穩(wěn)步發(fā)展的激光雷達(dá)技術(shù)將協(xié)助自動駕駛系統(tǒng)實現(xiàn)重大飛躍，但前提是必須降低成本。
顧名思義，激光雷達(dá)是指“以發(fā)射激光束探測目標(biāo)的位置、速度等特征量的雷達(dá)系統(tǒng)”。事實上，人類歷史中對激光雷達(dá)的應(yīng)用可以追溯回 1962 年 5 月。當(dāng)時，Louis Smullin 和Giorgio Fiocca 兩名電氣工程師曾在美國麻省理工大學(xué)的林肯實驗室中，利用一部 12 英寸的望遠(yuǎn)鏡向月球發(fā)射 50 焦耳的激光脈沖，并使用一部 48 英寸的望遠(yuǎn)鏡接收反射光，而后通過光的傳播速度估算出地球與月球的距離。
40 年后，除了在氣象學(xué)和阿波羅登月計劃中發(fā)揮重要作用外，激光雷達(dá)還作為關(guān)鍵傳感器，成為美國國防部DARPA 挑戰(zhàn)賽中機器人汽車原型中的關(guān)鍵組成部分。在最初的機器人汽車設(shè)計中，設(shè)計人員將多達(dá) 64 部激光雷達(dá)和傳感器集成至安裝在車頂?shù)木薮髨A柱體中。車輛在行駛時，這個圓柱體系統(tǒng)將不斷旋轉(zhuǎn)，向四面八方發(fā)射成千上萬條脈沖，從而提供 360 度車輛環(huán)視信息。具體來說，激光雷達(dá)掃射的目標(biāo)將被標(biāo)記為“點云”（Cloud of points），補充雷達(dá)和攝像頭的成像效果，共同引導(dǎo)無人駕駛汽車的行駛。
到 2007 年，也就是DARPA 挑戰(zhàn)賽的第四年時，賽事的冠軍作品及所有 6 件完成作品中的 5 件均沿用了激光雷達(dá)技術(shù)先鋒Velodyne Acoustics 公司推出的早期激光雷達(dá)。據(jù)報道稱，當(dāng)時每個激光雷達(dá)單元的成本高達(dá)近 8 萬美元。
盡管軟件和硬件方面都有了重大改變，但如今激光雷達(dá)的工作方式卻與之前的“祖輩”并無太大變化。專家表示，目前，汽車行業(yè)中主要采用久經(jīng)驗證的機械掃描型（MEMS）激光雷達(dá)，而新的固定式（無移動組件）激光雷達(dá)的性能更可靠、尺寸更緊湊，這對于將設(shè)備集成至車輛外飾中至關(guān)重要。對比來說，固定式激光雷達(dá)的視場（field of view），也就是可視范圍較窄，但憑借相對更低的成本，廠商也可以選擇同時安裝多款固定式激光雷達(dá)，從而達(dá)到擴大視場的目的。
通常而言，常見的汽車或卡車可以通過安裝大概8 個可視角度較小的激光雷達(dá)（車輛前后選擇可視角度 120 度的激光雷達(dá)、左右選擇可視角度 90 度的激光雷達(dá)）替代360 度的大型激光雷達(dá)，當(dāng)然具體數(shù)量仍與車輛的幾何外形有關(guān)。從另一方面來看，車輛對激光雷達(dá)探測距離的需求也可以降低。工程師可以根據(jù)激光雷達(dá)在車輛安全系統(tǒng)中發(fā)揮的作用，對安裝在車輛不同位置的激光雷達(dá)提出不同的探測距離要求，通常左右方向的激光雷達(dá)要求達(dá)到 30 米、前方要求200 米、后方要求 50 米。
隨著越來越多的初創(chuàng)公司和成熟的供應(yīng)商均紛紛“殺入”SAE 3 級、4 級和 5 級自動駕駛功能的開發(fā)中，激光雷達(dá)的研發(fā)正在如火如荼地開展。此外。這項技術(shù)在需求量巨大的 3D 地圖領(lǐng)域也有重要應(yīng)用。
Frost & Sullivan 公司預(yù)測，到 2025 年，激光雷達(dá)的年銷量將達(dá)到600 萬部，其中 50% 均將應(yīng)用至汽車中，市場規(guī)模預(yù)計將達(dá)到 20 億美元。經(jīng)《汽車工程》統(tǒng)計，激光雷達(dá)領(lǐng)域中的初創(chuàng)公司數(shù)量多達(dá) 30 多家，還有Velodyne 等老牌廠商及包括博世（Bosch）、法雷奧（Valeo）及大陸集團（Continental）等在內(nèi)的一級供應(yīng)商。此外，Aptiv 和麥格納（Magna）等頂級供應(yīng)商選擇與一些激光雷達(dá)專家公司合作。一些汽車廠商也開始通過收購或合作等手段，與激光雷達(dá)開發(fā)商建立聯(lián)系，包括BMW（通過麥格納與Innoviz 合作）、通用（與Strobe 合作）、豐田（與Luminar 合作），及福特（與Princeton Lightware 和 Velodyne 合作）。
