6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment
Maike Scholtes, Lukas Westhofen, Lara Ruth Turner, Katrin Lotto, Michael Schuldes, Hendrik Weber, Nicolas Wagener, Christian Neurohr, Martin Bollmann, Franziska Körtke, Johannes Hiller, Michael Hoss, Julian Bock, Lutz Eckstein
PPREPRINT, submitted to IEEE Access
Digital Object Identifier PREPRINT, submitted to IEEE Access
M. SCHOLTES , L. WESTHOFEN , L. R. TURNER , K. LOTTO , M. SCHULDES , H. WEBER ,N. WAGENER , C. NEUROHR , M. H. BOLLMANN , F. KÖRTKE , J. HILLER , M. HOSS , J.BOCK AND L. ECKSTEIN Institute for Automotive Engineering (ika), RWTH Aachen University, 52074 Aachen, Germany OFFIS e.V., 26121 Oldenburg, Germany ZF Friedrichshafen AG, 88046 Friedrichshafen, Germany fka GmbH, 52074 Aachen, Germany Corresponding author: M. Scholtes (e-mail: [email protected]).The research leading to these results is funded by the German Federal Ministry for Economic Affairs and Energy within the project ‘VVM- Verification & Validation Methods for Automated Vehicles Level 4 and 5’.
ABSTRACT
Verification and validation of automated driving functions impose large challenges. Currently,scenario-based approaches are investigated in research and industry, aiming at a reduction of testing effortsby specifying safety relevant scenarios. To define those scenarios and operate in a complex real-world designdomain, a structured description of the environment is needed. Within the PEGASUS research project, the6-Layer Model (6LM) was introduced for the description of highway scenarios. This paper refines the 6LMand extends it to urban traffic and environment. As defined in PEGASUS, the 6LM provides the possibilityto categorize the environment and, therefore, functions as a structured basis for subsequent scenariodescription. The model enables a structured description and categorization of the general environment,without incorporating any knowledge or anticipating any functions of actors. Beyond that, there is a varietyof other applications of the 6LM, which are elaborated in this paper. The 6LM includes a description ofthe road network and traffic guidance objects, roadside structures, temporary modifications of the former,dynamic objects, environmental conditions and digital information. The work at hand specifies each layerby categorizing its items. Guidelines are formulated and explanatory examples are given to standardize theapplication of the model for an objective environment description. In contrast to previous publications, themodel and its design are described in far more detail. Finally, the holistic description of the 6LM presentedincludes remarks on possible future work when expanding the concept to machine perception aspects.
INDEX TERMS
I. INTRODUCTION A S automated driving (AD) constantly increases in im-portance [1], a large challenge faced when implement-ing (highly) automated driving (HAD) functions is testingand validation of such functions. In [2], the issue arisingwhen trying to validate HAD through real-world drives, theso-called ‘approval trap’, is described. The resulting amountof kilometers that needs to be driven for this distance-basedstatistical validation approach is not feasible due to timeand cost reasons. This motivates the idea of scenario-basedverification and validation, in which specific safety-relevant scenarios guide the testing process [3].In order to use the scenario-based method for automatedvehicles operating in an open context [4], i.e. an unstructuredreal-world operational design domain, a sufficiently completedescription of the environment is needed. To decrease thecomplexity and provide a structured method the environmentcan be described by utilizing the 6-Layer Model (6LM)which was already introduced in previous work (see SectionII). Within the German research project PEGASUS and thecontext of defining highway scenarios [5], [6], the conceptwas applied to separate relevant aspects of the environment
VOLUME -, 2021 a r X i v : . [ c s . OH ] F e b choltes et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment description into different layers that are built upon each other.When extending the approach to urban use cases manyother aspects need to be discussed. Although urban structureswere already categorized before, see, in particular, [7], theywere – due to the different focus on highways – not furtherconsidered in PEGASUS and the following publications.Additionally, we could observe a rather discontinuous, evenconflicting, evolution of the Layer Model. Entities and prop-erties were shifted between layers and their naming was oftenchanged such that numerous ambiguities exist today. For thisreason, we see the need for a proper definition of the 6LMwhich will be given in this work to make the model a moreaccessible tool for structured environment description to re-searchers and safety engineers. The six layers presented inthis paper heavily build upon existing literature as indicatedin the definitions and examples of Section IV and V. Theclassification of objects from previous work is used whereappropriate, but clarified or adapted where needed with thegoal to provide a single and consistent reference of the 6LMfor all use cases.The remainder of this work is structured as follows. Anoverview of previous publications dealing with environmentdescriptions based on a layered model is given in Section II.Furthermore, in Section III, the scope of the refined model aswell as some motivation on where it can be used is presented.In the following section, we extend the model by a detaileddescription of each layer including the naming and a com-parison to previous definitions. We propose a categorizationof relevant traffic entities and their properties into a high-level classification supported by interesting examples. Thedescription of the model is followed by eight guidelines.Explanatory examples for each guideline reveal how to cat-egorize different entities with their properties and providejustification for the definition of the layers. Subsequently,the presented definitions and guidelines are applied to a real-world example. The work is concluded by an outlook ontofuture work and a summary of the described approach. II. RELATED WORK
In order to understand the 6LM and its applications, it isimportant to look more closely on the definition of the termscenario. Definitions of scene and scenario are given in [8],[9] and in a DIN SAE Spec [10]. While a scene is describedas a snapshot of the environment, a scenario describes thetemporal development of those snapshots. Therefore, a sce-nario features a certain time span. This is important to notefor the following description of the 6LM.The concept of using a layered model to structure anenvironment description for scenes and scenarios was firstintroduced in [7]. In his Ph.D. thesis [11], the lead authorof the previous publication refined and adapted the model.There, the model featured four layers: The first layer for thebase road network, the second layer for situation-specificadaptations of the road network, a third layer to describethe actors and their control and a last layer for environmentconditions. Within this concept, only hierarchical higher ranked layers could influence lower ranked layers. All layersintroduced in [11] can be found in Fig. 1, which provides anoverview of the development of the 6LM. In [11] markingsare included in the first layer while signs, guardrails and(urban) roadside structures are located in the second layer.In this reference, the second layer performs situation-specificadaptations to the road network required for special appli-cations and automated driving functions. This includes theplacement of different signs and safety structures as well asbuildings and street lamps to construct various environments.Roadwork related changes are also mentioned in this context.Therefore, Layer 2 of [11] is a combination of Layer 2 andLayer 3 of subsequent work that performs a more distinctiveclassification.Within PEGASUS [6] the concept of [11] was taken up forthe use case of highway scenarios. Subsequently, a fifth [12]and a sixth layer [13] were introduced, resulting in the 6LMdiscussed in this work. In [12] the former base road networkis split into two different layers, namely the road-level andthe traffic infrastructure, as to separate the road descriptionand the traffic rules. The road-level in [12] only containsthe layout of the road (geometry) and its topology while thelayer for traffic infrastructure contains structural boundaries,traffic signs and markings (the latter in contrast to [11]). In aGerman publication [14], which was published alongside theEnglish version [12], some small, but meaningful differencesare present. E.g., in [14], Layer 2 is named road equipmentwhile in the English translation this is changed to trafficinfrastructure.In all subsequent work ( [12]–[15]) Layer 3 is sepa-rated from the previous Layer 2 and describes solely tem-porary modifications of Layer 1 and Layer 2. This can,e.g., include modifications made when a construction siteis present. In [12] Layer 4 is named ‘Objects’. It containsall static, dynamic and movable objects that are not alreadypart of the traffic infrastructure. Furthermore, maneuvers andinteractions are situated in Layer 4. In [14] this layer isnamed slightly different (‘Movable Objects’), but containsthe same objects. However, in contrast to [12], [14] doesnot mention that ‘Movable Objects’ includes all potentiallymovable objects (static / stationary objects in [12]), i.e., alsotraffic participants that do currently not move. Layer 5, stickswith the previous definition of Layer 4 from [11] describingenvironmental conditions, such as weather.The concept in [13] is consistent with the basic concept of[12] using the five layers (and renaming them slightly): Streetlayer, traffic infrastructure, temporal modifications of Layer 1and Layer 2, movable objects and environment conditions.However, [13] introduces a sixth layer for digital information.This sixth layer is later renamed ‘Data and Communication’in [15], featuring the same definition as in the previouswork. As [13] focuses on the highway use case, structuralobjects along the road, such as buildings, are not mentionedexplicitly. Furthermore, the naming of Layer 4 as ‘MovableObjects’ makes a more definite classification. This latestversion of the model will be used as a starting point for the VOLUME -, 2021 choltes et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment
FIGURE 1.
