as ego vehicle or the like. It may also be possible to first determine the traffic situation from real-life data and then run, e.g. different, simulations for that traffic situation to obtain the driving data set 112. Further, the interface circuitry may be configured to receive location data 114 comprising map data of the deployment location. The location data 114 and / or map data may be provided by a suitable map service or the like and received from there or from a data storage or the like.

[0054] The processing circuitry 120 is configured to receive and process the driving data set 112 and / or the location data 114. Further, the processing circuitry 120 is configured to determine a number of risk incidents based on the driving data set 112. The processing circuitry 120 is further configured to determine, for the number of risk incidents, a respective severity of the risk incident. In addition, the processing circuitry 120 is configured to aggregate the severities of the number of risk incidents. Further, the processing circuitry 120 is configured to determine a risk score for deployment of the at least one traffic participant at the deployment location based on the aggregated severities. For instance, the processing circuitry 120 may be a single dedicated processor, a single shared processor, or a plurality of individual processors, some of which or all of which may be shared, a digital signal processor (DSP) hardware, an application specific integrated circuit (ASIC), a neuromorphic processor or a field programmable gate array (FPGA). The processing circuitry 120 may optionally be operatively connected to, e.g., read only memory (ROM) for storing software, random access memory (RAM) and / or non-volatile memory. Optionally, the processing circuitry 120 may be operatively connected to a network controller to communicate via a network, for example, in order to remotely control an autonomous vehicle, e.g. the at least one traffic participant 12, 14, perform traffic control, or the like.

[0055] Further, the processing circuitry 120 may be configured to generate output data 112 indicating the risk score. The output data indicating the risk score may be output in electronic and / or machine-readable form for further processing, visualization, or the like. In at least some embodiments, the risk score may be provided and / or used for deployment of an autonomous vehicle. The knowledge of the risk score for the deployment location may be used in a variety of ways in connection with the deployment of an autonomous vehicle. For example, the risk score may be used in the planning of a vehicle deployment or a vehicle fleet deployment, in route planning, in driving strategy planning, in driving trajectory planning, in planning an area of deployment of an autonomous vehicle, in the determination of a cost distribution for the vehicle deployment, in the calculation of insurance fees, or the like, wherein this is not limited herein. Further, the risk score may be used by city developers and regulators, which may be interested in the safety impact of an autonomous vehicle.

[0056] In at least some embodiments, the severities may be aggregated over time, per driving data set 112, e.g. by the processing circuitry 120. In other words, the risk score may be a series of aggregated severities of determined, e.g. detected, risk incidents and / or safety events for a single driving data set 112. than can be mapped over time. The severities may be optionally weighted, e.g. by closeness to become a collision when the risk incident and / or safety event is a near miss. As used herein, a near miss event, or more generally a near miss, may be understood as any event that came close to becoming a collision but did not become one, e.g. due to an evasive maneuver by either one or both the vehicles and / or traffic participants involved. Based on both real-world data and simulation data, near miss events occur more frequently than actual collisions. The worst point may also be referred to as a critical point in time, where the near miss is closest to becomes a collision. In at least some embodiments, the severities may be aggregated over space, per driving data set 112, e.g. by the processing circuitry 120. In other words, the risk score may be a series of aggregated severities of determined, e.g. detected, risk incidents and / or safety events for a single driving data set 112 that can be mapped over space. For example, a drivable area may be divided into road sections and intersections as defined by standard maps such as OpenStreetMap, OpenDrive maps, Lanelet2 maps or the like. Further, for example, a ground surface may be divided into easily scalable, uniformly discretized surfaces such as H3 discretization, S2 discretization. In addition, for example, the severities may be aggregated over trips or routes taken by something like an autonomous shuttle service where the risk over a block, zip-code, district or any other broader area which needs to be insured.

[0057] In at least some embodiments, each severity of the number of risk incidents may be associated with a corresponding risk incident location, e.g. by the processing circuitry 120. In other words, the respective severity of each risk incident determined and / or identified from the driving data set may be indicated by the incident location and / or the severity. This result may be stored in a database for further aggregation.