有一些支持者稱激光雷達(dá)是解決“自動駕駛汽車關(guān)鍵難題的靈丹妙藥”，這可能有些夸張，但無論如何，在自動駕駛汽車研發(fā)方面，激光雷達(dá)無疑是最熱門，也最有爭議的傳感器技術(shù)。
技術(shù)顧問Gerald Conover 在移動出行領(lǐng)域的經(jīng)驗非常豐富。他觀察道，“每隔幾周就會有一家新公司大肆鼓吹他們的全新激光雷達(dá)技術(shù)。但事實上，他們口中的成果很多仍僅基于實驗室項目結(jié)果，而真正的大批量生產(chǎn)可能還需要幾年時間。”只有那些成本低且性能高的產(chǎn)品才能真正存活下來。根據(jù)SAE 4 級和5 級自動駕駛車輛的發(fā)展情況來看，“未來，市場對激光雷達(dá)設(shè)備的需求可能很大。”
成本仍是汽車級激光雷達(dá)設(shè)備實現(xiàn)大規(guī)模部署的一大障礙。目前，開發(fā)版激光雷達(dá)的成本仍在10000 美元或更高 — “對于汽車生產(chǎn)來說，這么高的成本是不現(xiàn)實，”Conover 打趣道，“而降低激光雷達(dá)成本的關(guān)鍵技術(shù)僅掌握在少數(shù)幾家供應(yīng)商手中。”汽車廠商希望，未來，激光雷達(dá)的商品價格可以穩(wěn)步降低，正如曾經(jīng)的車載雷達(dá)和攝像頭一樣。
自2017 年末以來，Quanergy 公司一直在生產(chǎn)905 納米的固態(tài)激光雷達(dá)，其反射率為8 %，測距范圍為150 米。該公司聲稱，我們的激光雷達(dá)為光學(xué)相控陣類型，每秒可以掃描50 萬個點，100 米外的最小可監(jiān)測光點單位為3.5 厘米。Quanergy 公司認(rèn)為，目前，每個樣品的價格仍為數(shù)千英鎊，但公司有信心在未來的量產(chǎn)中，將產(chǎn)品的單位成本降低至300 美元之下。目前，Velodyne 公司最便宜的 16 束激光雷達(dá)每臺售價為4000 美元。
物盡其用，各司其職
雖然汽車激光雷達(dá)領(lǐng)域的競爭已經(jīng)達(dá)到白熱化程度，但并非所有原始設(shè)備制造商都認(rèn)為這項技術(shù)勢在必行。比如，特斯拉（Tesla）的自動駕駛系統(tǒng)就另辟蹊徑地采用了基于攝像頭的光學(xué)識別技術(shù)。這家公司的老板Elon Musk 似乎并不認(rèn)為激光雷達(dá)將改變自動駕駛領(lǐng)域的游戲規(guī)則。2017 年4 月的TED 演講舞臺上，“一旦你能解決攝像機的視覺問題，就能解決自動駕駛的問題；如果你不解決視覺問題，自動駕駛的問題自然無解。”Musk 表示，“只要有攝像頭，你絕對可以實現(xiàn)成為超人的夢想。”此外，這位CEO 還暗自給激光雷達(dá)貼上“拐杖”的標(biāo)簽。
本田北美公司（Honda North America）產(chǎn)品規(guī)劃助理副總裁Jay Joseph 指出，公司正在“利用激光雷達(dá)進行開發(fā)和測試”。他表示，我們的工程師認(rèn)為，激光雷達(dá)在短期內(nèi)是必要的。“當(dāng)然，從長遠(yuǎn)來看，我們希望看到其他解決方案——也許是一些借助網(wǎng)聯(lián)功能和共享信息的解決方案。但是，在新解決方案確保萬無一失之前，激光雷達(dá)可以為車輛提供高質(zhì)量的信息，讓車輛能夠基于這些信息做出正確的決策，因此是必不可少的。”
目前，對于包括Aptiv 等很多經(jīng)驗非常豐富的一級供應(yīng)商而言，很大程度上，他們的工作是將傳感器陣列組裝完美地集成至車輛中。Aptiv 公司高級技術(shù)副總裁Jada Smith 指出，“一些科技初創(chuàng)公司還對汽車不夠了解，但我們明白傳感器是如何與車輛協(xié)同工作的。”
Smith 表示，Aptiv 認(rèn)為攝像頭、雷達(dá)和激光雷達(dá)將在汽車自主技術(shù)中發(fā)揮協(xié)同作用。她稱，激光雷達(dá)是“一項必要技術(shù)，適用于所有用例，可作為一種冗余系統(tǒng)，并幫助車輛看清周圍發(fā)生的一切。”目前，Aptiv 已經(jīng)投資了Leddartech（flash 激光雷達(dá)公司）、Innoviz（MEM 型激光雷達(dá)）和Quanergy（光學(xué)相控陣型激光雷達(dá)）等公司，對各類激光雷達(dá)技術(shù)均有涉及。