Historic development of the 6LM. Four layers of Schuldt [11] (left) compared to the five layers of Bagschik et al. [12] (center) and the six layers of Bocket al. [13] (right). work at hand.How the 6LM, as described in [13], can be used to developa framework for scenario definition is shown in [15]. Thework performed in [15] utilizes Layer 4 of the 6LM to definelogical scenarios [10], [16] for controlled-access highways.
III. SCOPE AND MOTIVATION
Given the control loop of environment, driver and vehicle asdepicted in Fig. 2, the main focus of the 6LM as proposedin previous work is on the structured categorization of theenvironment which naturally interacts with both, the driverand the vehicle. In the field of verification and validation,
VOLUME -, 2021 et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment however, a series of additional applications of the 6LM exist.This extended scope will be motivated in this section.In the following, the notions ‘entity’, and ‘property’ areheavily used. For the scope of this work, we refer to anentity as anything that exists, has existed or will exist [17].The term ‘object’ denotes a material entity only. This meansthat traffic participants, their maneuvers, and their intentionsare all entities, but only the traffic participants themselvesare objects. A property is an attribute of an entity that canbe assigned some value, e.g., the position of a traffic sign,the size of a building, the visibility of a road marking, thevelocity of a vehicle, the intensity of a precipitation, orthe state of a traffic light. Besides properties, entities canalso have relations to each other, i.e., linking them together.For instance, a car can be connected to another car by therelation ‘drives behind’ [18]. Relations and classes of entities,e.g., the class of all vehicles, become more important in thecontext of ontologies.Consider an engineer designing scenarios for the verifica-tion and validation of AD functions. This engineer can usethe 6LM with its clear characterization of the environment,the entities and the properties as a basis for a scenariodescription. This holds independently of the utilized scenariodescription language itself, which could be natural, formal,or machine-readable. However, in order to allow for anautomatic conversion between the scenario description andthe language formats, we have to ensure that the structur-ing of the 6LM is compatible with the existing formatsOpenDRIVE [20] and OpenSCENARIO [21]. Having thisapplication in mind, the terms ‘environment’ / ‘environmentdescription’ and ‘scenario’ / ‘scenario description’, respec-tively, will often be used interchangeably within this paper.This in turn means that the 6LM will mainly concentrate onthe description of short time periods, i.e., the duration of atypical scenario. Note that clear and consistent rules for sce-nario design are particularly essential when reproducibility ofscenarios is required, e.g. for testing campaigns. Focusing onhazardous scenarios, the 6LM was already applied to struc-ture an environmental model used for iterative identificationof those scenarios [22].Similarly, the 6LM can serve as a basis for a traffic domainontology. In this paper, the authors examine how relevantdomain entities and their properties can be categorized intoa high-level classification of six layers, i.e., a flat, informallyspecified taxonomy. The canonical issue of detailing thesingle layers is part of ongoing research. Such an imple-mentation of the 6LM can be done in a formal and digitalontology, e.g., by using the Web Ontology Language [23],where the layers enable to classify entities by virtue of thewell-defined categorization.Test engineers are already utilizing the model as a struc-tured format for recording and analysis of measurement datato identify influencing factors on different layers and to fi-nally derive a scenario concept on that basis. For this purpose,it is necessary to develop a holistic, i.e., comprehensive,and well-structured environment description with all relevant environment aspects assigned to the corresponding layers.Given that, the 6LM should also be suitable to describeenvironments in different settings – urban, rural, and highway– and on different abstraction levels - macro-, micro- andnanoscopic [24].The environment description should be unbiased and actor-independent at any time. Therefore, the 6LM must not antic-ipate any function of an actor or any properties of later steps,as they might, for instance, occur in scenario extraction.Regarding the control loop of environment, driver and vehicle(see Fig. 2), this means that the 6LM is supposed to formulatea system-independent general environment description withan objective view on the traffic participants, without pro-viding information on their expected behavior. In the sameway, the 6LM should not contain any goals, values, or normssince the model is an ‘as it is’-description of the physicallyobservable only. As such the 6LM is not suitable to describesituations as defined in [9]. The above holds independentlyfrom the introduction of automation where the driver ofthe control loop is replaced with the automated system, asschematically pictured in Fig. 2. An outlook to possible newaspects of the 6LM introduced through automation such asmachine perception can be found in Section VII.The need for a method to construct (virtual) environmentswas frequently experienced in the past. In order to closethis gap and to support the generation of scenarios, therefinement of the 6LM focuses on a simple and unambiguousdesign. This also facilitates the comparison of scenarios anda subsequent reduction of given scenario sets. This iden-tification of similarities and differences can be performedon the entire description, but in most cases, it might beadvisable to perform it on single layers. The same holds fortesting and debugging of system failures. Consider a behavioror motion planning function that can be tested with roadnetwork, traffic guidance objects, and dynamic objects, butin contrast to a perception function, without any interferencethrough roadside structures and environmental conditions.A similar requirement on the layers’ independence couldbe imposed by a simulation engineer who wants to executesimplified environment simulations on selected layers or fullenvironment simulations on all layers, depending on thepower of the simulation tool at hand. Furthermore, there arecases where only the content of single layers is intendedfor exchange between different stakeholders like OEMs andTier-1s. There might be layers with common and such withmore customized content. These layers should be preferablyseparated from each other and the layers should be as self-contained as possible.To conclude this section, we look at the 6LM from adifferent – highly intuitive – perspective: Imagine the 6LMas a kind of board game where Layer 1 up to Layer 3 serveas a base description of the board. Then, Layer 4 describesthe actors with their behavior as well as comparable dynamicincidents on the board. Layer 5 and Layer 6 are somewhatseparated from this description and can be imagined abovethe board, but with influences on the board and its actors. Fi- VOLUME -, 2021 choltes et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment
FIGURE 2.
