Transmitted by the expert of Japan

Presentation on theme: "Transmitted by the expert of Japan"— Presentation transcript:

1 Transmitted by the expert of Japan
Informal Document VMAD-02-05 2nd VMAD IWG, April 1-2, 2019 Agenda item 7. How to consider Innovative Safety Validation Methods of Automated Driving System(2nd Report) April 1-2, 2019 Japan page1

2 Explain Scenario Structure Generation Process in detail today
Overall approach Reprint Explain Scenario Structure Generation Process in detail today page2

3 Perception limitation
Overall Scenario Structure Safety scenario structure can be classified from the view point of the physics of AD system, which includes perception, judgment and maneuver in order to cover holistic root cause. Scenario Structure Perception Judgement maneuver Perception limitation Traffic Disturbance Vehicle Disturbance sensor malfunction Traffic participants’ unsafe behavior Cause of vehicle instability Scenarios caused of performance limits of recognition. On board sensors are Camera, Radar, LiDAR. Scenarios caused of road geometry and surface condition. Explain this scenario progress, today ☛Data Sources and operational methods is planed to be reported next time. page3

4 Scope of the scenario Normal Pre-Critical Critical Unavoidable
When we evaluate Automated Driving System safety performance, the scope of scenario should be specified as critical situation. Normal Pre-Critical Critical Unavoidable Normal driving Potentially hazard emergency Scope Figure 1 shows a schematic of the posssible events that may occur from normal driving to a collision and highlights the events covered by the safety evaluation methodology scope specified in the current report. From the perspective of AD systems safety, Safety performance and Performance limits can be evaluated against unreasonable behavior of the surrounding vehicles. Safety performance refers to events that do not approach a hazard (sign) and occur before a safety margin decrease due to unreasonable behavior of the surrounding vehicles. Performance limits refer to events after the safety margin decreases due to the surrounding environment, or the behavior of other vehicles or traffic participants. Safety margin refers to time and space ranges required to secure safety of the ego-vehicle and the surrounding traffic participants. When considering the possibility that the ego-vehicle collides with the surrounding vehicles, Performance limitations in the current report includes situations in which drivers may not avoid a collision, but AD systems may do. Target zone of scenario page4

5 Traffic participants’ unsafe behavior
Traffic disturbance Scenario Structure Traffic disturbance structure can be classified into road geometry, ego-vehicle behavior and Other traffic participant’s position/motion. ⇒ Holistic scenario catalog can be derived by combination of these elements. Traffic disturbance Traffic participants’ unsafe behavior Overall approach A conceptual scheme of the traffic scenario structure is presented in Figure . The scheme accounts for traffic disturbances for traffic scenarios based on Driving Environment and Surrounding Environment. Driving environment comprises Road geometry and Ego-vehicle behavior. Surrounding environment comprises surrounding vehicles location (with respect to the ego-vehicle), and surrounding vehicles motion. page5

6 Driving environment classification
Comprehensively classified road geometry and ego-vehicle behavior from highway scene. 　 ・Road geometry is classified into 4 types (main road, merging lane, departure lane, ramp). ・Based on the premise that AD vehicles keeps appropriate distance and speed, spontaneous ego vehicle behavior related to safety can be integrated into lane keeping and lane changing. ・ Consider possibility of affecting the driving judgment of AD due to differences in the driving scene (road geometry) even with the same vehicle behavior. The same lane change operation, but the driving judgment of the AD vehicle may be different depending on the driving scene. Risk factor Lane decrease, Speed difference... Driving environment classification and parameterization The driving environment is defined by combining the classifications of the road geometry and of the ego-vehicle behavior. A)Road geometry In this document, highway geometry definition based on four major categories (Main road, Merging lane, Departing lane, and Ramp) is adopted from the road structure ordinance [8]. The ordinance includes further details on sub-categories and on detailed safety-relevant parameters such as cross sections, horizontal sections and others such as sight distance related. The modifications may be required based on the regions of applications. B)Ego-vehicle behavior A lane change maneuver from a contiguous line or from a merging lane, may differ in road geometry category, but share the ego-vehicle behavior. The same holds for lane keeping. Therefore, possible ego-vehicle behaviors are comprehensively categorized in Lane Keep and Lane Change categories. Risk factor Curve R（sort sight distance） page6