在選擇激光雷達(dá)類型時，最重要的是搞清楚激光雷達(dá)的角色。“我們希望讓激光雷達(dá)發(fā)揮什么作用？”Smith問道，“探測距離越遠(yuǎn)，視野越窄——這與攝像頭的情況一樣。因此，我們首先應(yīng)該確定到底要什么性能，然后再選擇Flash、MEMS 或其他類型的激光雷達(dá)。但關(guān)鍵是要根據(jù)我們希望達(dá)到的目標(biāo)，進行權(quán)衡。”
最近，《SAE 汽車工程》深入采訪了一群激光雷達(dá)創(chuàng)新研究人員，下文將介紹這些研究人員對激光雷達(dá)技術(shù)對了解和看法。
Innoviz Technologies
如同許多傳感器技術(shù)開發(fā)商一樣，Innoviz 公司的總部設(shè)在以色列，所有主要負(fù)責(zé)人均擁有以色列軍方的專業(yè)電子背景。這家公司成立于三年前，全球共有大約150 名員工。目前，Innoviz 公司已經(jīng)獲得超過8000 萬美元的投資，Aptiv、Magna 和三星在Innoviz公司均有持股。
激光雷達(dá)的基礎(chǔ)技術(shù)在于微機電系統(tǒng)（MEMS）的設(shè)計，即通過移動的鏡面投射來自固態(tài)芯片的掃描激光。更為重要的是，Innoviz 還推出了 905 納米激光雷達(dá)。Innoviz 北美總經(jīng)理Aditya Srinivasan 表示，905 納米激光雷達(dá)可以幫助公司有效降低成本。事實上，Innoviz 公司預(yù)計將在 2019 年推出一款定價僅在數(shù)百美元的汽車級激光雷達(dá)傳感器，這也是公司的首款汽車級激光雷達(dá)。
Srinivasan 表示，“有關(guān)我們的設(shè)計到底是不是固定式激光雷達(dá)，大家可能會有不同的看法，因為我們確實使用了一面移動的鏡子，但無論如何，這就是我們公司的‘固定式’激光雷達(dá)。”
Innoviz 公司的資料顯示，InnovizOne“經(jīng)過專門設(shè)計，可以輕松無縫集成至任何大眾市場車輛中”。該系統(tǒng)的水平和垂直場視角度分別為120° 和25°，可提供750 萬像素/秒的高清分辨率，幀率為25 幀/秒，探測范圍達(dá)250 米（820英尺）。從外觀而言，InnovizOne 高 50 毫米、寬 110 毫米、進深 100 毫米（即2 x 4.3 x 3.9英寸）。此外，系統(tǒng)還采用了一款專利信號處理芯片，進行數(shù)據(jù)管理工作，但Srinivasan 并未透露這家芯片合作廠商的名稱。
Innoviz 還聲稱，InnovizOne系統(tǒng)經(jīng)過專門優(yōu)化，可以識別反射率極低的物體，這也是目前很多其他激光雷達(dá)很難做到的。
4 月，Innoviz 還公布了通過公司供應(yīng)商合作伙伴麥格納（Magna）與BMW 簽訂的供貨協(xié)議。BMW 表示，公司計劃在2021 年推出一項SAE 3 級自動駕駛汽車叫車服務(wù)，而來自Innoviz 的激光雷達(dá)無疑將成為其中的關(guān)鍵組件之一。
此外，Innoviz 還與Aptiv 和三星共同合作，開發(fā)汽車激光雷達(dá)。6 月，Innoviz 表示，公司已經(jīng)與中國一級汽車供應(yīng)商恒潤科技建立了合作伙伴關(guān)系，而后者的客戶包括多家中國大型汽車廠商，目前正在將Innoviz 的產(chǎn)品集成至自家的自動駕駛平臺。
Aptive 公司的固定式激光雷達(dá)（圖片來源：Aptiv）。
TetraVue
TetraVue 公司的固定式 Flash 激光雷達(dá)技術(shù)借鑒了數(shù)碼相機相對成熟，且更為平價的傳感器技術(shù)，特別之處在于可以提供更高的分辨率。事實上，這家公司稱其汽車激光雷達(dá)為“高清4D 攝像頭”，可提供百萬像素級數(shù)字視頻捕捉，通過像素級深度信息支持遠(yuǎn)程傳感需求。
TetraVue 公司創(chuàng)始人兼執(zhí)行副總裁 PaulBanks 擁有應(yīng)用物理學(xué)博士學(xué)位。對他來說，分辨率是最關(guān)鍵的。Banks 先生用通俗易懂的語言解釋了公司理論上具備的技術(shù)優(yōu)勢。他摘下眼鏡說，我是個近視，加州決不允許我不帶眼鏡就開車上路，然而，市場上大多數(shù)激光雷達(dá)的分辨率卻低于加州對人類駕駛員的最低視力要求。
