• 検索結果がありません。

Use-case Generation and Analysis for Autonomous Driving in Urban Areas

N/A
N/A
Protected

Academic year: 2021

シェア "Use-case Generation and Analysis for Autonomous Driving in Urban Areas"

Copied!
8
0
0

読み込み中.... (全文を見る)

全文

(1)Takuya Nanri et al / International Journal of Automotive Engineering Vol.12, No.2(2021) 20214255. Technical Paper. Use-case Generation and Analysis for Autonomous Driving in Urban Areas Takuya Nanri1). Fang Fang1). Abdelaziz Khiat1). 1). Mobility &AI Laboratory, Nissan Motor Co., Ltd. 1-1 Morinosato-aoyama, Atsugi, Kanagawa 243-0123 JAPAN E-mail: {t-nanri, fangfang, khiat}@mail.nissan.co.jp Received on Nov. 9, 2020 ABSTRACT: Being able to generate adequate test scenarios in order to validate autonomous driving functions is of paramount importance for the deployment of autonomous vehicles. However, asserting the completeness of generated test cases and coverage of all possibilities has proven to be very difficult. Previously, few attempts that tried to generate representative possibilities have all been based on the designer′ s experience and thus inherently incomplete. This paper proposes a formal system inspired approach that solves this issue. With this approach, 2,544 basic use-cases were generated from a completeness perspective. Thereafter, for the sake of accelerating the development work of our autonomous vehicle, we used this asset to clarify which use-cases should be solved and in what order. KEYWORDS: Electronics and control, Autonomous driving, Reliability (E1) 1.. Introduction. Background: Autonomous vehicles in urban areas are an important factor in a Mobility-as-a-Service (MaaS) platform such as self-driving taxis[1] and unmanned ground vehicles for logistics[2]. An example of self-driving taxis is Easy Ride[1]; which is a field trial being conducted in Yokohama for a ride-hailing service using autonomous vehicles. Unmanned ground vehicles are being implemented in logistics field trials all over the world, including in Europe, the U.S., China and Japan[2]. A challenge for the introduction of autonomous driving in urban areas is to assure that these vehicle systems beFigure 1: Application area in Yokohama. have safely. Being able to generate adequate test scenarios in order to validate autonomous driving functions is of paramount importance for the deployment of autonomous. plete” use-case generation in urban areas, although an as-. vehicles. However, asserting the completeness of generated. sessment framework called a ”scenario-based approach” has. test cases and coverage of all possibilities has proven to be. been proposed[3] and some highway driving scenarios have. very difficult. Moreover, from the viewpoint of development. been generated[4].. work management, we cannot assess what the requirements. Purpose: In this paper, we propose a formal system. are and what should be done to meet them unless test sce-. inspired approach for generating use-cases for autonomous. narios are first clarified. Therefore, it is hard to judge the. vehicles in urban areas. Generating use-cases enables us to. progress status of the development work and to decide pri-. quantify the coverage of use-cases in urban areas for the de-. orities.. velopment work management and to clarify what use-cases. However, different from highway autonomous driving,. should be considered. Moreover, in order to apply it to. autonomous vehicles in urban areas face all kinds of use-. Yokohama area for Easy Ride[1], we defined an operational. cases due to many different road geometries and surround-. design domain (ODD) with the same ontology dictionary. ing objects. Previously, few attempts that tried to gener-. used in our use-case description and narrowed down on the. ate representative possibilities have all been based on the. use-cases that correspond to the targeted area. We describe. designer’s experience and thus incomplete. To the best. a framework that has been constructed to quantify the cov-. of our knowledge, no one has attempted to tackle ”com-. erage of use-cases for the development work management.. Copyright © 2021 Society of Automotive Engineers of Japan, Inc. All rights reserved. 54.