Schematic representation of the control loop between environment, automated system (for SAE Level 3/4/5 [19]) and vehicle and the focus of the 6LM. nally, this metaphor also gives an indication for the orderingof the layers. The numbering of the layers is not to rank themby importance, but purely for structuring and categorizationwhere one layer builds upon another.
IV. DEFINITION OF THE LAYERS
This section introduces the different layers of the 6LM. Foreach layer, it provides an explanatory name and categoriesof objects belonging to that layer. Table 1 provides an exem-plary, incomplete, overview of the entities within the differentlayers.Note that, together with the assignment of the entitiesto the layers, many properties like position, velocity, size,material or color can, in general, be described within therespective layers. There are, however, also compelling rea-sons to place certain properties in another layer than theobject itself. In some cases, for convenience only, this canbe bypassed by using so-called annotations. For more detailsand guidelines, we refer to Section V.
A. ROAD NETWORK AND TRAFFIC GUIDANCEOBJECTS (LAYER 1)
Layer 1 describes the road network together with all per-manent objects required for traffic guidance . For the defi-nition of ‘permanent’ compare the term ‘geo-spatially sta-tionary objects’ in [9], i.e., the objects and properties de-scribed in this layer remain unchanged within a scenario.Non-permanent traffic guidance objects are described fromLayer 3 upwards. Given the road network and traffic guidanceobjects, Layer 1 summarizes where and how traffic partici-pants can drive.The road network refers to the geometry, topology andtopography of the roads including road course, road linkage,road elevation and lateral profile. With the help of road mark- ings that are located in Layer 1, the semantics of the lanescan be derived, e.g., shoulders, cycle paths, and sidewalks.Furthermore, special areas with their boundaries such asparking spaces and keep-out areas can be clearly marked. Inthat regard, road markings also comprise instructions that arepainted on the road surface like speed limits, stopping lines,or turn arrows which are supplementary to given traffic signsor traffic lights.Both, permanently present (switchable) traffic signs andtraffic lights are placed in Layer 1 while their (changing)states will be described in Layer 6 (see Section IV-F). WithinLayer 1, we assume that the semantics of the traffic signs,traffic lights, and road markings is imposed by the GermanRoad Traffic Act (StVO) and the Vienna Convention on RoadSigns and Signals [25] and thus needs no further explanation.According to the VzKat, the catalog of traffic signs of theStVO [26], delineators and beacons like vertical panels withdistance information and before railroad crossings belongto the traffic signs and are, therefore, described in Layer 1.However, since many types of beacons, such as verticalpanels with integrated warning lights as well as guide barriersand separators, are used exclusively for the delimitation ofconstruction sites, these are no permanent guidance objectsand thus not relevant before Layer 3. In contrast to advertis-ing boards and other private signs positioned in subsequentlayers, all directional signs - e.g. city limit signs, officialtourist signs and direction signs - are placed in Layer 1.With the help of road markings and traffic signs, roundabouts,traffic islands, and bus stops can be identified unambiguously.Given all this information, it is now easy to plan a traf-fic participant’s mission: start and end point on the road,route to be followed, velocity profile, and handling of, e.g.,stopping requests. In addition to the mission planning onthis microscopic level, Layer 1 can also be used to design
VOLUME -, 2021 et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment
TABLE 1.
Layers of the 6LM including exemplary entities on the differentlayers
Layer 6 - Digital Information
State of traffic lights and switchabletraffic signsV2X messagesCellular network coverage
Layer 5 - Environmental Conditions (Artificial) IlluminationPrecipitationRoad weather (dry, wet, icy etc.)Wind
Layer 4 - Dynamic Objects
Vehicles (moving and non-moving)Pedestrians (moving and non-moving)TrailersAnimalsTrees falling over(at the current point in time)Miscellaneous objects such as balls,coke cans etc.
Layer 3 - Temporary Modifications of L1 and L2
Roadwork signsTemporary markingsCovered markingsFallen trees laying on the street
Layer 2 - Roadside Structures
BuildingsVegetationGuardrailsStreet lampsAdvertising boards and pillars
Layer 1 - Road Network and Traffic Guidance Objects
Roads including shoulders, sidewalks,parking spaces etc.Road markingsTraffic signs and traffic lights complete traffic flows on a macroscopic level. On that basis,different road networks can be compared and urban onescan be distinguished from rural ones. Moreover, an initialassessment of the complexity of scenarios is already possibleon this layer using the information on geometry, topology andtopography of the roads as well as traffic guidance objects.Lastly, we assign the road surface material, e.g., asphalt,cobblestone and gravel to Layer 1. It also includes irregulari-ties of the road surface, e.g., damage and potholes. Additionalstructures such as manhole covers and grids modifying theroad surface are described as well. The same holds for speedbumps which are often a combination of street elevation androad markings.From the board game perspective, Layer 1 is the board onwhich elements of higher layers can be placed. It extendsthe definition of the base road network [11] by including thetraffic infrastructure of [12] and [13]. To allow for a clearseparation from the following layer, however, the focus is onthose traffic infrastructure objects which are used for trafficguidance and regulation.
B. ROADSIDE STRUCTURES (LAYER 2)
In contrast to PEGASUS, current projects like V&V-Methoden [27] and SET Level [28], as well as researchand development in industry, focus on urban environments.Compared to highway scenarios, where the number of objectsbeyond the road is very limited, there are many roadsideobjects that need to be described in the urban setting. Thisjustifies a separate layer to subsume all these objects. Ourdefinition of Layer 2 contains many urban structures alreadyconsidered in the situation-specific adaptations of [11].Layer 2 addresses the roadside structures and contains all static objects that are usually placed alongside - and not onto- the road . Examples of such static objects are buildings,vegetation like trees and bushes, walls and fences, streetlamps, above ground hydrants, bollards, and other typesof fixed poles. Bus shelters with benches and surroundingconstructions like tunnels and bridges are also grouped inthis layer. The same holds for so-called vehicle restraint sys-tems that prevent vehicles from leaving the road and reducecrash severity. The safety structures include, for instance,guardrails, concrete step barriers, and impact attenuators.Bridge barriers for cyclists and pedestrians shall also be listedhere.Analogously to Layer 1, we suppose that all objects ofLayer 2 are permanently installed at a designated position.For reasons of simplicity, however, an oscillating motion ofleaves, banners or flags can also described in this layer (seealso the guidelines in Section V-A). In cases where we needa more detailed motion description or there are deviationsfrom this designated position, we refer to Layer 3 or Layer 4,respectively.Coming back to the board game example, Layer 2 canbe thought of being on top of the basic Layer 1 changingthe design of the board and increasing the complexity of thescenario at hand. Keeping simulation applications in mind, VOLUME -, 2021 choltes et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment this allows to perform simplified or detailed simulationsdepending on the chosen layers.