“對我們來說，高分辨率才是最重要的，”Banks直言，“實際上，我們使用的就是你手機里的（圖片）傳感器，其他產(chǎn)品根本不是我們的對手。”
事實上，Banks 的觀點有一定道理，目前博世和三星等投資者對此都很感興趣。TetraVue 公司的固定式Flash 激光雷達(dá)技術(shù)借鑒了數(shù)碼相機領(lǐng)域相對成熟且更為平價的CMOS 和CCD 傳感技術(shù)。TetraVue 激光雷達(dá)采用肉眼不可見的 800 納米波長，發(fā)射速度達(dá)每秒 30 幀（30 fps）。此外，這款激光雷達(dá)還配備了“照明”功能，可以提供像素級的深度信息。
Banks 向我們展示了一位舞蹈演員通過“常規(guī)”激光雷達(dá)和 TetraVue激光雷達(dá)看到的不同數(shù)據(jù)場景，TetraVue 產(chǎn)品所提供的更大視角和深度著實讓人驚嘆。
Banks 表示，“我們的激光雷達(dá)更像是一款視頻攝像頭，無論是從外觀，還是使用感受而言。”事實上，Banks 并沒有夸張，TetraVue 系統(tǒng)每秒可以展示 6000 萬個點，最終的成像效果讓其他競爭對手的畫質(zhì)對比來看就像是一些早期的視頻游戲界面。
現(xiàn)階段，對于這家位于美國加利福尼亞洲的激光雷達(dá)公司而言，最大的挑戰(zhàn)在于探測距離。Banks 表示，按照當(dāng)下的設(shè)計，這款激光雷達(dá)的探測距離在 150 米左右（492 ft），而其他競爭對手的產(chǎn)品，比如Velodyne 最新激光雷達(dá)產(chǎn)品的距離幾乎是TetraVue 的 2 倍。不過，Banks 表示，我們也可以提高探測距離，但“最根本的問題還是成本。”他還表示，公司對這款激光雷達(dá)的定位是大眾應(yīng)用市場，也就是說定價將在大眾可以接受的范圍。
Ouster
內(nèi)戰(zhàn)期間的著名軍事學(xué)家Nathan Bedford 曾提出，“孤注一擲，占得先機”。在如今的科技時代，Ouster 公司可能是對這一名言貫徹地最為徹底的公司了。
Ouster 公司創(chuàng)始人兼 CEOAngus Pacala 曾稱贊稱，我們的市場優(yōu)勢在于：我們已經(jīng)開始向世界各地輸送汽車激光雷達(dá)了。
“我們的產(chǎn)品根本無需額外推銷，”Pacala 夸口說，“我們是市面上最智能，也最輕質(zhì)的 360 度3D 傳感產(chǎn)品”。此外，他還說，Ouster 公司是唯一一家愿意向任何買家，開誠布公地公開技術(shù)價格的公司。
Pacala 此前曾在 Quanergy公司擔(dān)任工程總監(jiān)。他表示，Ouster 的OS-1 是目前市面上分辨率最高的激光雷達(dá)，具有一流的功耗、尺寸和重量指標(biāo)。該系統(tǒng)每秒可測量130 萬點，但功耗還不到15 W。正如TetraVue 一樣，該公司的技術(shù)也植根于相對成本更低也更為成熟的CMOS 技術(shù)（在智能手機和數(shù)碼相機中均有多年應(yīng)用歷史）。
為了將成本控制在合理范圍內(nèi)，OS - 1激光器的工作波長為850 納米，成本在某種程度上則根據(jù)客戶對渠道的不同需求，有“階梯式”差異：其中，成本最高的版本采用了64 個發(fā)射器，可提供每個垂直視場的“切片”，而性能較低的版本僅有16 個通道，因此成本較低。Pacala 表示，該公司預(yù)計將在2018 年底出貨10,000 至20,000 部OS-1 產(chǎn)品。
OS–1 的尺寸與兩個冰球相當(dāng)，高 2.5英寸（63毫米），直徑3.14 英寸（80毫米），重約330 克。這款激光雷達(dá)并不屬于固定式激光雷達(dá)，而是使用了一個可以旋轉(zhuǎn)的發(fā)射器，可提供360° 覆蓋范圍和近32° 的垂直視場，精度約為3 厘米（1.3英寸），但測距范圍相對較短，只有120 米（394英尺）。
隨著OS–2 的問世，Ouster 的測距范圍有所提升，也就是說OS–2 的測距范圍將達(dá)到200米（656英尺），采用64 通道設(shè)計，視場范圍 15.8°。不過，OS–2 的尺寸更大，也更重。Pacala 表示，OS–2 將在今年第三季度起售。
Valeo
