許多自動駕駛技術發展計劃都要求在車輛上安裝固態激光雷達傳感器，但由于當前原型車所采用的激光雷達模塊都是由活動部件組成的機械系統，因此這一需求引發了人們對激光雷達的巨大興趣，OEM和供應商競相投資開發非機械組裝的激光雷達技術。
一些小型公司已經研發出了固態激光雷達技術，但還無法實現在汽車上的應用，并且其中一些技術已經在過去18個月內被大型汽車公司所收購：福特投資了Velodyne，采埃孚收購了Ibeo 40%的股份，大陸集團和Analog Devices則分別并購了Advanced Scientific Concepts和Vescent Photonics。
激光雷達通過發射激光來測量物體間的距離，該功能與雷達類似，但激光可以支持系統在夜晚和雨雪天氣下提供高分辨率的圖像，這激發了市場對固態激光雷達技術的興趣。
“高分辨率‘快閃光’激光雷達的功能適用于所有照明和天氣條件，因此成為了自動駕駛的一項必要技術。”Continental北美高級輔助駕駛系統的客戶項目主管Dean McConnell介紹道，“我們以30赫茲的速度捕捉圖像，然后以每秒30次的頻度構建3D點簇。”
該技術還有助于安全系統對值得注意的對象進行歸零校正，這對確定一個對象是否會對駕駛構成威脅至關重要。
“激光雷達的作用就像人類的眼睛，視野寬廣，可快速進行掃描，然后專注于感興趣的對象，”ADI的汽車安全總經理Chris Jacobs說道。
激光雷達供應商正在加速開發緊湊型固態模塊。由于目前自主駕駛研究人員采用的大型機械裝置體積過大，生產成本太高，研究人員正在努力縮小模塊尺寸，并將距離和視野完美結合。
“我們的固態塻塊尺寸是9×6×6厘米，大約相當于兩盒撲克牌的大小，”Quanergy的首席執行官Louay Eldada介紹道。“固態塻塊目前有120度的視野范圍，所以三個可以構成360度視野覆蓋。三個固態塻塊的位置安排，可以是左前方和右前方各一個，第三個位于后方中間位置，或者每個角落各一個。”
車輛與物體間的距離是關鍵的安全參數，而縮小視野范圍有助于提高該參數的測量精度。開發人員正在努力實現與相機和雷達相同的測距能力，目標約為200米（656英尺）。要獲得理想的測距能力，需要權衡不同方面，位置點是幫助確定視野覆蓋范圍的關鍵參數；側視模塊不需要與前視模塊擁有相同的測距能力，所以其視野范圍可以更寬。
“我們已經實現了在15度視野內進行70米（230英尺）范圍內的精確測定，但這明顯不夠，”采埃孚主動和被動安全部門產品規劃負責人Aaron Jefferson說。“視野寬度至少需要達到50或60度。成本下降后，可以將模塊與尾燈和前大燈合并。”
激光雷達可以彌補攝像機和雷達的不足，提供的信息在與其他傳感器信息“融合”后，能夠生成可靠的汽車環境圖像。所有這些傳感器會產生大量的數據，所以通信和數據管理將成為整體設計中的一個重要因素。
“3D激光雷達傳感器將生成大量數據，但與雷達和攝像機類似的是，軟件技術可以幫助減少數據量，剔除無用或不重要的數據，并從關注的數據中提取細節，”Jefferson說道。“此外，用于數據過濾、數據分組/分類、對象識別的技術也決定了需要處理的數據量，對于數據量管理而言，這才是真正的問題。”
市場充滿興趣，但仍在觀望
盡管出現了很多新發展，但預計市場在未來一段時間內不會有太多活動。許多工程師都表示，激光雷達技術可以在等待自動駕駛車輛設計更加成熟的同時慢慢發展。目前，系統設計人員可以在等待下一代模塊推出的時間內，采用機械組件創建原型。
“固態激光雷達將于今年晚些時候投入生產，但測試和軟件開并不需要固態產品，”Eldada說道。“雖然我們計劃固態模塊在9月出貨，但要到一年以后我們才能準備好車用級別的零件。”
激光雷達車輛的大規模推廣和自動駕駛汽車一樣，前景尚未明朗。在主流OEM開始訂購激光雷達傳感器之前，企業車隊項目可能會擴大規模，并帶來市場機會，優步（Uber）在匹茲堡正在進行的自動駕駛測試項目就是一個例子。
“我們正在研究2021年以前的批量生產情況，但不同領域的實現時間可能會有所差異，”McConnell說道。“一些車隊服務公司正積極地推動在已進行地圖測繪的區域應用自動駕駛技術。”
許多開發人員認為，激光雷達投入使用后并不會大量取代其他傳感器，因為我們仍需要一系列技術，以支持在各種天氣條件下實現自動駕駛所需的能力和冗余水平。
“我們并不認為3D激光雷達會替代現有傳感器，而是將其看成是一種能夠提供高分辨率感知系統、幫助實現SAE4級自動駕駛的創新技術，”Jefferson說道，“3D固態激光雷達、攝像機、雷達、超聲波傳感和其他技術將繼續發揮作用，這些技術是在駕駛過程中360°實時感知車輛周圍狀況所必須的。
然而，這并不是一個普遍公認的結論。
“超聲波技術將會消失，”Eldada反駁道。“攝像機可以支持色彩辨認需求，如查看交通信號燈。將激光雷達和攝像機融合，對數據進行‘著色’，可以使數據更有價值。雷達可以支持冗余需求，而進行轉向或剎車決策還需要其他傳感器信息的支持。”
Many autonomous-driving development plans call for deploying a handful of solid-state Lidar sensors on each vehicle, but the Lidar modules used for today’s prototype vehicles all are mechanical systems with moving parts. That’s prompted huge interest in Lidar, with OEMs and suppliers racing to invest in non-mechanical technologies.
Several small companies have developed solid-state Lidar technologies that aren’t ready for automotive applications—and some of those have been gobbled up by major automotive companies past 18 months. Ford made a large investment in Velodyne, while ZF bought a 40% stake in Ibeo. Continental acquired Advanced Scientific Concepts. Analog Devices Inc. (ADI) acquired Vescent Photonics Inc.