7 Surrounding environment classification
Defines the position and behavior of surrounding vehicles involved in safety. ・Define the initial position of the surrounding vehicle in 8 directions around ego-vehicle (Define the vehicle ahead as "+1" in consideration of secondary obstacles) ・Classify the movement of surrounding vehicles into 5 types of vertical and horizontal movement. ・Focus on the location and behavior of nearby vehicles that may affect the driving decisions of AD vehicles.　 ⇒　Rationalize the evaluation range Direction May affect the driving judgment of AD vehicles Surrounding vehicle motion transition Surrounding vehicle initial location vehicle position Cut in Cut out Acceleration Deceleration (Stop) Sync 1.Lead(L) 2.Following(F) 3.Parallel(Pr-f) 4. Parallel(Pr-s) 5. Parallel(Pr-r) 6. Parallel(Pl-f) 7. Parallel(Pl-s) 8. Parallel(Pl-r) Surrounding environment classification and parameterization The surrounding environment of traffic disturbance is defined as a combination of the surrounding vehicles location and motion (Figure 12). For the definition of the surrounding vehicles location, additional information to account for vehicle type specific characteristics (e.g. motorcycle and car partially sharing a lane) may need to be incorporated. The location of the surrounding vehicles to be considered in the safety evaluation is defined according to adjacent locations in eight directions around the ego-vehicle, as these may invade the ego-vehicle’s trajectory. In addition, when there is a large speed difference between the leading vehicle and the vehicle ahead of the leading vehicle, the former may cut out to avoid a collision. If this cut out occurs suddenly, the oncoming ego-vehicle may also need to intervene for crash avoidance. To consider for this possibility, the location of the vehicles ahead of the leading vehicle is also considered and is noted as ‘+1’（Figure 13, left) The motion of the surrounding vehicles is classified into five types: cut-in, cut-out, acceleration, deceleration/stop, and synchronization. From the perspective of safety evaluation, it is possible to minimize the number of evaluation tests by focusing on the motion of the target participants that may obstruct the ego-vehicle's behavior (case 1 to 8, Figure 13, right). page7

8 Concept of scenario creation
Basically, traffic disturbance scenario can be described with two vehicles: ego-vehicle and the subject vehicle. ・In response to the movement of traffic participant around ego-vehicle (1st Act), ego-vehicle makes a cognitive decision and takes some action (2nd Act). ・On a traffic environment where multiple traffic participants exist, it is important to evaluate severe scene caused by this secondarily generated vehicle movement. Example) 1st act Deceleration Scene.1 Scenario Ego L(1) a) b) c) Deceleration Steering Acceleration Crash avoidance by ego-vehicle steering Figure A.6 illustrates an example of crash avoidance by ego-vehicle steering in a combined scenario involving a primary lead vehicle deceleration (L1), followed by an ego-vehicle lane change in conflict with a vehicle traveling straight in the contiguous lane (Prr5). The primary deceleration by the leading vehicle (L1) influences lateral acceleration of the ego vehicle. A most-severe condition can also be defined for safety evaluation by fixing the safety target to the maximum possible lateral speed. Also, similar to the previous example, the safety of the ego-vehicle lane change towards Pr-r5 can be treated as a simple cut-in scenario according to the scenario classification. From the examples provided, the safety evaluation primary action is covered by means of most severe approach. The secondary action of the ego-vehicle in a traffic environment involving multiple vehicles, by using the specified maximum value of the ego vehicle as a parameter, it is possible to cover by the scenario structure. Therefore, it is regarded that there is no need to define new scenarios combining multiple vehicles. Ego L(1) Scene.2 Pr-r(5) Ego 2nd act Steering Pr-r(5) page8