2017 年底，法國一級供應(yīng)商法雷奧（Valeo）旗下一款激光雷達(dá)正式登陸奧迪（Audi）A8 轎車，這也被視為激光雷達(dá)首次登陸一款具備SAE 3 級駕駛員協(xié)助功能的量產(chǎn)車型。奧迪A8 的交通堵塞領(lǐng)航系統(tǒng)（Traffic Jam Pilot system）可以在法雷奧 Scala 激光雷達(dá)的協(xié)助下，在發(fā)生交通堵塞的情況下，即車速不超過每小時37 英里（60公里/小時）時，控制車輛的加速、制動和轉(zhuǎn)向功能。
Scala 是法雷奧與 LeddarTech合作開發(fā)的創(chuàng)新固定式激光雷達(dá)，并榮獲2018 年佩斯獎（2018 PACE award）。法雷奧稱，Scala 的水平視場范圍為145°，測距范圍為150 米。LeddarTech 表示，正如許多常見的傳感技術(shù)一樣，Scala 的優(yōu)勢主要在于具有專利的處理方式和算法：“本質(zhì)上來說，我們采用了一整套軟件、算法及專門知識，用于設(shè)計或優(yōu)化各種不同類型的固態(tài)激光雷達(dá)傳感器。”
然而，一些熱切的早期技術(shù)嘗試者可能還無法立刻享受到法雷奧/LeddarTech 激光雷達(dá)陣列。在最初階段，這套系統(tǒng)在許多國家均不支持。奧迪并不愿意在美國及其他法律和監(jiān)管框架尚不夠明確的地區(qū)，貿(mào)然推出這項功能。
5 月，法雷奧宣布了與中國互聯(lián)網(wǎng)巨頭百度 2017 年成立的開放自動駕駛平臺阿波羅公司（Apollo）簽署“戰(zhàn)略合作”。法雷奧在一份新聞稿中表示，公司將為阿波羅項目提供傳感器方面的專業(yè)知識，以及“傳感器清潔系統(tǒng)，及自動駕駛汽車互聯(lián)功能方面的豐富經(jīng)驗”。
Steadily evolving lidar sensor technology will offer significant leaps forward in autonomous capability—once cost is reduced.
The acronym stands for Light Detection and Ranging, and LiDAR first showed its potential in May 1962, when electrical engineers Louis Smullin and Giorgio Fiocca used the 12-inch telescope at MIT’s Lincoln Laboratory to bounce 50-joule laser pulses off the Moon, receiving the reflected light with the lab’s 48-inch telescope. Using time-of-flight calculations, the two researchers were first to measure the distance between Earth and its only natural satellite.
Four decades later, after use in meteorology and the Apollo program, lidar emerged as a vital sensor in robotic-vehicle prototypes competing in the U.S. Department of Defense’s DARPA Challenge. With up to 64 lasers and sensors packed in large, roof-mounted cylinders, the ungainly prototypes resembled spinning soup cans when operating. The sensor units swept a 360° field of view around the vehicle, firing thousands of light pulses. Objects that reflected light within the sweep were identified as a cloud of points, supplementing imaging from the multiple radars and cameras also fitted to guide the driverless vehicles.