(2) Takuya Nanri et al / International Journal of Automotive Engineering Vol.12, No.2(2021) Finally, we show that our use-cases described with our on-. proach” and have generated use-cases for highway driving. tology have the potential to be applied to logical reasoning. based on the framework[3][4].. about encountered traffic scenes.. The PEGASUS project for standards of autonomous. Contributions: Our contributions are summarized in. driving, in which many OEMs (Original Equipment Manu-. the following four points: Firstly, the proposed approach. facturers) and Tier 1 suppliers in Europe participated, pro-. generates use-cases for autonomous driving in urban ar-. posed an assessment framework called a ”scenario-based. eas considering completeness. This enables us to calculate. approach” for autonomous vehicles and generated a lim-. the coverage from a use-case perspective and to generate. ited number of highway driving scenarios to validate the. complete test scenarios by adding necessary detailed pa-. framework[3]. The work done was largely theoretical, and. rameters. The second contribution is that we have defined. generated a limited number of use-cases for driving on Ger-. an ontology to describe the generated use-cases. The ontol-. man highways. The Japan Automobile Manufacturers As-. ogy dictionary reduces the ambiguity of use-case meanings. sociation (JAMA) has also generated highway driving sce-. in natural language, making it easy to search for relevant. narios using a similar ”scenario-based approach” and has. use-cases using human-understandable keywords. Besides,. derived 32 critical situation scenarios[4] for highways; and. our ontology-based description of use-cases has the poten-. has also generated a limited number of use-cases for driving. tial to be applied to logical reasoning about traffic scenes,. on Japanese highways. Meanwhile, the National Highway. especially in complicated situations. The third one is that. Traffic Safety Administration (NHTSA) in the U.S. has. we have defined a general ODD description based on our. reservations about designing guidelines for use-case gen-. ontology dictionary in order to narrow down generated use-. eration of autonomous vehicles. It requested the Society. cases for corresponding targeted situations. Since the ODD. of Automotive Engineers (SAE) to summarize best prac-. is described using the same ontology, the use-cases corre-. tices and would-be de facto standards, although it has pro-. sponding to the ODD can automatically be extracted and. posed several important concepts for representing scenar-. can easily be linked to situations that need to be dealt. ios, such as ODD and object and event detection and re-. with. The last contribution is the construction of a frame-. sponse (OEDR)[5].. work that enables us to calculate and clarify the status of. Ontology representation: Ontology here refers to. development work by adding more information to the gen-. the representation of knowledge-based systems; which are. erated use-cases such as the relevant necessary techniques. widely applied in semantic web applications.. to solve them. Since it can clarify what use-cases should be. an important concept for consistent use-case representa-. solved and what techniques should be pursued, it leads to. tion because natural language description is too ambigu-. better visibility and acceleration of the development works.. ous to represent use-cases of autonomous driving. There. This paper is organized as follows: related studies are. are some studies that have defined ontologies for describ-. summarized in section 2., after which our approach and. ing use-cases of autonomous vehicles with consistency. For. results for use-case generation are presented, followed by. example, Bagschik et al.[6] defined an ontology in the field. discussions and conclusions.. of automated vehicles to help experts completely create a. It is also. wide range of scenarios, and proposed a traffic scene cre2.. Related studies. ation approach based on their ontology. They modeled the ontology with 284 classes, 762 logical axioms and 75. In this section, some related studies are first described Then, some. semantic web rules to generate scenes for German motor-. ontology-related studies in the field of autonomous vehicles. ways. They proposed a process for generating traffic scenes. are mentioned since ontology is a key component for consis-. by defining an ontology and generated a limited number. tent use-case representation. Moreover, some ODD-related. of scenes only for German motorways for the purpose of. studies are presented. Finally, we clarify the differences. validating the process and their ontology. Then, they dis-. between these related studies and our approach.. cussed the requirements for the representation of scenarios. from a use-case generation perspective.. Use-case generation: Autonomous driving (AD) use-. in the different process steps defined in ISO 26262, and. cases has been generated by some groups and projects for. proposed a consistent terminology to create scenarios[7].. standardization purposes all over the world. However, to. However, they did not consider urban areas. Besides, there. the best of our knowledge there are few activities for gen-. are other papers[8][9][10][11] that used ontologies to rea-. erating use-cases completely from an urban area perspec-. son about some driving situations or road structures in au-. tive, although some groups have established and utilized an. tonomous driving application in order to handle specific,. assessment framework referred to as a ”scenario-based ap-. complicated, although limited situations, such as a straight. Copyright © 2021 Society of Automotive Engineers of Japan, Inc. All rights reserved. 55.