C. TEMPORARY MODIFICATIONS OF LAYER 1 ANDLAYER 2 (LAYER 3)
Layer 3 is comprised of temporary modifications of elementsof Layer 1 and Layer 2 . Therefore, Layer 3 does not introduceany new object classes that have not already been defined inthe previous layers. This explicitly does not mean that no newobjects can be introduced, just that they are of the same classas Layer 1 and Layer 2 objects . An explanatory example isa construction site with corresponding traffic signs and roadmarkings, modifying the course of the lane. Those elementsare guidance elements of Layer 1, i.e., of an object classalready described, but we assume they are newly introducedin Layer 3. As mentioned in Section IV-A, beacons withinconstruction sites are included in the catalog for traffic signs.The same holds for traffic cones. Therefore, they are of theclass ‘traffic guidance object’ (Layer 1), but are instantiatedin Layer 3 as they are non-permanent in the real world.Moreover, a fallen tree (compare also [29]), boulders,contamination of the road through soil or sand as well ascollapsed building parts that are lying on the street (e.g., dueto an elementary event) are described in Layer 3. The sameholds for the state change of a road marking which is coveredby the elements mentioned before. Note that all modificationsin Layer 3 are supposed to be constant for the entire durationof the scenario .Given that the scenario duration only represents a lowerbound for the word ‘temporary’, a specification of the upperbound seems to be more difficult. In [12] ‘temporary’ isdefined as a time frame of one day, although a thoroughevaluation on why this time interval was chosen is not given.To the opinion of the authors, such a justification is necessary.The following paragraph provides possible sources that canserve as orientation on the basis of which the time frame canbe chosen.In order to devise a suitable time frame for temporary mod-ifications and to decide if the choice of one day is justified,different regulations were consulted. These include guide-lines for roadwork [30], a handbook for road markings [31]and regulations for temporary buildings [32]. The Germanguidelines for roadwork [30] only differentiate between road-work of longer duration (minimum one day and stationary)and roadwork of short duration (matter of hours shorter thanone day). As already stated, this definition is not practicablewhen thinking of a holistic environment description, as thetime span of one day is too short to be an upper thresholdfor, e.g., construction work. The handbook for road markings[31] defines a category for temporary markings that existlonger than 180 days. In general, this is more in accordancewith our understanding of possible duration of roadworks andshows that the term ‘temporary’ can also include a ratherlarge time frame.Temporary building regulations [32] claim that non-permanent structures are defined as being intended to be disassembled and assembled at different locations. A specialpermit is necessary, if such an object is located in the sameposition for more than three months. Therefore, we suggestusing this threshold as an orientation for roadside structures.To conclude, the above shows that existing regulations canonly give an indication. ‘Temporary’ cannot be defined inde-pendent of the context of the modification and can be a ratherflexible notion possibly including large time frames. Someprominent public building projects are perfect examples inthis regard. Therefore, it is advisable not to determine a fixedthreshold, but to perform this classification on a case by casebasis, keeping the specific application of the 6LM in mind.Fig. 3 shows an image of an intersection that is cur-rently under construction. In the figure, entities belongingto Layer 3, i.e., representing temporary changes of Layer 1and Layer 2 due to the construction site, are highlighted inred. Those are the actual construction site itself, roadworktraffic signs, temporary traffic lights and temporary roadmarkings. It is important to note that the excavator and thelittle roadwork trailer (on the far right-hand side), whichcan also be assigned to the construction site, are not part ofLayer 3. They are defined as movable, dynamic objects andare, therefore, located in Layer 4.
D. DYNAMIC OBJECTS (LAYER 4)
Layer 4, ‘Dynamic Objects’ , is the first layer that introduces a time-dependent description . It is roughly speaking the ‘traf-fic layer’ as it includes movable objects whose movementscould evolve over time and are described by trajectories ormaneuvers. This is consistent with the definition in previouspublications within PEGASUS such as [13] and [15]. How-ever, when extending the application of the 6LM to the urbancontext, not only focusing on the SAE Level 3 [19] functionson highways, many additional entities must be described onthis layer.Layer 4 contains all “dynamic element” that, according to[9], “move (having kinetic energy), or possibly being ableto move (having sufficient energy and abilities to move)”.The latter refers to the objects that can potentially move, butdo not necessarily have to within the scenario. These objectsmight be stationary, resting in a fixed position such as parkedvehicles, pedestrians standing still and garbage cans sitting intheir pickup-position at the street etc. Note that the definitionincludes entities that are designed to perform movements, butalso comprises those entities that move on a regular basis ordue to an external trigger.The definition in [9] together with the fact that Layer 4 alsodescribes state changes that are not necessarily associated tothe entity’s movement led us to a renaming of Layer 4 as‘Dynamic Objects’ (in contrast to the ‘Movable Objects’ of[13]). An example for such a ‘dynamic’ (or time-dependent)state change is the visibility of a road marking which iscovered by increasing snowfall. In this regard, modifying theoriginal definition of dynamic objects in [8] and describingthem as “elements whose state changes in-between scenes”
VOLUME -, 2021 et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment
FIGURE 3.
Image recorded using a drone showing an intersection with roadwork. Elements of Layer 3 are marked in red. Boxes drawn by hand for illustrationpurposes only. (i.e., within a scenario) complements our understanding ofLayer 4.All traffic participants present good examples for dynamicobjects on Layer 4. These include vehicles (including trail-ers), motorcycles, bicyclists, pedestrians and rail vehiclessuch as trams. Furthermore, objects on Layer 4 includeanimals and miscellaneous objects that are movable.Let us mention coke cans lying on the road or being kickedby a child and household garbage cans that are generallyplaced near houses or pushed to and from the street for pick-up. Similar examples are balls rolling towards the street ortrees currently falling over. As part of the vegetation, a treethat is planted in a certain position is located on Layer 2. Ifthis tree happens to fall over and find its resting position onthe road, this can be understood as a temporary modificationand the fallen tree would be situated in Layer 3. However,while the tree is falling over, it features a trajectory and willbe placed on Layer 4.An image of a regular urban intersection is shown in Fig. 4.In the image, all Layer 4 elements are highlighted through redboxes. It contains vehicles (parked and moving), bicycles anda tram. In the board game example, those are the actors on theboard that has been designed through Layer 1 to Layer 3.Previous publications [12] and [13] included interactionsin their description of Layer 4. Considering the overall aimof the 6LM of giving a general environment description,i.e., a description of the physical observable only, the terminteractions might be misleading. The 6LM does not includeany non-physical entities such as goals and values.In [8] light and weather conditions are included in the term‘dynamic object’. In the case of the 6LM, light and weatherconditions are not part of Layer 4 as they are defined in a separate layer, Layer 5, ‘Environmental Conditions’.
