[情報] 2019自駕報告:豐田跳脫率比百度高270倍

作者: Scape (non)   2020-02-28 08:48:24
來源:
http://bit.ly/2VsBFHv
Toyota has the highest degree of self-driving cars, 27,000% more than Baidu's
Apollo
Toyota 的無人駕駛車輛有著最高的跳脫率,比起百度的Apollo 高出270倍
Baidu's open source Apollo self-driving car platform is the king of 2019 with
a minimum separation distance per mile.
百度的Apollo 開源無人駕駛測試車在2019年稱王,有著最小的跳脫率。
Baidu
One of the things that makes California unique in the development of
autonomous vehicles is that it requires developers to submit some
Disengagement report Per year.
百度
對於無人駕駛車輛測試來說,每年要求提供無人駕駛脫離報告是加州的一項與
眾不同之處。
This disengagement report-don't confuse with DC's outstanding performance
Dismemberment plan -Is a detailed log of human safety drivers being forced to
control self-driving cars under any circumstances during tests on public
roads, and the circumstances that caused this. This is one of the best
indicators of the public's understanding of how advanced a given company's
self-driving car program is.
這份脫離接觸報告詳細記錄了人類安全駕駛員被迫接管無人駕駛車輛的狀況。
這是大眾了解特定公司的無人駕駛車輛計劃有多先進的最好指標之一。
in the past few years, Waymo Usually, it is dominated by the least
disengagement per mile, but it looks like Chinese Baidu, And Open source
Apollo The self-driving car platform went away and pulled Jon Snow from a
relative in Denvers Targaryen, Mountain View.
過去幾年,Waymo 保持著每英里跳脫率最少的紀錄,但今年來自中國的百度其
Apollo 開源無人駕駛車平台其表現著實讓我們眼前一亮。
Specifically, Baidu achieved a staggering 0.06 departures per thousand miles
driven, and its four development vehicles traveled a total of 100,300.2 miles
in 2019. All its disengagements (only six times in total) occurred on ground
streets, not on highways or freeways. This is a common theme for most
reporting companies.
具體來說,2019年百度Apollo 平台實現了每一千英里里程只有0.06次的驚人跳
脫率,該公司的四輛開發測試車總共行駛了100300.2 英里。所有的(只有六次)
的跳脫狀況都是在地面街道上發生,而不是在高速公路或快速道路上。這也是
大多數公司所報告的普遍現象。
Waymo and Cruise are tied for second. Their clutch per 1,000 miles is only
0.08. For brevity, we will call them Wallys from now on. Baidu. Cruise has
227 cars on the road and travels 831,039.9 miles. Waymo only conducted 147
car tests, but In a typical Waymo wayThey did go a long way-exactly
1,453,137.3. This is equivalent to three round trips to the moon.
Waymo 與Cruise 則是並列第二。他們的每千英里跳脫率只有0.08次。Cruise
共有227輛測試車並且共跑了831039.9 英里。Waymo 有147輛測試車,但Waymo
的總測試里程數更多,有1453137.3英里,相當於可以往返月球三次。
How have these numbers improved since 2018? Facts have proven to be
significant. Best of Waymo-and best of all reports Submitted to 2018 -It's
0.09 Wallis. Baidu's 0.06 is a very serious bump, and even Cruise and Waymo's
own 0.01 Wally decline should not be underestimated.
那這些數據自從2018年後改善了多少?在2018年時,Waymo 最好(也是所有公司
中最好)的跳脫率是0.09。百度今年的0.06跳脫率是一個非常大的衝擊,甚至Cruise
與Waymo 的0.01次的降幅也不可小看。
So now we have seen the best. Which major companies are underperforming in
producing self-driving cars? Well, unfortunately, the top three of this year
are all traditional car manufacturers. Mercedes Benz It was the worst one,
with 140 of them coming from 18 cars and driving 14,238 miles.
前幾名看過了。那哪些大公司在研發無人駕駛車輛時表現不佳?不幸的是,今
年的倒數三名都是傳統汽車製造商。Benz 是其中之一,他們的18輛無人駕駛測
試車一共行駛了行駛了14238英里,跳脫率高達140次/千英里。
The second worst is the Munich people-I mean BMW, of course -Scored 370
Wallys. This was produced by five cars that only traveled 21.36 miles. Maybe
if they pay more, the results will improve, but we have to see how they
perform in 2020.
第二糟的是BMW,跳脫率是370次/千英里。他們只有五輛車共行駛了21.36英里。
Finally, at the bottom of the proverbial bucket is Toyota cars the study. Its