(3) Takuya Nanri et al / International Journal of Automotive Engineering Vol.12, No.2(2021) road with crossing, an intersection with traffic light and T-. and scenario, proposed by Ulbrich et al[16]. They defined. junction. In other words, they created ontologies and rules. these terms for a consistent representation of traffic scenar-. for a limited application scenario and they did not cover all. ios. Then, we define the ontology for autonomous vehicles. possible use-cases in urban areas.. to achieve a consistent description of use-cases and confirm. ODD definition: SAE J3016 provides the following def-. their completeness. Moreover, we define an ODD with the. inition of ODD: ”Operating conditions under which a given. same ontology as the one used to generate use-cases. By. driving automation system or feature thereof is specifically. using the same ontology, we can easily link use-cases with. designed to function, including, but not limited to, environ-. ODD and extract use-cases corresponding to an ODD.. mental, geographical, and time-of-day restrictions, and/or. In this section, we first describe the hierarchical classi-. the requisite presence or absence of certain traffic or road-. fication and the granularity of use-cases to generate them.. way characteristics[12].” The use-cases to be dealt with de-. After that, we describe our ontology dictionary used in con-. pend on an ODD. Several companies that have been do-. firming the completeness of generated use-cases. Finally,. ing a field test of autonomous vehicles have defined their. we present an ontology-based ODD definition used to nar-. respective ODD[13][14][15]. For example, an OEM in Eu-. row down generated use-cases for a specific application.. rope has used the following categories as an ODD for au-. 3.1.. tonomous vehicles: speed range, time-of-day, weather con-. Hierarchical classification. In order to assert completeness of use-cases, we define. ditions, the presence or absence of certain types of roadway. a MECE hierarchical classification axis.. infrastructure and the geographical area in which the au-. The hierarchi-. cal classification axis consists of upper and lower levels.. tomated vehicle is designed to operate[13]. Also, a white. The upper level includes basic vehicle behavior (maneuvers). paper by several OEMs and Tier 1 suppliers shows the fol-. from a lane perspective and the ego vehicle’s pre-planned. lowing ODD categories for autonomous vehicles: weather. navigation path. Although geographical road structures. conditions, geographic domain, background scene and dy-. can be used as a classification axis such as intersections,. namic properties of the scene[14]. These ODDs are used as. straight roads and so on, it is hard to classify use-cases. a requirements for autonomous vehicles, but there seems. from a MECE point of view because there are too many. to be no direct link between those requirements and corre-. perspectives to define in one axis. On the other hand, the. sponding use-cases.. ego vehicle’s behavior can be treated as a MECE classifi-. Our approach: Our purpose is to generate complete. cation axis because the subject of behavior is clear.. use-cases for autonomous vehicles in urban areas in the. The upper level includes keeping lane, changing lane,. real world. We have generated use-cases in urban areas and. crossing lane (intersection) and so on. Crossing lane (in-. confirmed its completeness by describing use-cases with our. tersection) means the ego vehicle crosses another lane such. defined ontology. Moreover, this approach makes it possi-. as at an intersection. Crossing lane have the other prop-. ble to visualize the development work status based on the. erty in the upper level, ego vehicle’s pre-planned navigation. generated use-cases by adding information on the progress. path. For example, crossing lane has three possible paths:. level of each related technical aspect. Afterwards, we in-. crossing, which means going straight, driver side turn and. troduce an ontology-based ODD definition that enables us. passenger side turn. Since we have left-hand traffic and. to easily narrow down generated use-cases corresponding. right-hand traffic in the real world, we refer to a right turn. to a specific ODD and to clarify what use-cases need to be. in left-hand traffic and a left turn in right-hand traffic as. solved.. a driver side turn, and a left turn in left-hand traffic and 3.. right turn in right-hand traffic as a passenger side turn.. Proposed method. Fig.2 shows a part of the upper level of the hierarchical Completeness is the ability to cover all possible use-cases. classification, including basic behavior and the ego vehi-. that any autonomous vehicle could encounter. To ensure. cle’s pre-planned navigation path. Fig.3 shows an example. completeness of use-cases, we define a hierarchical classifi-. of upper-level classification.. cation axis from a Mutually Exclusive, Collectively Exhaus-. Next, we define the lower level, which includes factors. tive (MECE) point of view. We also define the granularity. that induce change in the ego vehicle’s basic behavior at. of use-cases in order to clarify how deeply they should be. upper level.. classified. We can alleviate combinatorial explosion and. three types of attributes: environment, road and object. obtain tractable use-cases by defining the granularity. In. attributes. The first level of environment attributes, for. order to define the granularity of use-cases, we use the fol-. example, is classified into visibility and weather. The first. lowing concepts to represent traffic scene: scene, situation,. level of road attributes is classified into the road surface. The factors at the lower level consist of. Copyright © 2021 Society of Automotive Engineers of Japan, Inc. All rights reserved. 56.