E. ENVIRONMENTAL CONDITIONS (LAYER 5)
Layer 5 contains environmental conditions . These consist of weather, atmospheric and lighting conditions . Weather andatmospheric conditions, e.g., include precipitation, visibility(fog etc.), wind, and cloudiness. Layer 5 also includes roadweather conditions . These are weather related modificationsof the road surface like dry, wet or icy roads. As part ofthe lighting conditions, daytime related aspects such as po-sitioning of the sun (or the moon) and sunrays can be listed.Furthermore, artificial light sources like lighting throughstreet lamps or through advertising boards are included. Thecurrent layer also features a time-based description , as itis possible that environmental conditions change during ascenario.By this definition, Layer 5 is consistent with previouspublications such as [13]. However, while the content hasnot changed, the naming was adjusted to be in accordancewith the English term and various corresponding publicationsdescribing operational design domains (ODDs) [34]–[36].Note that we define Layer 5 to include globally perceptibleenvironmental conditions. This, e.g., covers whether fog ispresent or not. Actor-dependent occlusions through, for in-stance, fog or traffic participants are not covered in the 6LM.They can, however, be derived for different actors from theinformation given in the 6LM.Effects of Layer 5 conditions that fall into the category of‘globally perceptible conditions’ are presented in Layer 3 orLayer 4. If an environmental condition such as wind inducesa movement, this must be described in Layer 4 (comparealso [11]). As a consequence, the traffic cone blown away by VOLUME -, 2021 choltes et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment
FIGURE 4.
Image of an intersection recorded using a drone highlighting all Layer 4 elements. Boxes drawn by hand for illustration purposes only. strong winds or the motion of individual leaves - if desired todescribe in detail - are addressed in this layer. Layer 4 alsoincludes a road marking that continuously loses visibility.If the effect of the environmental condition is without anydynamic component and constant for the entire duration ofthe scenario, we formulate this change in Layer 3. In otherwords, the road marking which is covered by snow for the en-tire duration is described in Layer 3. Note that this approachallows the orchestration of information within the 6LM sincethe categorization of entities and properties into the differentlayers takes place in an objective way. The observable resultis the same even if the triggering event causing it is different:A traffic cone currently being overturned by wind or knockedover by a construction vehicle are both located on Layer 4, aroad marking covered by snow or lost cargo is located oneither Layer 3 or Layer 4 (depending on whether the roadmarking’s covering state is constant or not).
F. DIGITAL INFORMATION LAYER 6
Layer 6 is defined to focus on all kinds of information ex-change, communication, and cooperation on basis of digitaldata only . A sixth layer was first introduced in [13] as ‘Digital Information’ and later renamed as ‘Data and Com-munication’ in [15]. In the publication at hand, we choose theoriginal name.Analogous to previous publications, Layer 6 addressesall information between vehicles, infrastructure, or both,emerging from V2X modules and for instance transmittedwirelessly. V2X is a rather new concept without high marketpenetration today whose importance will increase with thefuture developments [37]. Let us mention information aboutroad closures due to accidents, extreme weather conditions and truck-to-truck communication in platooning applicationsas examples. Due to its importance for digital information ex-change, wireless signal coverage and strength is also assignedto this layer.In the 6LM, we differentiate between the classification ofentities and properties on the one hand and the classificationof the underlying information source at the other hand:When describing that a vehicle reduces its velocity due toan upcoming traffic jam, the vehicle’s braking action and thetraffic jam itself are assigned to Layer 4. Thereby, it does notmatter whether the traffic jam was observed by the driver (ora set of sensors in case of automation) or received through aV2X message. If, however, the information was transmittedthrough V2X, the fact that the V2X message was sent,together with its content, must be represented in Layer 6. Welike to stress that this is particularly important when the V2Xinformation differs from the ground truth. Since the receptionof V2X messages requires specialized hardware, a placementof V2X messages in Layer 6 is justified as well.Fig. 5 gives an example where the V2V message is supple-mentary, but identical, to the ground truth information. Twovehicles are approaching an intersection with an occlusionpresent. The vehicle that has the right of way intends to turnleft and, therefore, crosses the planned path of the secondvehicle. The intention of turning left can be seen throughthe indicator light (visible to the other vehicle only withoutocclusion), but it could also be conveyed through a V2V mes-sage. Without the V2V message the environment descriptionwould be incomplete. However, for the approaching vehiclethat is supposed to yield, the V2V message is relevant due tothe occlusion.Intelligent traffic management systems are assigned to
VOLUME -, 2021 et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment
FIGURE 5.
Example for relevant V2V communication at an intersection.
Layer 6, too. Part of traffic management systems are thestates of traffic lights and switchable traffic signs as they are,for instance, used for highway sign gantries. Those statesare included in Layer 6, no matter if they are controlledcentrally for dynamic traffic routing, use V2X communica-tion or feature none of the above. In any case, these statesare encoded as digital information. With increasing V2Xcommunication another example are traffic lights that informthe traffic participants about the duration of a green phaseand an appropriate velocity to cross the intersection withoutthe necessity to decelerate or stop. Similarly, we can mentionthe flashing lights at railway crossings. Note that only thevariable information, i.e., the changing states of traffic signsand traffic lights, is described here, while the respectiveobjects are already placed in Layer 1.
V. GUIDELINES FOR THE 6-LAYER MODEL
The previous section provided definitions of the layers, theirnaming and a description of what should be included inthe layers. Working with the model as a tool to generatea holistic description of the environment has shown thatambiguities in the process of assigning entities to layers,even though classifications are carefully made. The followingguidelines intend to give clarification on this matter and aremeant to show that the model is capable of providing anoverall categorization. All guidelines are stated first and aresubsequently explained along with examples in the followingsection. Furthermore, Fig. 6 reveals some aspects formulatedin the guidelines in a graphical manner.
A. GUIDELINES
1) Layers 1, 2 and 3 conduct a spatial-based description.They do not contain any time-variable aspects. Time-based descriptions are introduced from Layer 4 up-wards.2) Layer 3 contains temporary changes of Layer 1 and 2.These changes are fixed for the whole duration ofthe scenario. They are not permanent in the sense ofLayer 1 and 2.3) From Layer 3 upwards, state changes are introduced.Additionally, from Layer 4 upwards state changes can be time-dependent.4) If an entity has time-dependent properties (potentiallyvariable during a scenario), it should be placed onLayer 4 upwards. However, not all its properties needto be time-dependent.5) Not all properties of an entity are necessarily in thesame layer. The same property of a given entity should,however, not be located on different layers. If in doubtwhere to locate a property, it is placed in the layerwhere it matches the description of the layer best andhas the largest influence.6) Annotations can be used for reasons of simplicity or inorder to add extra information.7) Allegedly global properties need to be thoroughlychecked whether they are truly objective. If they arenot, they are not part of the 6LM.8) Properties of all layers can influence properties onother layers. There is no single direction of influence.
B. EXPLANATORY DESCRIPTION AND EXAMPLES FOREACH GUIDELINE
Guideline 1.