By 2007, the annual event’s fourth year, the winning vehicle and five of the six finishers had early lidars from technology pioneer Velodyne Acoustics mounted atop their roofs. Each unit reportedly cost nearly $80,000.
Today’s steadily evolving lidar typically functions in a similar fashion as the early units, albeit with new hardware and software. The proven mechanical-scanning types are in the practical lead for automotive use, experts say, while new solid-state (no moving parts) devices are expected to deliver greater reliability, dependability and a compact form factor—crucial for integrating the unit within the car’s exterior skin. Solid-state types typically have a more limited field-of-view (FOV). Their lower cost, however, offers the possibility to employ multiple sensors to cover a broader area.
Depending on vehicle’s exterior geometries determined by styling, a typical car or truck could need up to eight compact lidar units of narrower acceptance angle—120° on the front and rear, 90° on the side—compared with the big 360° unit on the roof. Range can be reduced; engineers are aiming for 30 meters on the side, 200 meters to the front, 50 meters to the rear, depending on how microwave radar is integrated into the vehicle’s safety suite.
Lidar development is booming, as start-ups and established Tier suppliers race to enable SAE Level 3, 4 and 5 automated-driving capability in future vehicles. The technology is also in big demand for 3D mapping.
Frost & Sullivan forecasts sales of 6 million lidar units in 2025—half of them for use in autonomous vehicles, for a projected $2-billion market. Automotive Engineering counts more than 30 start-ups in the field, along with market leader Velodyne and some Tier 1s including Bosch, Valeo, and Continental. Aptiv and Magna are among top suppliers that are partnering with LiDAR specialists. OEMs continue to acquire and build ties with lidar developers, including BMW (Innoviz via Magna), GM (Strobe), Toyota (Luminar) and Ford (Princeton Lightware and Velodyne).
While some advocates have dubbed lidar “the essential piece of the puzzle for self-driving cars,” it is certainly the most discussed and perhaps controversial sensor technology related to autonomous vehicles.
“Every couple of weeks a new company is touting new lidar technology adaptations,” observed veteran mobility-tech consultant Gerald Conover. “Many of their claims are based on lab-project results, so designs producible in high volume may still be some years away.” Only those that deliver high performance at low cost will survive. Depending on the uptake for autonomous vehicles in SAE Levels 4 and 5, however, “the demand for lidar devices could be significant,” Conover noted.
Cost remains the nagging impediment to the mass deployment of automotive-grade units. Development units still cost $10,000 or more—“not a sustainable number for automotive production,” Conover quipped. “The thing standing in the way of this is the necessary expertise to produce working lidar, which is in the hands of only a few supplier firms.” OEMs eventually expect a steady cost-reduction path to commodity status, similar to those of onboard radar and cameras.
Since late 2017, Quanergy has been producing a 905-nanometer solid-state lidar with a range of 150 meters at 8% reflectivity. An optical phased-array type, it can scan half a million points per second with a spot size of 3.5 cm at 100 meters, the company claims. While single-unit samples are priced in the thousands, Quanergy believes high-volume scale will drive per-unit cost below $300. Velodyne’s lowest-priced 16-laser unit costs $4,000 per unit.
Horses-for-courses tech choices
\While the automotive lidar space is white-hot, not all OEMs see the technology as an imperative. Tesla’s Autopilot system uses camera-based optical recognition, and company boss Elon Musk appears unconvinced that lidar is a game-changer. “Once you solve cameras for vision, autonomy is solved; if you don’t solve vision, it’s not solved,” Musk said during a TED Talk in April 2017. “You can absolutely be superhuman with just cameras,” he added. He obliquely labeled lidar “a crutch.”
Honda North America is “doing development and testing with lidar,” noted Jay Joseph, assistant VP of Product Planning. He said Honda engineers believe lidar is necessary in the short term. “Longer-term, of course, we’d like to see other solutions—probably more dependent on connectivity and shared information. But until that’s reliable, lidar is probably necessary to provide good information to the vehicle so it can make good decisions.”
Assembling and integrating the sensor array into the vehicle is an important role that experienced Tier 1s including Aptiv are playing. “We understand how it works with the vehicle; some of the tech start-ups don’t understand vehicles well,” noted Jada Smith, Aptiv’s VP of advanced technology.