(4) Takuya Nanri et al / International Journal of Automotive Engineering Vol.12, No.2(2021) Upper level. Lower level Environment attributes. Keeping lane. Road attributes Object attributes. Changing lane. Figure 4: Lower level of use-cases Crossing. Crossing lane (Intersection). behavior typical of an average human driver. The last pre-. Driver side turn. condition is that the ego vehicle always has a defined goal, i.e., it has a pre-planned navigation path. This path is not. Passenger side turn. a lane-level path, but a road-level navigation path. The granularity of use-cases is defined as follows. The use-case definition used in this paper adopts the relevant. Figure 2: Upper level of use-cases. concepts for representing the traffic environment proposed by Ulbrich et al., namely, scene, situation and scenario[16]. Before we define the use-cases, these concepts are first defined as follows. Scene: Based on Ulbrich et al’s definition[16], a scene is defined as a snapshot, including the ego vehicle, driving environment, surrounding objects, and the qualitative relationships among them. Fig.5 schematically shows the Keeping lane. Changing lane. Crossing … lane (Intersection). concept of a scene, including the ego vehicle, objects, environment, and the qualitative relationships between the ego vehicle and objects, between the ego vehicle and the. Figure 3: Examples of upper level of use-cases. environment, between objects and the environment, and finally among objects. The scene resolution is defined ac-. and road furniture. The first level of object attributes is. cording to whether the information can be distinguished by. classified into dynamic and static objects. All the factors. a snapshot or not.. that induces changes in basic behavior at the upper level. Examples of each component are as follows:. are listed in the lower level. Fig.4 shows three types of • Ego vehicle: an autonomous vehicle. attributes at the lower level of use-cases.. • Driving environment: this includes the number of 3.2.. Granularity of use-case. lanes, presence of traffic lights, traffic signs, etc.. The finer the classification axis becomes, the larger is the. • Surrounding objects: they consist of dynamic objects. number of use-cases. The preconditions and the granular-. (vehicles, pedestrians), and static objects(parked ve-. ity of use-cases are defined in order to alleviate combina-. hicles, falling objects, construction-related materials). torial explosion of the number of use-cases and to achieve. • Qualitative relationships between the ego vehi-. tractable use-cases.. cle/objects and driving environment: this includes a. The preconditions are defined as follows:. self-driving lane, distance to stop lines (i.e. far, close),. • The ego vehicle is autonomous.. logical drivable area (lateral lane width), etc. • Qualitative relationships among objects: this includes. – The ego vehicle complies with traffic rules.. their relative orientation (oncoming, crossing), etc.. – The ego vehicle displays behavior typical of an average driver.. For example, a scene without pedestrian and another scene with pedestrian are different because the objects differ in. – The ego vehicle has a pre-planned navigation. terms of the snapshot point of view.. path.. Situation: A situation is derived from a scene, including. Needless to say, the main precondition is that ego vehicle is. the information, for example, path and vehicle type, needed. autonomous. Of the other three preconditions, the first one. for the ego vehicle to make an appropriate decision. The. is that the autonomous vehicle fully complies with traffic. following are examples of the components of a situation:. rules. This is the most important of the other three precon-. • Path: this includes keeping the same lane, changing. ditions. It is assumed that the autonomous vehicle displays. Copyright © 2021 Society of Automotive Engineers of Japan, Inc. All rights reserved. 57.