researchers managed a staggering 1,620 of Wally among six cars traveling only
1,817 miles. Especially one car-VIN number JTHDU1EF3G5020098, which was a
liar-completed an amazing 900 breaks on its own in just 534 test miles.
最後,至於備受關注的Toyota,該公司用6輛測試車行駛了1817英里,但卻給出
了跳脫率1620次/千英里 的墊底成績。特別是一輛VIN 為JTHDU1EF3G5020098
的測試車,竟然在短短534英里的測試里程中,跳脫了驚人的900次。
It's easy to joke now, but the truth is that developing self-driving cars is
unusual, crazy, incredible, very, very difficult. Toyota hasn't done this in
a while, and although it obviously has a way to go, we implore it:
Toyota's self-driving cars plan to underperform in 2019, but there will
always be next year!
Oh if you are curious Tesla That's all, rest assured, Elon & company
submitted a disengagement report. The Big T has zero detachment (yes, zero),
but it's important to note that it is only equipped with one car, and the car
only travels 12.2 miles. So yes, fully autonomous driving is Imminent, you
guys.
=====================================================================
加州車輛管理局(DMV) 每年都會要求在該州測試無人駕駛車輛的公司提供數據,
其中平均每一千英里需要人工接管幾次的跳脫率一直倍受關注。
過去一直都是Waymo 領先,而且領先幅度不小。另外一家新創公司Cruise 成績
通常也不錯,然後前幾年就被GM 收購了,所以GM 一直對他們的自駕技術很有
自信。
在加州測試的公司還有一大堆,除了像是Apple、Tesla 這些美國本土公司外,
來自中國的公司也非常的多,2019年的DMV 無人駕駛報告裡前十名有快半是中
國公司:
https://i.imgur.com/Oh5Ykf6.png
另外有趣的是,不曉得是不是因為脫離率被百度追上,Waymo 在這報告發佈不
久後就寫了推特,說脫離率並不能夠代表各公司實際的研發狀況好壞。
╭───────────────────────────────╮
│https://twitter.com/Waymo/status/1232758355082375168 │
│ │
│Waymo │
│ │
│1/7 We appreciate what the California DMV was trying to do │
│when creating this requirement, but the disengagement metric │
│does not provide relevant insights into the capabilities of │
│the Waymo Driver or distinguish its performance from others │
│in the self-driving space. │
╰───────────────────────────────╯
不過Waymo 會這樣說是正確的,因為這份報告是由各公司自己提供,其實際狀
況可能會不一樣,或許有些公司選擇的路線比較簡單、或許有別的因素,簡單
的說就是測試的基準並不相同。
DMV 報告連結:
http://bit.ly/32zsert
作者: ghjkl5566 (56不能王)   2020-02-28 09:23:00
教主做的好!像偷油塔這種不思進取的車廠,就該讓它像Nokia一樣消失
作者: lucky945 (液友)   2020-02-28 09:30:00
什麼系統不會出錯 就算跳脫也是駕駛該負責馬上接管
作者: bluemkevin (WHO WHO)   2020-02-28 09:56:00
Toyota 很積極在開發阿
作者: arcross (阿插)   2020-02-28 09:57:00
標題黑豐田ㄛ 哪家的系統是100%不會出錯的 銷量說明一切愛作弊德國車墊底不意外 做不到100%就不要開上路害人了
作者: awayaway (away)   2020-02-28 09:58:00
“什麼系統不會出錯,人要cover啊”這句話在航空版討論737-max 兩起墜機意外的時候也有人說過.
作者: bluemkevin (WHO WHO)   2020-02-28 09:59:00
@@ 沒有開上路害人吧?而且跳脫率高不代表不安全阿
作者: G8man (中肯伯)   2020-02-28 10:04:00
所以跳脫率是什麼QQ?
作者: bcs (= ="frailty..gggg XD)   2020-02-28 10:07:00
正常,日系對意外反應是差。學納牌直接台灣三寶對決,AI功力一定進步一甲子
作者: d8613518 (張森)   2020-02-28 10:18:00
總比不跳脫直接撞毀害死車主好吧?
作者: yytseng (yytseng)   2020-02-28 10:34:00
lv2的就是煩,但有些可以假手做弊騙偵測,所以常出事
作者: hiphoprover (浪子)   2020-02-28 10:41:00
這是在美國測試的,有種來台灣測看看平面道路的 LOL
作者: kenation (BBS-KEN)   2020-02-28 11:13:00
車子跳脫率高,駕駛才不會真的跳脫。跳脫率低,車主就跑去後座睡覺吃便當了(逆向思考一下)
作者: thid5335 (討推專家)   2020-02-28 11:19:00
教主怎沒翻最後一段 特只有12.2麥的紀錄 不知道是不是因為有廣大特粉幫忙debug
作者: DYE (跟腦殘爭最後都會變腦殘.)   2020-02-28 11:59:00
關閉這陰影模式 是否能提升續航里程?畢竟電跟車是車主的
作者: bluemkevin (WHO WHO)   2020-02-28 12:03:00
電腦模擬無法反映出硬體故障的處理機制,特斯拉一定有實驗車運行
作者: darktasi (桃園達克大叔)   2020-02-28 12:53:00
中國文章看的頭好痛,電車好棒棒?
作者: streit (淺紫彩雲)   2020-02-28 21:19:00
我只知道中國不可信。
作者: supereva (eva)   2020-02-28 21:42:00
報告由各公司自行提供?? 那還有什麼參考價值..至少要統一標準測試吧
作者: Atkins13 (Atkins13)   2020-02-28 23:13:00
中國的基本上不可信 其它的可再討論
作者: axlfun (撿到一塊錢)   2020-03-02 14:35:00
百度也很認真搞這塊一陣子了

Links booklink

Contact Us: admin [ a t ] ucptt.com