The separation of spatial-based and time-baseddescription allows the reusability of Layer 1 and 2 entitiesif one location is used multiple data recordings. The roadnetwork, traffic guidance objects and roadside structures canbe kept constant and are only changed by modifications inLayer 3. This separation of spatial description (Layers 1to 3) and temporal description within a scenario (Layer 4upwards) is consistent with the separation of entities inthe OpenDRIVE [20] and OpenSCENARIO [21] descriptionformats [38].
Guideline 2.
As explained in Section IV-C, Layer 3 con-tains temporary changes prevailing for the whole duration ofthe scenario. As such, objects of Layer 1 and Layer 2 thatare modified can be added as well as new objects of classesalready contained in Layer 1 and Layer 2.
Guideline 3.
In physics, a state describes the collection ofall information needed to describe a system at a certain pointin time. Then, a state change describes the change of onesystem state to another system state. If one considers entitieswithin an environment, states can continuously change orremain constant after a single change. The changed state ofan object from Layer 1 or Layer 2 is described in Layer 3if the following holds: A modification to the original state isvisible, but the change process itself could not be observed,and the modified state lasts for the duration of the scenario.As an example consider road markings covered by soil forthe entire scenario. In the 6LM, only those states whosechanges are explicitly observable over time are understood asbeing time-dependent and thus, described in Layer 4 upwardsaccording to Guideline 2. There can be a single state changeor continuous state changes as the road marking can becovered all at once or step by step. Of course, there are alsostate changes described in Layers 5 and 6, for instance, thechange of a weather condition or the state change of a trafficlight. VOLUME -, 2021 choltes et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment
FIGURE 6.
Overview of the layers including spatial and temporal separation.
Guideline 4.
Guideline 4 requires to locate entities fea-turing time-dependent changes on Layer 4 upwards. Thisincludes movable objects, as they can change their position.At the same time, however, these entities can also haveconstant properties, such as size and color, which can befurther detailed in an ontology on the basis of the 6LM (seeSection VII).
Guideline 5.
According to the definition of the 6LMand the previous guidelines, it is obvious that there can beproperties of a single entity assigned to different layers. Aprominent example are traffic lights. Traffic lights are bydefinition part of the traffic guidance objects and, therefore,located on Layer 1. In Layer 1, they are spatially situatedin a specific position. However, traffic lights can also changetheir states over time by switching, e.g., from green to yellow.This state change is located on Layer 6. This can easily beunderstood when thinking of the traffic light as two parts:The stationary element and the controller performing statechanges (this separation is also in accordance with previouspublications, e.g., [11]). A similar example are street lamps(positioned on Layer 2) that can be switched on and off (statechanges in Layer 5 as part of the lighting conditions).Always locating a specific property of an entity on thesame layer is definitely desirable. As such, movements of traffic participants are always mapped to Layer 4 and weatherconditions are exclusively addressed in Layer 5. However,there might be examples where the same property is de-scribed in different layers. If this is the case, it is due to andin accordance with the definition of the 6LM. Let us, e.g.,consider the visibility of a road marking. As stated in theexplanation of Guideline 4, it depends on the observabilityof the state change whether the property is located in Layer 3or Layer 4.We provided detailed and concise definitions for the 6LM.Nevertheless, we acknowledge that ambiguous situationsmay arise and would also like to give a guideline in caseof doubt. Anticipating the function of an entity or propertywithin a scenario and using it for the categorization cannotbe done as the 6LM is an unbiased description that does notcontain any interpretation. If it is not clear where to placean object or property, we will choose the layer where theobject or property has the largest influence. Influences onother layers are, however, still possible.The control and state change of a traffic light will belocated in Layer 6 as part of the ‘Digital Information’ Layer.They are not located in Layer 5 although one could arguethat the state change, e.g., from red to green, might influencethe lighting condition of the surroundings. That being said,
VOLUME -, 2021 et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment the influence of the state as traffic regulation element is ratedmuch higher and is, therefore, placed in Layer 6.
Guideline 6.
Annotations can be used for two reasons ifdesired. On the one hand, they can be used to add extra- interesting, but not necessarily essential - information re-garding entities. Prominent examples for this are annotationsto emergency responders. A police officer is described onLayer 4, as he is a pedestrian. However, we might want toannotate that he is on duty fulfilling some regulatory task.Similarly, the fact that emergency vehicles are executing theirprivileges through siren or blue lights can be annotated withthe entity on Layer 4.On the other hand, annotations can be used for reasonsof simplicity when an information is valuable, but does notneed to be given in detail. One example for this was alreadymentioned in the definitions of Layer 2 and Layer 5 whendescribing the motion of leaves. In general, an object forwhich we want to describe motion needs to be placed inLayer 4. This is, e.g., obvious for traffic participants. Think-ing of a bush, however, it might be desirable to only note thatit is moving in the wind, but it is not necessary to describethe movement of each leaf in detail on Layer 4. Therefore,the general information of an oscillating motion can also beannotated with the object on Layer 2. Analogously, warninglights at railroad crossings or on safety beacons can beflashing. If desired, this state change can be described onLayer 6. However, if the state does only undergo simpleperiodic changes (‘flashing’) this can also be marked as anannotation to the entity itself on the same layer the entity isoriginally placed.
Guideline 7.
A prominent example for an allegedly globalproperty is the friction coefficient. On first sight, it wouldbe possible to place the friction coefficient with the roador the environmental conditions. However, when consideringproperties that influence the friction coefficient, it becomesclear that it does not only depend on the road surface (asphalt,cobblestone, gravel, etc.) and the condition of the road (dry,wet, icy, etc.), but also on the material that is in contact withthe road. In case of the vehicle that would be the tire. At thispoint, it becomes clear that the description is not a globaland objective part of the general environment descriptionanymore. Therefore, such values are not part of the 6LM. It is,however, possible to determine such values by using severalproperties present in the description.Another striking example for the described actor-independence might be occlusions. The 6LM itself does notdescribe the occlusion for any traffic participant, but it givesthe possibility to perform an objective and complete environ-ment description that allows to calculate such occlusions forindividual instances of interest in a later step.
Guideline 8.
Each layer can influence previous layers andfollowing ones. This idea can also be found in [12]. Thepossibility of influence is independent from the numbering ofthe layers which are not ranked by importance, but purely forstructuring of the categorization. For instance, dense trafficcould cause that the shoulder is used for driving. This would be an influence of Layer 4 onto Layer 1 that is recorded inLayer 3. Other examples are the traffic light states that influ-ence the trajectories / maneuvers of traffic participants andthe weather conditions having an influence on the visibilityof road markings.With the guidelines presented in this section, the authorshope to facilitate the use of the 6LM and to encourage usersfrom research and industry to apply the model as a tool forstructured environment description.