[0058] In at least some embodiments, location discretization of the severity and / or the risk score may be performed, e.g. by the processing circuitry 120. For example, discretized locations may be used to allow for an easy aggregation and comparison of multiple driving data sets. Accordingly, the risk score may be determined as a function of location discretization. For example, the location discretization may comprise at least one of determining tracklets based on a route of the at least one traffic participant and determining squares or hexagons over map data. This is not limited herein.

[0059] Further, in at least some embodiments, the processing circuitry 120 may be configured to determine the number of risk incidents and its severity for a number of timesteps with respect to processing of the driving data set 112, and each of the number risk incidents and its severity may be mapped to a location in the map data of the location data 114.. For example, the risk (score) may be computed based on severity, and optionally the exposure, further optionally including duration, of certain events at a given time t.

[0060] In at least some embodiments, the risk score may be determined location based, e.g. by the processing circuitry 120. In other words, the risk score may be determined for e.g. an entire, location, route, intersection, or the like. Further, the location-based determination may be carried out for at least one of a geographical location, a fictional location, a geographical area, a road section, a road area, a traffic area, a road route, and a travel route. Further, the risk score may be determined condition based, e.g. by the processing circuitry 120. In other words, the risk score may be determined for, e.g. a certain weather condition, traffic volume, lighting, or the like. For example, the condition-based determination may be carried out for at least one of a certain weather condition, a certain traffic density or traffic volume, and a certain lighting.

[0061] Further, in at least some embodiments, the processing circuitry 120 may be configured to receive multiple driving data sets 112. The number of risk incidents and their severities may be determined across the multiple driving data sets for a specific location and / or specific condition. The risk score may be determined based on the severities of the multiple driving data sets 112. In other words, the, optionally weighted, severities of multiple driving data sets may be aggregated for a given scenario, e.g. for a specific route and / or under certain conditions and turned into a risk score. For example, in each individual, discretized location, risk incidents, their severities, and optionally closeness, across all driving datasets of the same scenario, e.g. traffic, weather, lighting or illumination, etc., are aggregated, e.g. averaged, ensuring statistical sufficiency. By way of example, costs as a measure of severity for all safety events occurred in a discretized location may be aggregated, e.g. accumulated, divided by the number of drive throughs. Further, this value may then be weighted, e.g. by length or the like, and e.g. linearly, mapped into a [0,1] set.

[0062] In at least some embodiments, the processing circuitry 120 may be configured to aggregate the risk scores of the multiple driving data sets 112 across multiple locations and / or multiple conditions and / or scenarios. In other words, once risk scores of [0,1] are available for all discretized locations and scenarios, aggregations of choice may be conducted in two dimensions, namely location and / or condition based, in an arbitrary order. For example, aggregating across locations may be understood that for a list of discretized locations, the risk scores may be combined, e.g. averaged. Optionally, the risk of each discretized location may be weighted, e.g. by the length of a road segment, number of events or average time spent in this section. This aggregation may be conducted on a subset of scenarios, e.g. when interested in the risk score of a route under a specific condition, e.g. in rain. Alternatively or additionally, to reflect the risk of a deployment involving several scenarios, e.g. weather conditions, traffic conditions etc., the risk scores can be aggregated by weighting them, e.g. according to their likelihood of occurrence.

[0063] In at least some embodiments, the processing circuitry 120 may be configured to determine an overall risk score for deployment of the at least one traffic participant at the deployment locations. In other words, once aggregated over all locations and / or scenarios, a single number between [0,1], wherein the deployment risk score may characterize the driving risk of the deployment locations and operating domain of choice.

