Full automatic testing concept for Autonomous Driving System based on Deep learning approach
- 提供方法
- 本サイト上にてダウンロード・閲覧可
- 形態
- 価格
- 一般価格(税込)：￥1,100 会員価格(税込)：￥880
- 文献番号
- 20219074
- 文献・情報種別
- その他の国際会議
- 掲載ページ
- 1-5(Total 5 p)
- 発行年月
- 2021年 9月
- 出版社
- (公社)自動車技術会
- 言語
- 英語
書誌事項
| 著者（英） | 1) Hadj Hamma Tadjine, 2) Mohamed Marouf, 3) Sofiane Ouanezar, 4) Sylia Baraka |
|---|---|
| 勤務先（英） | 1) IAV GmbH, 2) IAV France, 3) IAV France, 4) IAV France |
| 抄録（英） | Autonomous driving is one of the most complex problematic that is currently tackled. It is expected to increase road safety, redesign urban areas and push new industry branches. The industry rely on cloud and Big Data technologies not only for storage but also for data analysis and patterns finding. Thus, with the increasing volume of the data generated by the vehicle sensors, it became more and more important to process it to get the most value out of it for safety guarantee, test and validation of autonomous driving system. Two of the most interesting applications from the sensors data are automatic scenarios detection and modelization of the scenarios in a Domain Specific Language for test and validation in simulation environment. Automatic use case detection responds to both of the problematic of the growing volume of the data generated by vehicle sensors and to the problematic related to the scenario manual annotation. For, it is still a common practice to annotate manually the detailed scenarios out of the vehicle sensor data. This kind of annotation is prone to human error, to over focus on common scenarios at the expense of new ones that were either never seen before or not taking into account because of biased presumptions. Modelization of scenarios in a Domain Specific Language enable to ensure the reliable execution of planning and control functions in all situations for which they can be activated. It enables not only to have a more efficient catalogue test of scenarios that are generated from vehicles sensors but also enable the injection of the model in simulation environment. On the other hand, the level 3-5 of the SAE level driving automation are based on multiple perception functions and AI based functionalities that cannot be tested in the same way for the level 1 and 2. The challenge for the introduction of higher levels of automations is to guarantee that the system functions behave in a safe way. For ADAS, the proof is provided by driving many tests kilometers and test grounds and public roads. However, for higher levels of automation a distance-based validation is not an acceptable solution for safety insurance. Moreover, it has been affirmed by Wachenfeld and Winner [1] that it is not economically acceptable, for 6.62 billion test kilometers have to be made in order to hypothetically prove (with 50% chance) that the Autonomous Driving Systems are twice as good as a human driver. Hence, the large researches that are conducted in a way to introduce new metrics for higher levels automation systems validation. Scenarios based test – proposed by [2], is applied more and more as a mean to test and validate higher levels of automation. A new approach which address a full automation test methods based on CNN for autonomous systems AD level 3+ is addressed and presented. 翻訳 |