VI. EVALUATION THROUGH REAL-WORLD DATA
The previous sections gave a definition of the different layersof the 6LM and provided guidelines for a clear classification.This section applies the definitions and guidelines to a real-world example in order to show the practicability of the 6LM.An environment description is given for a real-world mea-surement recorded by a drone taken from the IntersectionDrone (inD) - Dataset [33]. Fig. 7 shows four frames outof the video footage for which we analyze the physicaloccurrences. While each of the frames establishes a scene,a sequence of them is printed here showing the temporalevolvement, i.e. the scenario. The recording features an ur-ban, four-armed intersection. Buildings are blurred in therecordings in order to meet regulations on privacy of infor-mation.In the following, the visible elements in the scenes pre-sented in Fig. 7 are listed according to the 6LM. Please notethat all described elements of Layer 1 to Layer 4 featuredescriptions of their material. Those go hand in hand withthe object itself and are put into the same layer.
Layer 1 • Geometry and course of the road (including sidewalks) • Traffic signs (hardly visible in Fig. 7) • All markings: stop lines, crosswalk, markings for park-ing zones and keep-out areas • Road surface with irregularities
Layer 2 • Street lamps (e.g. next to the crosswalk) • Buildings and roadside structures, such as fountains(bottom middle), bicycle stands and vegetationThe presented recording does not contain any contentfor
Layer 3 as no temporary modifications of Layer 1 andLayer 2 elements are made. After considering Layers 1, 2and 3 the description of the spatial and non-temporal prop-erties is complete. The description made on Layer 1 andLayer 2 remains invariant when further real-world data of thisintersection is recorded. This is consistent with the modelingin OpenDRIVE [20] and an advantage of the clear separationbetween spatial and temporal properties that is established inthe paper.From Layer 4 upwards a time-based description is intro-duced. The temporal development of the events is revealedthrough the series of images in Fig. 7.
Layer 4 • All vehicles, moving and non-moving: The four vehicleson the street as well as the parked vehicles VOLUME -, 2021 choltes et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment
FIGURE 7.
Time sequence of snapshots from real-world data recorded at an intersection by utilizing a drone. • All bikes: driving on the street and parked at the bicyclestand • All pedestrians, moving and non-moving, e.g., the groupof four pedestrians using the crosswalkWithout enriching the recorded data with additional in-formation, obtaining a complete description on
Layer 5 isdifficult. Apparent in the recording is that no precipitationor harsh weather condition such as low visibility is present.Therefore, as part of the road weather the street can be de-scribed as dry. Furthermore, shadows are visible for differentobjects. Those shadows are part of the lighting conditions onLayer 5.This intersection does not feature any traffic lights orswitchable traffic signs, therefore, there is no requirement todepict their status on
Layer 6 . In the same way any other typeof Layer 6 information is also difficult to recover from thereal-world data recording. Cellular network coverage couldbe gathered through data enrichment or actual measurementsalong with the recording of the video footage. Additionally,the recorded intersection is not equipped with any V2Xinfrastructure and, to the best of our knowledge, there are notraffic participants featuring this technology. Therefore, thereis no digital information present
VII. FUTURE WORK
In order to achieve a standardized description, the authorsplan to implement the 6LM as part of a domain ontology forthe verification and validation of highly automated vehicles,including a taxonomy of relevant traffic entities as well as their properties and relations. According to the given classi-fication of the 6LM, this ontology will detail the differententities, properties and relations in a comprehensive way,e.g., by introducing subclasses for traffic participants andadding specific weather conditions.Further, it can be helpful to distinguish between envi-ronment description for driving functions and machine per-ception. Factors triggering actions within a scenario, whichmight be worth investigating, can be very different for thetwo groups. Furthermore, when considering perception re-lated aspects the current 6LM might need some adaptationsor add-ons as the description of perception aspects cannotnecessarily be made actor-independently. Future work willthus concentrate on possible environment description for ma-chine perception. For instance when looking at driving func-tions, it is often sufficient to look at road surface material.When looking at perception aspects material descriptionsfor several other entities increase in importance. Moreover,occurrences of reflection and contamination [39] become ofinterest. Therefore, it is then necessary to give much moredetailed descriptions of all surroundings, not only of the roadsurface. This could possibly introduce new properties intothe 6LM or require a more detailed description of alreadymentioned properties. This issue could also be addressed inthe aforementioned domain ontology.Additional future work will deal with possible data sourcesto enrich recorded test data.
VOLUME -, 2021 et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment
VIII. SUMMARY
The categorization and guidelines given in this work refinethe 6LM for environment description that was originally es-tablished in PEGASUS. Furthermore, the model is extendedto serve a variety of new applications from verification andvalidation and to address more complex scenarios than justhighway scenarios. Applying the 6LM to urban environmentsrequired integrating a concept for roadside structures, othertypes of dynamic objects and traffic light states.The definitions and guidelines given help to gain a stan-dardized categorization for a generally usable, unbiased andobjective environment description. The work covers all layersand gives explanatory examples along with guidelines. Theclear structure established in the 6LM can be used as a basisfor scenario descriptions and ontologies.
ACKNOWLEDGMENT
The research leading to these results is funded by the GermanFederal Ministry for Economic Affairs and Energy withinthe project ‘VVM - Verification & Validation Methods forAutomated Vehicles Level 4 and 5’.
REFERENCES [1] ERTRAC Working Group, “Connectivity and automated driving”,
Con-nected Automated Driving Roadmap , Burssels, Belgium, 2019, ver. 8.[2] W. Wachenfeldt, and H. Winner, “The release of autonomous driving”,in
Autonomous Driving , Berlin, Heidelberg, Germany: Springer-VerlagGmbH Berlin Heidelberg, 2016, pp. 425-449.[3] T. Menzel, G. Bagschik, and M. Maurer, “Scenarios for development,test and validation of automated vehicles”,
IET ITS ,2014, vol. 8, Iss. 3, pp.183-189[9] S. Ulbrich, T. Menzel, A. Reschka, F. Schuldt, and M. Maurer, “Defin-ing and Substantiating the Terms Scene, Situation, and Scenario forAutomated Driving”,
IEEE 18th International Conference on IntelligentTransportation Systems , Las Vegas, 2015, pp. 982-988.[10] DIN SAE SPEC 91381:2019-06, “Terms and definitions related to testingof automated vehicle technologies”, 2019.[11] F. Schuldt, “Ein Beitrag für methodischen Test von automatisiertenFahrfunktionen”, Ph.D. dissertation, Fakultät für Elektrotechnik, Informa-tionstechnik, Physik, TU Braunschweig, Braunschweig, 2017.[12] G. Bagschik, T. Menzel, and M. Maurer, “Ontology based scene creationfor the development of automated vehicles”, , 2018, pp.1813-1820.[13] J. Bock, R. Krajewski, L. Eckstein, J. Klimke, J. Sauerbier, A. Zlocki,“Data basis for scenario-based validation of HAD on highways”,
AachenColloquium Automobile and Engine Technology , Cangshu, 2018. [14] G. Bagschik, T. Menzel, C. Körner, and M. Maurer, “WissensbasierteSzenariengenerierung für Betriebsszenarien auf deutschen Autobahnen”,
Uni-DAS e.V. Workshop Fahrerassistenzsysteme und automatisiertesFahren , 2018, Bd. 12.[15] H. Weber, J. Bock, J. Klimke, C. Rösener, J. Hiller, R. Krajewski, A.Zlocki, and L. Eckstein, “A framework for definition of logical scenariosfor safety assurance of automated driving”,
Traffic Injury Prevention
IEEE Intelligent Systems
Model-Based Safety and Assessment. IMBSA 2020. LectureNotes in Computer Science VOLUME -, 2021 choltes et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment , Paris, France, 2019, pp. 2383-2390.[39] R. Felton, “Google has a wiper for its LiDAR to handle bird shit andrain and look at it go”, 2017. [Online]. Available: https://jalopnik.com/google-has-a-wiper-for-its-lidar-to-handle-bird-shit-an-1795346982.Accessed on: March 03, 2020.