[0064] Further, in at least some embodiments, the number of risk incidents may be determined by applying at least one metric. For example, the at least one metric may be based on at least one of a time-to-collision between traffic participants involved in the driving data set, a near miss event traffic participants involved in the driving data set, and a reachability of traffic participants involved in the driving data set. In other words, the risk may be calculated as a function of frequency and severity values of a road safety metric or a combination of multiple road safety metrics with a severity choice and a closeness choice such as: a. time-to-collision (TTC) with cost of potential collision weighted by the maximum collision probability, b. Responsibility- sensitive-safety (RSS), c. near-collisions and / or near misses with the associated costs weighted by the closeness to become a collision, d. collisions with associated severity ranking, MAIS level or associated costs, and / or e. reachability with cost of potential collision weighted by the likelihood of a trajectory overlap, etc.

[0065] For example, the at least one metric may be based on a near collision and / or near miss, i.e. information rich non-collision interactions of traffic participants. The processing circuitry 120 may be configured to receive trajectory data indicating trajectory information of a traffic participant, e.g. traffic participant 12, under consideration and at least one other traffic participant, e.g. traffic participant 14, of a traffic situation at a specific location, e.g. the deployment location. The processing circuitry 120 may be configured to determine at least one near miss event involving the traffic participants based on the trajectory data. Further, the processing circuitry 120 may be optionally configured to determine, for the at least one near miss event, a worst point where the near miss is closest to become a collision. In addition, the processing circuitry 120 may be configured to determine a safety criticality of the at least one near miss event based on the worst point, the safety criticality considering a closeness to becoming a collision and an estimated severity in case of such collision. This uses information rich non-collision interactions of traffic participants, i.e. the at least one near miss event, which are more frequently than actual collisions between traffic participants.

[0066] In at least some embodiments, the determining of the number of risk incidents may be further based on an exposure radius or distance around the at least one traffic participant. The exposure radius may be understood as a radius around a given traffic participant, e.g. an ego vehicle or the like, considered for the determination. The exposure radius may either be static, e.g. x meters or the like, or dynamic, e.g. with an increase or decrease with the traffic participant’s velocity by e.g. a factor of y, or the like.

[0067] Fig. 2 illustrates an exemplary system 200 for determining a deployment risk of at least one traffic participant in a deployment location. The apparatus 100 described herein may be or may form part of the system 200.

[0068] Functionally and / or structurally, the system 200 may be divided into three subsystems 210, 220, 230, as indicated in Fig. 2. By way of example, subsystem 210 may be or may form a data access layer. Subsystem 220 may be or may comprise a server or other suitable computing device. Subsystem 230 may be or may comprise a client computer. The apparatus 100 may form, for example, the subsystem 220 and / or at least a part of the subsystem 220 and / or the subsystem 230. The subsystems 210, 220, 230 may be operationally coupled with each other, e.g. in the way as indicated in Fig. 2.

[0069] The subsystem 210, e.g. data access layer, comprises a file storage 212. For example, the file storage 212 may be configured to receive, store and / or provide one or more driving data sets 112 and / or location data 114. Further, the subsystem 210 comprises a database 214. For example, the database 214 may be configured to receive, store and / or provide computation results of the subsystem 220.

[0070] The subsystem 220, e.g. server or other computing device, comprises one or more functional modules, which may be implemented in software and / or hardware. The subsystem 220 may implement the functions of the apparatus 100 as described herein. It comprises a driving data processing module 222, a risk analysis module 224, a location discretization module 226, and an aggregation module 228. In the driving data processing module 222, the driving data is read from the file storage 212 and processed for further analysis. In the risk analysis module 224, the risk computation is performed at the configured frequency, with a given exposure radius using a given safety and severity metric. The results of this step, incident location and severity, are stored in the database 214 for further aggregation. If a type of location discretization is known prior to aggregation, the results may also be stored on a discretized location level (option 1), which may be more resource efficient. If the location discretization is defined after the risk analysis is performed, e.g. to keep multiple options open, the results may be stored along with the coordinates of the incident that may later be mapped to a discretized location (option 2). Within the risk aggregation module 228, normalized risk scores are computed, e.g. for different types of configurations. These configurations may, for example, be location-based, e.g. for an entire location, route, intersection, etc., and / or condition-based, e.g. for certain weather conditions, traffic volumes, lighting, etc.