(5) Takuya Nanri et al / International Journal of Automotive Engineering Vol.12, No.2(2021) Use-case 1. Scene Ego vehicle. Scenario 1. Situation 1. Objects. Scenario 2 Scene Scenario 3 :. Environment. Use-case 2 Situation 2. Scenario 4 Scenario 5. Figure 5: The concept of scene. Scenario 6. :. :. Figure 7: Concept of use-case. Scenario 1. Scene. Situation 3. Situation 1 Scenario 2. resenting use-cases in natural language can lead to ambigu-. Scenario 3. Situation 2. ity. Therefore, an ontology is defined that describes all the. Scenario 4. use-cases. Ontology means a representation of knowledgebased systems, which are widely applied in semantic web. Figure 6: Concept of scenario. applications. Ontology is also an important concept for consistent use-case representation. The ontology defined. lanes, crossing lanes (going straight, turning toward. here consists of six categories: class, entity state, relational. driver side or passenger side), etc.. element, future behavior element, decision making element. • Vehicle type: this includes a passenger vehicle, truck,. and parameter.. bus, taxi, two-wheeled vehicle (motorcycle), an emer-. The class category has environment, object and road at-. gency vehicle.. tributes. These are the same as the lower levels for classify-. Moreover, the objects that do not affect the behavior of the. ing use-cases. The entity state category is divided into ob-. ego vehicle are removed from the situation. For example, if. ject states and road states. The relational element category. a pedestrian does not affect the behavior of the ego vehicle,. shows the relations between objects and between objects. a scene without the pedestrian and another scene with the. and road furniture. The future behavior element shows the. pedestrian are the same from a situation perspective.. behavior of other object entities and road furniture in the. Scenario:. A temporal development between several. future. For example, this category is expressed using the. scenes in a sequence is defined as a scenario. A single situa-. verb ”will”. The decision making element category shows. tion may contain several scenarios. Fig.6 shows the concept. the behavior of the ego vehicle in the future. The behavior. of a situation and a scenario. Each node represents scene,. of the ego vehicle can be decided by itself. The parameter. and a temporal development between several scenes in a. category shows the parameters included in each category.. sequence constitutes a scenario. One situation can include. From scene, situation and scenario definition perspective,. several scenarios.. the first three categories (class, entity state and relational. Use-case: A use-case is considered here at the situation. element) include scene-related ontologies.. For example,. level. Situations in which the ego vehicle has the same be-. class includes ontologies about the definition of surround-. havior policy are defined as a single use-case. One use-case. ing objects themselves, and entity state includes ontologies. may contain several scenarios. Fig.7 shows the granular-. about the definition of surrounding objects’ states. Also,. ity of use-case. There are levels of granularity for scene,. relational element includes ontologies about the definition. situation and scenario. A scenario may have many varia-. of qualitative relationships among the ego vehicle, driving. tions corresponding to one situation, because it could eas-. environment and surrounding objects. The next two cat-. ily change depending on the behavior of the ego vehicle. egories (future behavior element and decision making el-. or other objects, such as their acceleration or deceleration,. ement) include situation-related ontologies. For example,. especially in negotiation situations. This means that the. future behavior element includes ontologies about the def-. number of scenarios can easily become much larger. In. inition of surrounding objects’ path, and decision making. contrast, use-cases at the situation level are easier to treat.. element includes ontologies about the definition of the ego. Therefore, a use-case is defined here at the situation level.. vehicle path. Since the use-case is considered here at the situation level, categories do not include scenario-related. 3.3.. ontologies.. Ontology representation. Our ontology is based on previous ontologies[8][17] that. Although we can generate use-cases using the hierarchi-. have been applied for understanding scenes involving au-. cal classification axis and the granularity of use-cases, rep-. Copyright © 2021 Society of Automotive Engineers of Japan, Inc. All rights reserved. 58.