Smith said her company wholly believes in the tech triad of cameras, radars and lidar for vehicle autonomy. Lidar, she said, is “a necessary piece of technology, to handle all use cases, provide redundancy, and to help the vehicle see everything going on around it.” Aptiv is covering multiple technology bases with its investments in Leddartech (flash-lidar), Innoviz (MEM type) and Quanergy (optical phased array).
Choosing lidar types is a horses-for-courses engineering exercise. “What roles do we expect them to play?” Smith asked. “The longer the range, the narrower its FOV—same concept as a camera. Depending on what performance we want, we may choose a flash type for one and MEMS for another. It’s tradeoffs, depending on what we’re trying to accomplish.”
Automotive Engineering recently spent time with a group of lidar innovators and brings the following insights into their backgrounds and technologies.
Innoviz Technologies
Like so many sensor-tech developers, Innoviz is based in Israel and several of its principals have specialized-electronics background with the Israeli Defence Forces. Innoviz was founded three years ago and has about 150 employees globally. The company has garnered more than $80 million in investment funding, including stakes by Aptiv, Magna and Samsung.
The foundation technology is a microelectromechanical systems (MEMS)-based design in which movement of the mirror that projects the scanning lasers comes from a solid-state chip. Critically, Innoviz promotes laser scanning at 905 nm, a long-established wavelength that Aditya Srinivasan, general manager, North America, said allows the company to keep a lid on cost—to the point at which Innoviz can offer its first automotive-grade lidar sensor, InnovizOne, starting in 2019 at a cost in the hundreds of dollars.
There may be some contention about whether the Innoviz design should be defined as solid-state, since the system uses a moving mirror, but, “Rightly or wrongly, we’re calling this ‘solid-state,’” said Srinivasan.
The InnovizOne is “designed for seamless and easy integration into any mass-market vehicle,” the company’s literature describes. The system delivers 120°-horizontal and 25°-vertical FOV for high-definition resolution of 7.5 million pixels/sec; the frame rate is 25 frames/sec. Claimed range is up to 250 m (820 ft). The unit’s footprint measures 50 mm high by 110 mm wide by 100 mm deep (2 x 4.3 x 3.9 in.). Data is managed by a proprietary signal-processing chip that’s been developed with “a partner” that Srinivasan chooses not to name.
The company also claims its system is adept at identifying objects with extremely low reflectivity—a performance aspect that to now has been a challenge for many lidar developers.
In April, Innoviz announced a supply agreement with BMW through Innoviz’ supplier partner Magna. BMW said it intends to offer an SAE Level 3 autonomous ride-hailing service in 2021 and Innoviz-derived lidar apparently will be a key component.
Aptiv and Samsung also are partnered with Innoviz for automotive lidar development and in June the company said it formed a partnership with Chinese automotive Tier 1 HiRain Technologies, which supplies several major Chinese automakers and is integrating the components for its own autonomous-driving platform.
TetraVue
The primary differentiation point for TetraVue’s solid-state “flash” lidar technology is high resolution—as well as reliance on long-proven and relatively-inexpensive sensor technology derived from the digital-camera world. In fact, the company refers to its automotive lidar as a “high-definition 4D camera” that essentially fuses mega-pixel digital video capture with lidar for long-range sensing with pixel-level depth information.
Resolution is everything to TetraVue founder and executive VP Paul Banks, who has a Ph.D in applied physics but explains the company’s presumptive technology advantage in plainspoken terms. Banks removes his eyeglasses, saying the state of California would not certify him to drive without them—yet most competing lidar technologies “see” with less resolution than the state’s minimum vision requirement for humans to legally drive.
“For us, that’s what’s important,” Banks said flatly. “High resolution. We actually use the same (image) sensor that’s in your cell phone. We cheat,” he deadpanned.
His argument is a compelling one that’s interested investors such as Bosch and Samsung. Appropriating the well-developed and extremely cost-driven complementary metal oxide semiconductor (CMOS) and charge-coupled device (CCD) sensing technology of digital cameras, TetraVue’s lidar flashes the environment at up to 30 fps with lasers operating at the invisible-to-the-eye 800-nm wavelength. This “illumination” is merged with the high-resolution video-capture to derive depth information at the pixel level.
Banks’ demonstration borders on amazing, as he shows a data scene of a dancer seen with “conventional” lidar and TetraVue’s lidar; the additional perspective and depth from the TetraVue image is patently startling.
“It looks and feels more like a video camera,” Banks said. And he does not exaggerate—showing up to 60 million points per second, the images from the company’s system make the techy-but-still-scratchy visual representations from competitors seem like the visuals from an ancient video game.