[0071] The subsystem 230 may comprise or may be configured to run a client application 232 to start the processing of the driving data 112 etc. and / or the analyzing of the risk.

[0072] Fig. 3 illustrates an example 300 of aggregating severities over time for determining a risk factor.

[0073] In Fig. 3, the abscissa axis, i.e. the x-axis, represents time, e.g. seconds (s), minutes (m), etc. The ordinate axis, i.e. the y-axis, represents in the lower part event severity in the upper part aggregated severity. Reference sign 310 denotes the number of risk incidents, which may also be referred to as events, determined from the driving data set 112.

[0074] Fig. 4 illustrates an example 400 of aggregating severities over space for determining a risk factor according to an embodiment.

[0075] In Fig. 4, the abscissa axis, i.e. the x-axis, represents space, e.g. a road section, route, etc. The ordinate axis, i.e. the y-axis, represents in the lower part severity critical events, which are commonly denoted by reference sign 410.

[0076] Fig. 5 illustrates in a flowchart an example 500 of an aggregation process for determining a deployment risk. For example, the aggregation process may be implemented in at least one of the apparatus 100 and the system 200 as described herein.

[0077] At operation 502 (Start aggregation), the aggregation process starts. At operation 504, each individual, discretized location, risk incidents, their severities, and optionally closeness, across all driving datasets 112 of the same scenario, e.g. traffic, weather, illumination, etc., are aggregated, e.g. averaged, ensuring statistical sufficiency. For example, this may comprise accumulating costs as a measure of severity for all risk incidents occurred in a discretized location, divided by the number of drive throughs. At operation 506, this value may then be weighted, e.g. by length or the like, and, at operation 508, e.g. linearly, mapped into a [0,1] set, i.e. the risk score.

[0078] Once risk scores of [0,1] are available for all discretized locations and scenarios, aggregations of choice may be selected in two dimensions, location based and condition based, in an arbitrary order. In Fig. 5, operations 510, 512 and 514 form a first branch and operations 516, 518 und 520 form a second branch. The first branch and the second branch form alternative aggregation operations. In the first branch, i.e. operations 510, 512 and 514, the aggregation is performed across locations. For example, for a list of discretized locations, the risk scores may be combined, e.g. averaged. Optionally, the risk of each discretized location may be weighted, e.g. by the length of a road segment, number of events or average time spent in this section. This aggregation may be conducted on a subset of scenarios, when interested in the risk score of a route under a specific condition, e.g. in rain. In the second branch, i.e. operations 516, 518 and 520, the aggregation is performed across scenarios. For example, to reflect the risk of a deployment involving several scenarios, e.g. weather conditions, traffic conditions etc., the risk scores can be aggregated by weighting them, e.g. according to their likelihood of occurrence.

[0079] At operation 522, once aggregated over all locations and scenarios, a single number between [0,1], i.e. the risk score for deployment, which may also be referred to as deployment risk score, is output. It characterizes the driving risk of the deployment locations and operating domain of choice.

[0080] Fig. 6 illustrates in a flow chart an exemplary method 600 for determining a deployment risk of at least one traffic participant in a deployment location.

[0081] The method 600 comprises receiving 610 at least one driving data set comprising dynamic information of the at least one traffic participant, the at least one driving data set being associated with the deployment location. The method further comprises determining 620 a number of risk incidents based on the driving data set. Further, the method comprises determining 630, for the number of risk incidents, a respective severity of the risk incident. In addition, the method comprises aggregating 640 the severities of the number of risk incidents. Further, the method comprises determining 650 a risk score for deployment of the at least one traffic participant at the deployment location based on the aggregated severities.

Claims

Patent claims1. A method (600) for determining a deployment risk of at least one traffic participant (12, 14) in a deployment location, the method comprising: receiving (610) at least one driving data set comprising dynamic information of the at least one traffic participant, the at least one driving data set being associated with the deployment location; determining (620) a number of risk incidents based on the driving data set; determining (630), for the number of risk incidents, a respective severity of the risk incident; aggregating (640) the severities of the number of risk incidents; and determining (650) at least one risk score for deployment of the at least one traffic participant at the deployment location based on the aggregated severities.