(6) Takuya Nanri et al / International Journal of Automotive Engineering Vol.12, No.2(2021) Road attributes Environment attributes. Road geometry…. Table 1: The number of use-cases Upper level # of use-cases. Time of day…. Keeping lane. 300. Changing lane Operational design domain. Ego vehicle attributes Object attributes. Navigation path…. 69. Crossing lane. Crossing. 478. (Intersection). Driver side turn. 847. Passenger side turn. 572. Static object…. : Local rule attributes. :. Total. Local rule…. 2,544 4.. Figure 8: Our definition of ODD. Application result. In this section, we show the result of use-case generation tonomous vehicles. Our ontology is extended from pre-. and confirmation of use-case ”completeness”. Then, we de-. vious ones to cover all possible use-cases for autonomous. fine an ODD in order to drive autonomously for Easy Ride. driving, as the latter just cover a limited number of driv-. in Yokohama[1] and narrow down use-cases to correspond. ing situations. Since previous ontologies do not cover all. to an ODD in Yokohama.. use-cases, they can be subsumed under our ontology. Our. 4.1.. ontology categories have been adapted from previous ones in order to cover all possible use-cases. An ontology con-. cases for autonomous driving. In defining this order of. sists of terminological boxes and assertional boxes[8], and. priority, we started with the usual situations, that is, nat-. all the categories here are assigned to the terminological. ural environment conditions of sunny or cloudy daytime. boxes. Sensor data are used as assertional boxes for our. weather. While other vehicles usually comply with traf-. ontology. 3.4.. Use-case generation. An order of priority is defined in order to generate use-. fic rules, we consider actual frequent deviations from traf-. Ontology-based ODD definition. fic rules, e.g., over-speeding, ignoring stop lines/signs and. Our ODD is defined with five categories: road, environ-. changing lanes in prohibited areas. At this point, we do. ment, ego vehicle, object and local rule attributes. These. not consider parking-related use-cases.. categories are a general definition because they are based on. Based on the defined classification, preconditions, order. the ODD definition[12] and contain all the items in the pre-. of priority, and granularity, 2,544 use-cases were generated.. vious usages[13][14][15]. Namely, road attributes include. Table.1 shows a part of the number generated at each upper. road geometry, environment attributes include time of day. level.. and weather conditions, ego vehicle attributes include nav-. An example of a use-case is shown in Fig.9. The use-. igation path, object attributes include static object and. case here is described based on first-order logic format and. dynamic object and local rule attributes include local traf-. is classified into three parts: ego vehicle’s decision making. fic rules and regulations as shown in Fig.8. Since these. based on pre-planned navigation path, future behavior of. attributes are described with our ontology dictionary, we. the other objects and ego vehicle’s decision making consid-. can easily extract the related use-cases with an ODD.. ering future behavior of the other objects. The first part is. ODD here is the set of conditions which define the exis-. decision making behaviors related to pre-planned naviga-. tence or not of an item in the area of interest, for example. tion path. The second part is future behaviors of the other. intersection without traffic light, roundabout, pedestrians. objects that can be predicted by ego vehicle. The last part. and U-turn are items that could exist or not in a certain. is decision making behaviors considering the future behav-. area. These items are extracted from our definition of on-. iors of the other objects. Using this format, we can ap-. tologies; where each item in the ODD corresponds to a. ply several reasoning methods such as first-order logic and. specific definition in the ontology (e.g., road attributes of. Markov logic. Since it covers all possible use-cases in ur-. ODD include the set of ontologies of road subcategory in. ban areas, it could reason about all the traffic situations in. class category of ontology). On the other hand, a use-case. urban areas.. represented with an ontology which does not exist in an. 4.2.. ODD, is out of scope from the ODD’s perspective.. Confirmation of use-case completeness. The completeness of the use-cases generated for autonomous driving was also confirmed. AD use-cases can be. Copyright © 2021 Society of Automotive Engineers of Japan, Inc. All rights reserved. 59.