MAIKE SCHOLTES received her B.Sc. andM.Sc. (Dean’s List Award) degrees in Computa-tional Engineering Science from RWTH AachenUniversity, Germany in 2015 and 2017, respec-tively. During semesters abroad she, inter alia,studied at the University of California, San Diego.After working one year in the area of drivingsimulation at fka SV Inc. in the Silicon Valley, shestarted her Ph.D. at the Institute for AutomotiveEngineering (ika) at RWTH Aachen University.She works on the assessment of automated driving and ADAS with a focuson machine perception.
LUKAS WESTHOFEN received the degrees ofB.Sc. and M.Sc. (with honors) in computer scienceby the RWTH Aachen University in 2015 and2019, specializing on the topics of probabilisticprograms and software verification. Since 2019,he is a PhD student at OFFIS e.V., Oldenburg,Germany. His general work focuses on developingmethods to establish confidence in the safety ofautomated vehicles. More specifically, his researchinterests include the formalization of knowledgeas well as its exploitation for safeguarding automated driving functions.
LARA RUTH TURNER received both herdiploma degree and Ph.D. in mathematics fromthe University of Kaiserslautern in 2008 and 2012,respectively. After a postdoctorate at the Univer-sity of Vienna, she joined ZF Friedrichshafen AGin 2014. She currently works on validation forassisted and autonomous driving functions with aspecial focus on scenario-based methods.
KATRIN LOTTO graduated with diploma inmathematics at Technische Universität München,Germany. Since 2009 she has been working atZF Group in Friedrichshafen, Germany, in variousareas of transmission development to operate nowin the field of validation methods for automatedvehicles.
MICHAEL SCHULDES received the B.S. andM.S. degree in electrical engineering from RWTHAachen University, Germany, in 2016 and 2019respectively. He is currently a Research Assistantwith the Institute for Automotive Engineering,RWTH Aachen. His research interest lies in thearea of assessment of autonomous driving func-tions with a focus on data-driven scenario-basedapproaches.
HENDRIK WEBER received his B.Sc. in Me-chanical Engineering in 2015 and his M.Sc. in Au-tomation Engineering in 2017, both from RWTHAachen. Since 2017, he has been a research as-sistant at the Institute for Automotive Engineering(ika) at RWTH Aachen University, where he isalso pursuing his PhD. His work focusses on ver-ification and validation approaches for automatedvehicles as well as on methods and tools for as-sessing their impact on safe traffic. Currently, he isleading the evaluation subproject in the European research project L3Pilot,which pilots L3 vehicles on public roads and evaluates their behavior intraffic and resulting impacts.
NICOLAS WAGENER received the B.Sc. andM.Sc. degrees in Computer Science from RWTHAachen University, Germany, in 2014 and 2016,respectively. Since the end of 2016, he is a re-search scientist and PhD student at the Insti-tute of Automotive Engineering (ika) at RWTHAachen University. Currently, he is Group Leaderfor Simulation within the research area VehicleIntelligence & Automated Driving at the Institute.Moreover, he is the scientific coordinator of thesub-project ‘Databases’ within the project VVMethods. His research interestinclude simulations and driving simulators, safety validation methods andautomated driving.
VOLUME -, 2021 et al. : 6-Layer Model for a Structured Description and Categorization of Urban Traffic and Environment
CHRISTIAN NEUROHR received the B.Sc. andM.Sc. degrees in mathematics in 2011 and 2013respectively, both from Technische UniversitätKaiserslautern, Germany and the Ph.D. degree(Dr. rer. nat.) from Carl von Ossietzky UniversitätOldenburg, Germany in 2018. After a short periodas a visiting researcher with the MAGMA groupat the University of Sydney, he started his currentoccupation as a postdoctoral researcher at OFFISe.V., Oldenburg, Germany, where he is working inthe area of scenario-based verification and validation of automated vehicles.Specifically, he is the scientific coordinator of the sub-project ’CriticalityAnalysis’ within the project VVMethods.
FRANZISKA KÖRTKE graduated with diplomain mechanical engineering at the Technical Uni-versity of Dresden, Germany. Since 2017, she hasbeen working in the field of test automation forscenario-based validation of automated vehicles atZF Group in Friedrichshafen, Germany.
MARTIN HERBERT BOLLMANN received theDipl.-Ing. in mechanical engineering from Tech-nical University Dresden in 2012. Until 2019,he worked as a calculation and reliability engi-neer at ZF Group in Friedrichshafen, Germany.Currently, he works as test architect, developingtest strategies and searching for test methods, thatare aligned with SOTIF and able to fulfill re-quirements coming from safety-related standards.Moreover, he is the industrial coordinator of thesub-project ’Criticality Analysis’ within the project VVMethods.
JOHANNES HILLER studied Electrical Engi-neering, Information Technology and ComputerEngineering at RWTH Aachen University. Cur-rently, he is Group Leader Data & IntelligentInfrastructure within the research area Vehicle In-telligence & Automated Driving at the Institute forAutomotive Engineering (ika) at RWTH AachenUniversity. He works on the assessment and eval-uation of automated driving and advanced driverassistance systems with a focus on data analysis,enrichment and the analysis of video.
MICHAEL HOSS received the B.Sc. and M.Sc.degrees in Computational Engineering Sciencefrom RWTH Aachen University, Germany, in2015 and 2017, respectively. Since the beginningof 2018, he has been a research associate at theInstitute for Automotive Engineering (ika) of thesame university, where he is also pursuing a Ph.D.degree. His research interest focuses on safety-aware test methods for the perception subsystem.
JULIAN BOCK received the B.Sc. degree inComputational Engineering Science from RWTHAachen University in 2012 and the M.Sc. degreein Computational Engineering Science in 2014.From 2014 to 2019, he was research scientistand PhD student at the Institute for AutomotiveEngineering at RWTH Aachen University. Since2020, he is Manager Artificial Intelligence at fkaGmbH in Aachen. His research interests includeautomated driving and its fields safety validation,real-world scenario data acquisition, machine learning and pedestrian pre-diction.
LUTZ ECKSTEIN is head of the Institute forAutomotive Engineering (ika) at RWTH AachenUniversity. He studied mechanical engineering atStuttgart University and received his Ph.D. fromthe same university in the year 2000. Duringhis time in the industry he worked on activesafety systems and human-machine interactionsfor Daimler AG and BMW AG. He was appointedto his professorship and current position at RWTHAachen University in 2010.16