9 Cut in scenario Lane change scenario of ego vehicle or parallel running vehicle 　・The scenario trajectory intersects is a possibility of a collision occurring 　・Approach to the rear of ego vehicle takes into consideration of affecting AD vehicle path planning surrounding vehicle position Ego vehicle behavior description Lane keep Lane change 1.Lead(L) There is no Cut in scenario because two vehicles drive on the same lane. 2.Following(F) 3.Parallel(Pr-f) Cut in scenario ahead of ego- vehicle. Trajectory crosses in the way. Consider possibility of collision occurrence. 4. Parallel(Pr-s) Pull over scenario for ego-vehicle. Assume unreasonable lane change of surrounding vehicles. 5. Parallel(Pr-r) Lane change to the rear of ego-vehicle. Consider possible affect AD vehicle path planning. 6. Parallel(Pl-f) 7. Parallel(Pl-s) 8. Parallel(Pl-r) Ego L Ego L F Ego F Ego Ego Ego Pr-f Pr-f Ego Ego Pr-s Pr-s カットインシナリオは自車の前後(L:1,F:2)以外の他車両が、自車の走行起動と交差する状況である この結果、自車と他社が衝突するリスクがある 例えば、自車の挙動はレーンキープしている状態と右前に走行している他車両(Pr-F：３)の相対速度や相対距離によって衝突する リスクを持っている 同様に、自車がレーンチェンジする挙動では、他社との走行経路がクロスすることによって衝突のリスクがある また、右後方の他車両(Pr-r:5）で自車がレーンキープもしくはレーンチェンジの挙動では、AD車両の行動によっては危険な状況に陥る場合も あることを考慮し確認すべきシナリオであると考えた Ego Ego Pr-ｒ Pr-ｒ Same as above 3, 4, 5 (Symmetrical scenario) page9

10 surrounding vehicle position
Cut out scenario Lane change to ego vehicle 　・In the cut-out scenario of the leading vehicle (L), a collision with the vehicle ahead is assumed 　・In the cut-out scenario of the following vehicle (F), influencing AD LC-judgment is considered surrounding vehicle position Ego vehicle behavior description Lane keep Lane change 1.Lead(L) In the cut-out scenario of the leading vehicle (L), a collision with the vehicle ahead is assumed. 2.Following(F) No trajectory intersects, it doesn’t affect any safety. In the cut-out scenario of the following vehicle (F),influencing AD LC-judgment is considered. 3.Parallel(Pr-f) 4. Parallel(Pr-s) 5. Parallel(Pr-r) 6. Parallel(Pl-f) 7. Parallel(Pl-s) 8. Parallel(Pl-r) Ego L +1 Ego L F Ego カットアウトシナリオは、自車の前方車両(L:1)が他レーンに移動した場合を想定したシナリオである この場合のリスクとして、例えば自車がレーンキープしていて、前方車（L)がカットアウトした先の車両（＋１）が低車速もしくは停止していると 自車が前々車と衝突する恐れがある。また、自車が何らかの判断によっては同時にレーンチェンジを行い、前車（L:1）と衝突する恐れがある。 同様に後車（F:２）がカットアウトしたとき、自車がレーンチェンジを行うと衝突する恐れがある page10

11 Acceleration scenario
Acceleration to ego-vehicle 　・The scenario where trajectory intersects may cause an collision 　・Approaching to ego vehicle with s/vehicle acceleration, influencing AD LC-judgment is considered surrounding vehicle position Ego vehicle behavior description Lane keep Lane change 1.Lead(L) It is an operation to leave ego-vehicle, it doesn’t affect any safety. 2.Following(F) Ego-vehicle LK assumes the rear collision by the following vehicle. Ego-vehicle LC is considered surrounding vehicle approaching from behind. 3.Parallel(Pr-f) 4. Parallel(Pr-s) 5. Parallel(Pr-r) No trajectory intersects, it doesn’t affect any safety. Acceleration scenario of surrounding vehicle when ego- vehicle LC. Trajectory crosses in the way. Consider possibility of collision occurrence. 6. Parallel(Pl-f) 7. Parallel(Pl-s) 8. Parallel(Pl-r) F Ego F Ego 加速シナリオは、他車両が自車に対し相対速度が早い場合かつ自車の走行軌跡と交差するときに起こるリスクである 例えば、自車がレーンキープしているとき、後車が自車に向かって加速してくる場合や、自車がレーンチェンジしようとしたときに右後方車両( Pr-r)が加速した時を想定している。この状況は、自車が追い越しレーンに入るときに起こる。 Pr-r Ego Same as above 3, 4, 5 (Symmetrical scenario) page11