The current downside to Vista, California-based TetraVue’s lidar may be a comparative lack of range. The current design, Banks said, has a range of about 150 m (492 ft)—Velodyne’s latest system, for example, boasts a range twice that far. TetraVue’s range may be improved, he said, “but for us, it all comes down to cost.” He said the company is intent on delivering its advanced technology at a price conducive to mass-market application.
Ouster
Founder and CEO Angus Pacala’s Ouster could be the tech age’s embodiment of the military doctrine of the Civil War’s Nathan Bedford Forrest: “Get there firstest with the mostest.”
Pacala extols one of his company’s market advantages as just that: Ouster is shipping automotive lidars today, he said.
“We let our products do the talking,” he boasted of the “smartest, lightest 360-degree, 3D sensing in the market.” He also said his company is the only one to openly and transparently price its technology for any buyer.
Pacala, formerly the director of engineering at Quanergy, said Ouster’s OS-1 is the highest-resolution lidar currently commercially available, and it has best-in-class power consumption, size and weight. The system measures 1.3 million points per second, yet consumes less than 15W. And like TetraVue, the company’s technology is rooted in comparatively low-cost, highly-developed CMOS technology used for years in ever-advancing smartphones and digital cameras.
To keep costs reasonable, the OS-1 lasers operate on the 850-nm wavelength; cost is “laddered” to some degree according to the customer’s need for channels: the highest-cost versions use 64 emitters to deliver each vertical field-of-view “slice,” while lower-performance requirements can cut cost with just 16 channels. Pacala said the company expected to ship 10,000 to 20,000 units by the end of 2018.
The roughly double-puck-sized OS-1 weighs just 330 g and is 2.5 in (63 mm) tall and 3.14 in (80 mm) in diameter. It is not solid-state: the unit spins to emit over the 360° coverage range and nearly 32° vertical FOV. Its accuracy is about 3 cm (1.3 in)—but range is a comparatively abbreviated 120 m (394 ft).
Range is improved with the coming OS-2, which Ouster indicates will have a 200-m (656 ft) range and 64 channels spaced at 15.8°, although the unit is correspondingly larger and heavier. Pacala said the OS-2 would be available in the third quarter of this year.
Valeo
French Tier 1 supplier Valeo late in 2017 achieved the distinction of supplying what is believed to be the first lidar sensing system to be deployed on a series-production vehicle, Audi’s A8 sedan, widely described as using SAE Level 3 driver assistance. The A8’s Traffic Jam Pilot system controls the A8’s acceleration, braking and steering at speeds up to 37 mph (60 km/h), using the company’s Scala lidar.
Scala, a solid-state design developed in cooperation with LeddarTech, won a 2018 PACE award for supplier innovation. Valeo said Scala has a 145° horizontal field of view and range of 150 m. As is typical for many sensing technologies, LeddarTech says its advances are largely in proprietary processing and algorithms: “essentially an ensemble of software, algorithms and know-hows that are used to design or optimize various types of solid-state lidar sensors,” according to company literature.
Some anxious early-adopters won’t find for Traffic Jam Pilot and the Valeo/LeddarTech lidar array just yet, however; the system initially is not available in many countries. Audi was reluctant to introduce the technology in the U.S. and other markets that do not have clearer legal and regulatory frameworks to address conditional autonomy.
In May, Valeo announced a “strategic cooperation” with Apollo, the open autonomous driving platform created by China’s Baidu in 2017. The company said in a release it will contribute to the Apollo project its expertise in sensors, not to mention its “skills in sensor cleaning systems and connectivity between autonomous vehicles.”
Author: Lindsay Brooke & Bill Visnic
Source: SAE Automotive Engineering Magazine
通用 Cruise 自動駕駛汽車集成了大量激光雷達(dá)，用于自動駕駛汽車的公共道路測試。 (圖像: GM)
為了將傳感器完全整合至汽車中，縮小激光雷達(dá)的尺寸將至關(guān)重要。麥格納的概念車（如圖所示）已經(jīng)邁出了整合的第一步。（圖片：麥格納）
Aptiv在CES17上使用的BMW 5系自動演示車的傳感器布局。 綜合陣列包括四個短程激光雷達(dá)，五個遠(yuǎn)程激光雷達(dá)，六個掃描雷達(dá)（ESR），四個短程雷達(dá)（SRR），一個三焦攝像頭，一個交通燈攝像頭，兩個GPS天線，一個DSRC模塊和兩個 計算機軟件堆棧用于冗余。(圖像：Aptiv)
Aptive 公司的固定式激光雷達(dá)（圖片來源：Aptiv）。