2. The method of claim 1 , further comprising: generating output data indicating the at least one risk score.

3. The method of claim 1 or 2, wherein the at least one risk score is provided and / or used for deployment of an autonomous vehicle.

4. The method of any one of the preceding claims, wherein the deployment location is at least one of a geographical location, fictional location, a geographical area, a road section, a road area, a traffic area, a road route, and a travel route.

5. The method of any one of the preceding claims, wherein the severities are aggregated over time, per driving data set.

6. The method of any one of the preceding claims, wherein the severities are aggregated over space, per driving data set.

7. The method of any one of the preceding claims, wherein each severity of the number of risk incidents is associated with a corresponding risk incident location.

8. The method of any one of the preceding claims, further comprising: receiving location data comprising map data of the deployment location.

9. The method of claim 8, further comprising: performing location discretization based on the map data.

10. The method of claim 8 or 9, wherein the number of risk incidents and its severity are determined for a number of timesteps with respect to processing of the driving data set, and each of the number risk incidents and its severity is mapped to a location in the map data.

11. The method of claim 9 or 10, wherein the location discretization comprises at least one of determining tracklets based on a route of the at least one traffic participant and determining squares or hexagons over map data.

12. The method of any one of the preceding claims, wherein the at least one risk score is determined location based.

13. The method of any one of the preceding claims, wherein the at least one risk score is determined condition based.

14. The method of claim 13, wherein the condition-based determination is carried out for at least one of a certain weather condition, a certain traffic density or traffic volume, and a certain lighting.

15. The method of any one of the preceding claims, wherein multiple driving data sets are received, the number of risk incidents and their severities are determined across the multiple driving data sets for a specific location and / or specific condition, and the risk score is determined based on the severities of the multiple driving data sets.

16. The method of claim 15, wherein the risk scores of the multiple driving data sets are aggregated across multiple locations and / or multiple conditions.

17. The method of claim 16, further comprising: determining an overall risk score for deployment of the at least one traffic participant at the deployment locations.

18. The method of any one of the preceding claims, wherein the number of risk incidents is determined by applying at least one metric based on at least one of a time-to-collision betweentraffic participants involved in the driving data set, a near miss event traffic participants involved in the driving data set, and / or a reachability of traffic participants involved in the driving data set.

19. The method of claim 18, wherein the determining of the number of risk incidents is further based on an exposure radius or distance around the at least one traffic participant.

20. The method of any one of the preceding claims, wherein the at least one traffic participant (12, 14) comprises at least one of a vehicle, pedestrian, motorbike, and a bicycle.

21. The method of any one of the preceding claims, wherein the at least one driving data set further comprises static information of the at least one traffic participant (12, 14).

22. The method of claim 22, wherein the static information of the at least one traffic participant (12, 14) comprises at least one of a dimension and a mass of the at least one traffic participant.

23. The method of any one of the preceding claims, wherein the dynamic information comprises trajectory data of the at least one traffic participant (12, 14).

24. An apparatus (100) for determining a deployment risk of at least one traffic participant (12, 14) in a deployment location, the apparatus comprising: interface circuitry (110) configured to: receive at least one driving data set (112) comprising dynamic information of the at least one traffic participant (12, 14), the at least one driving data set being associated with the deployment location; and processing circuitry (120) configured to: determine a number of risk incidents based on the driving data set; determine, for the number of risk incidents, a respective severity of the risk incident; aggregate the severities of the number of risk incidents; and determine at least one risk score for deployment of the at least one traffic participant (12, 14) at the deployment location based on the aggregated severities.

25. A non-transitory machine-readable medium having stored thereon a program having a program code for performing the method according to any one of claims 1 to 23, when the program is executed on a processor or a programmable hardware.

26. A computer program having a program code for performing the method according to any one of claims 1 to 23, when the program is executed on a processor or a programmable hardware.

Images