12 Deceleration scenario
Deceleration to ego-vehicle 　・The scenario where trajectory intersects may cause an collision surrounding vehicle position Ego vehicle behavior description Lane keep Lane change 1.Lead(L) Deceleration scenario of forward vehicle. Ego-vehicle slow down and avoid when ego-vehicle LK. Ego-vehicle LC may contact forward vehicle. 2.Following(F) No trajectory intersects, it doesn’t affect any safety. 3.Parallel(Pr-f) Deceleration scenario of ego-vehicle LC. Trajectory crosses in the way. Consider possibility of collision occurrence. 4. Parallel(Pr-s) It is an operation to leave ego-vehicle, it doesn’t affect any safety. 5. Parallel(Pr-r) 6. Parallel(Pl-f) 7. Parallel(Pl-s) 8. Parallel(Pl-r) Ego L L Pr-ｆ 減速シナリオは、自車の走行軌跡上で他車が何らかの状況で減速もしくは急減速した場合に起こるリスクである 例えば、自車がレーンキープしていて、他車（L:2)が急減速した場合、自車は他車に衝突する恐れがある これはレーンチェンジした場合も同様である Same as above 3, 4, 5 (Symmetrical scenario) page12

13 Sync scenario on the main road Sync scenario at merging
Synchronized scenario Synchronization to ego-vehicle 　・Assumed safety evaluation in the case of ego vehicle LC not being possible due to parallel vehicle behavior Sync scenario on the main road Sync scenario at merging Ramp road is same Branch is same Sync Scenario ■Assumed scene in the Sync scenario (All are included in Sync scenario as "LC impossible scenario") ①Concession 並走シナリオは、目的地に向かうため予めレーンを変更し、分岐に備えるためのレーンチェンジや、本線での合流路で、並走する車両によって走行経路を 塞がれる場合を想定している また、このシナリオは、一般的に行われる“譲り合い”または“悪意のある妨害”によってよくある状況である さらに、合流路が渋滞の場合も、自車にとっては同様な状況である ②Malicious interference ③Traffic jam page13

14 Influence on LC judgment?
Scenario description example Critical scenes can only be described with a one-to-one relationship, and evaluation scenarios are the time series of them. Example) ■Cut-out scenario of the leading vehicle (L) Scene 1 Scene 2 Scene 3 Scene 4 Ego L +1 Ego-vehicle follows forward vehicle (L) For the last car (+1) deceleration, forward vehicle (L) overtakes (Cut out) Ego vehicle suddenly approaching decelerating vehicle ahead Collision may occurs because ego-vehicle's deceleration is not in time ■Cut-out scenario of the following vehicle (F) ここまでの各シナリオについてはクリティカルなシーンで表現してきたが、実際のテストシナリオは、シーンの時系列によって表現される 例えば、カットアウトシナリオでは 自車、他車の配置が設置され、次に前者が設定時間にカットレーンチェンジを開始し、レーンチェンジ完了後、前々車両の状態が現れ（例えば停止している） 自車が回避出来なかった場合は衝突する 下の段の場合のカットアウトでは 自車の後ろに他車が配置されており、ある設定時間で自車、他車がレーンチェンジし始め、レーンチェンジ後、自車、他車の相対速度、相対距離によっては 衝突する このように、本来のシナリオは時系列で表されており、その状況でクリティカルになるパラメータ（ここでは例えば、他車のカットアウト開始時間、自車、他車の速度 、相対距離、前々車両の速度など）を設定するので膨大なシナリオが出来上がる Scene 1 Scene 2 Scene 3 Scene 4 Influence on LC judgment? F Ego Ego-vehicle is LC judgment Following vehicle (F) follows its ego-vehicle Start ego-vehicle LC operation At the same time, following vehicle (F) starts LC (Cut out) In the consequences, following vehicle approaches behind ego-vehicle page14

15 Traffic disturbance scenario on highway
Extract 32 types of scenarios according to combination of driving environment and surrounding environment 　・It is necessary to define and quantify the parameters of surrounding vehicles such as longitudinal and lateral vehicle speed, acceleration, deceleration and distance for each scenario Taxonomy for Automated Driving Vehicle Scenario As a result of the systematization process proposed, a streamlined methodology to structure traffic scenarios that considers road geometry, ego-vehicle behavior, and surrounding vehicles location and motion can be developed. For each scenario, parameters of the surrounding vehicles such as acceleration, deceleration, speed, and distance should be defined and quantified. page15

16 Summary Traffic disturbance Scenario on highway:
Road geometry is classified into 4 types：main road, merging lane, departure lane, ramp Ego-vehicle behavior is 2 types; Lane keep and Lane change. Scenarios can be described by 8 (+1) locations of surrounding vehicles. Critical scenes can be described with a one-to-one relationship, and evaluation scenarios are the time series of them. Please check if there are any incomplete parts about Traffic disturbance Scenario of today’s proposal. page16