(7) Takuya Nanri et al / International Journal of Automotive Engineering Vol.12, No.2(2021) Decision making Oncoming based on pre-planned navigation path vehicle TurnToDriverSide(EgoVehicle) => ShiftPositionInLaneToDriverSide(EgoVehicle). - Pre-planned nav. path: Driver side turn - Traffic light state: Green - Object: Oncoming vehicle. 727. 3. 10. Future behavior of object isPassengerVehicle(V1) & isRespectingLegalSpeed(V1) & PositionInLaneOnCenter(V1) & isOnIntersectionWithTrafficLight(V1) & isTrafficLight_Green(EgoVehicle) => willGoStraight(V1) Decision making based on future behavior of object willGoStraight(V1) & TurnToDriverSide(EgoVehicle) isBlocking (EgoVehicle, V1). 4,784,428 6. 10. 2,544. TurnToDriverSide(EgoVehicle) => Decelerate(EgoVehicle). Ego vehicle. 43,059,852. 1 Combination number. Considering just usual situations. Unifying cases & eliminating improvable cases. ODD constraint for Easy Ride in Yokohama. Figure 10: Confirmation of completeness =>. Table 3: The example of ODD Yokohama. willGoStraight(V1) & TurnToDriverSide(EgoVehicle) => hasRightOfWayOver(V1, EgoVehicle) isBlocking(EgoVehicle, V1) & isApproachingTo(EgoVehicle, V1) & hasRightOfWayOver(V1, EgoVehicle) => StopFor(EgoVehicle,V1). Intersection w/ traffic light. red. Roundabout. x. Pedestrian. o. Bus priority lane. x. :. :. Figure 9: Example of use-case hicle that will turn to driver side are different situations. However, they are unified into a single use-case from use-. Table 2: The definition of variables Symbol Meaning E. Set of environment attributes. O. Set of object entities. R. Set of road entities. Os. Set of object entity states. Rs. Set of road entity states. Rel. Set of relational elements. F. Set of future behavior elements. Si. Set of situations. U. Set of use-cases. case definition perspective if the ego vehicle goes straight because these behaviors of oncoming vehicle do not affect the ego vehicle behavior policy. Equation.2 represents this unifying process.. We then eliminated improbable cases. among them, for example straight road x turn to driver side. Fig.10 shows the number of cases in each process. The number of use-cases was confirmed as shown in this figure. Moreover, since we generated use-cases completely, we can automatically narrow down the number of use-cases to be considered by defining an ODD for the target area. When we applied a defined ODD in Yokohama for Easy. completely derived from situation definitions. Situations. Ride to the use-cases, 727 use-cases were obtained as shown. can be derived from objects, states of them and future be-. in Fig.10. We were able to design use-cases to be considered. havior elements of them. Therefore, the number of situa-. because we defined an ODD with the same ontology as use-. tions can be calculated by simply multiplying the number. cases. Table.3 shows some elements in the defined ODD for. of components of the situations as a combination number.. Yokohama.. Equation.2 is used to calculate the number of situations. Table.2 shows the meaning of each symbol in Equation.2. Si U. = =. E × (R, Rs) × (O, Os, Rel) × F f (Si). 5.. Discussion. Based on the tractable use-cases obtained, we can visual-. (1). ize the overall development status by tracking the develop-. (2). ment status for each use-case. With this framework, we can However, this combination number is too large to treat.. clarify which use-cases should be considered. In our case,. Therefore, after considering the order of priority (visibil-. we divided the development status into four technical fields:. ity/weather attributes), we unified the cases where the ego. perception, cognition, decision making and trajectory con-. vehicle has the same behavior policy as an autonomous ve-. trol in order to follow the progress in details. First of all, all. hicle based on our use-case definition. For example, an. relevant technical elements with the use-cases were listed. oncoming vehicle that will go straight and an oncoming ve-. up in each technical field. We then added the development. Copyright © 2021 Society of Automotive Engineers of Japan, Inc. All rights reserved. 60.

(8) Takuya Nanri et al / International Journal of Automotive Engineering Vol.12, No.2(2021) Acknowledgement. status (e.g. done, created, ongoing, not yet) to each technical element in each technical field. After all the technical. The authors would like to thank Alexandre Armand and. elements were linked to corresponding use-cases, the de-. Javier Iba˜ nez-Guzm´ an for the simulating discussions.. velopment status of use-cases was calculated by using the status of corresponding technical elements. In this case,. References. the development status of a use-case was defined with the. (1) Easy ride. https://easy-ride.com/ (3 Nov. 2020). (2) Japanese Ministry of Economy, Trade and Industry. Toward social implementation on autonomous driving robot (in Japanese). 2019. https://www.meti.go.jp/shingikai/mono_info_ service/jidosoko_robot/pdf/pre_001_04.pdf (3 Nov. 2020). (3) Roman Henze. Validation of automated driving on highways. In Society of Automotive Engineers of Japan Spring Forum, 2019. (4) JAMA. Introduction of research results on high automated driving safety validation (in Japanese). In Society of Automotive Engineers of Japan Spring Forum, 2019. (5) NHTSA. A framework for automated driving system testable cases and scenarios. In U.S. Department of Transportation report, 2018. (6) Gerrit Bagschik, Till Menzel, and Markus Maurer. Ontology based scene creation for the development of automated vehicles. CoRR, abs/1704.01006, 2017. (7) Till Menzel, Gerrit Bagschik, and Markus Maurer. Scenarios for development, test and validation of automated vehicles. CoRR, abs/1801.08598, 2018. (8) A. Armand, D. Filliat, and J. Iba˜ nez-Guzman. Ontologybased context awareness for driving assistance systems. In IEEE IV, pages 227–233, June 2014. (9) M. H¨ ulsen, J. M. Z¨ ollner, and C. Weiss. Traffic intersection situation description ontology for advanced driver assistance. In IEEE IV, pages 993–999, June 2011. (10) L. Zhao, R. Ichise, T. Yoshikawa, T. Naito, T. Kakinami, and Y. Sasaki. Ontology-based decision making on uncontrolled intersections and narrow roads. In IEEE IV, pages 83–88, June 2015. (11) Y. Akagi and T. Morikawa. Simultaneous description of logical design and implementation of automated driving systems. In IEEE IV, pages 1565–1570, June 2019. (12) Society of Automotive Engineers. J3016: Taxonomy and definitions for terms related to driving automation systems for on-road motor vehicles. 2018. (13) Mercedes-Benz Research & Development North America, Inc. Reinventing safety: A joint approach to automated driving systems. 2018. https://www.daimler.com/documents/innovation/ other/vssa-mercedes-benz-and-bosch.pdf (3 Nov. 2020). (14) Aptiv et al. Safety first for automated driving. 2019. www.aptiv.com/docs/default-source/white-papers/ safety-first-for-automated-driving-aptiv-white-paper. pdf (3 Nov. 2020). (15) Waymo LLC. On the road to fully self-driving. 2017. www.mtfchallenge.org/wp-content/uploads/2017/ 02/waymo-safety-report-2017-10.pdf (3 Nov. 2020). (16) S. Ulbrich, T. Menzel, A. Reschka, F. Schuldt, and M. Maurer. Defining and substantiating the terms scene, situation, and scenario for automated driving. In IEEE ITSC, pages 982–988, Sep. 2015. (17) F. Fang, S. Yamaguchi, and A. Khiat. Ontology-based reasoning approach for long-term behavior prediction of road users. In IEEE ITSC, pages 2068–2073, Oct 2019.. slowest status among the ones of corresponding technical elements. The decomposition of development status enabled us to specify what technical elements should be solved by identifying the affected use-cases. Based on the complete use-cases, we can easily extend a development management framework by adding more information to each use-case in this way. Therefore, the obtained AD use-cases have great potential in several applications. Moreover, we can apply generated use-cases described with our ontology to situation understanding and decision making. Since our use-case description includes the prediction of intention of the other objects, it is quite effective, especially in complicated situations. Although previous methods can cover limited use-cases for specific situations[8][17], our use-cases can cover all possible ones in urban area. Ontology description of use-cases allows us to apply first-order logic, Markov logic or Bayesian networks and to help autonomous vehicles to reason about the behavior of other objects in their surrounding. Finally, we can create test scenarios based on our usecases using several detailed parameters. Although test scenarios need more detailed parameters, all the test scenarios can be created by considering careful addition of specific parameters to each corresponding use-case. 6.. Conclusion. In order to clarify which use-cases need to be solved for autonomous driving in urban areas, we generated 2,544 canonical use-cases in an exhaustive way using our completeness-based approach.. These use-cases, defined. with our ontology, enabled us to automatically narrow down the use-cases to consider in a corresponding operational design domain, such as in Yokohama for the Easy Ride project. Consequently, we were able to use it to manage the progress of the development works. These usecases have great potential because they could be applied directly to situation understanding for autonomous vehicles and they could be used in generating test scenarios by varying relevant parameters. In the future, we would like to pursue some new interesting application domains leveraging the generated use-cases.. ∗1 This paper is written based on a proceeding presented at JSAE 2020 Annual Congress.. Copyright © 2021 Society of Automotive Engineers of Japan, Inc. All rights reserved. 61.

(9)

Figure 1: Application area in Yokohama
Figure 3: Examples of upper level of use-cases and road furniture. The first level of object attributes is classified into dynamic and static objects
Figure 5: The concept of scene
Figure 8: Our definition of ODD
+2

参照

関連したドキュメント

When making early preplant surface applications (15 to 45 days prior to planting), use a tank mix of Satellite HydroCap herbicide with other herbicides registered for use in a

Fill the spray tank ½ full with clean water; add the appropriate detergent (follow manufacturer’s directions for use), then fill tank to capacity and operate the sprayer with

Some insects are known to develop resistance to insecticides after repeated use. As with any insecticide, the use of this product should conform to resistance management

• Application Method: Apply specified dosage in sufficient water volume to ensure uniform application and incorporation into the soil using one of the following methods:..

If weeds are emerged at the time of the Incinerate Herbicide application, the addition of a crop oil concentrate (COC) type adjuvant at the rate of 1% v/v or a nonionic surfactant

Zaltus SX, applied as part of a burndown program, may be used for residual weed control, as well as to assist in postemergence burndown of many weeds where field corn will be

When tank mixing or sequentially applying atrazine or products containing atrazine to corn or sorghum, the total pounds of atrazine applied (lbs ai/A) must not exceed the

 Failing to provide return transportation or pay for the cost of return transportation upon the end of employment, for an employee who was not a national of the country in which