The interest stems from Lidar’s advanced use of emitted laser light to measure the distance of objects, functioning much like radar. The laser lets the system provide high resolution imagery at night and in rain or snow.
“High-resolution 'flash' Lidar is a necessary technology for autonomous driving because its capabilities are available in all lighting and weather conditions,” said Dean McConnell, Director of Customer Programs, Advanced Driver Assistance Systems, at Continental North America. “We’re capturing images at 30 Hz, constructing 3D point clusters thirty times per second.”
The technology also helps safety systems zero-in on objects of interest. That’s important to determine whether an object is a threat to driving.
“Lidar acts more like the human eye: it views a broad scene, doing a quick scan, then if it sees something interesting, it can focus in on that,” said Chris Jacobs, General Manager of Automotive Safety for ADI.
Lidar providers currently are racing to develop compact solid-state modules because the large mechanical pucks now used by autonomous-driving researchers are too bulky and costly to go into production vehicles. Researchers are striving to shrink sizes and come up with a good combination of distance and field of view.
“Our solid-state box measures 9 x 6 x 6 cm, about the size of two decks of cards,” said Louay Eldada, Quanergy's CEO. “Currently, it has a 120-degree field of view, so with three you have 360 degree coverage. There will always be two in the front, on the right and left sides, and one in the back middle or one on each corner.”
Determining the vehicle's distance to objects, a key parameter for safety, can be increased by narrowing the field of view. Developers are trying to achieve the same distance levels as cameras and radar, with a goal of around 200 m (656 ft). To achieve desirable distance performance, several tradeoffs are being considered. Location points are key parameters that help determine field-of-view coverage; modules looking to sides, for example, won’t need the same range capability as forward-facing units, so their field of view can be wider.
“We’ve demonstrated 70 meters (230 ft) with a 15-degree field of view, which is clearly not sufficient,” said Aaron Jefferson, Director of Product Planning for ZF’s Active and Passive Safety Division. “It needs to go up to 50 or 60 degrees to start. When the cost gets down, it’s conceivable that they could be integrated into taillights and headlights.”
Lidar will complement cameras and radar, providing information that typically will be "fused" with that from other sensors to create a reliable image of vehicle surroundings. All these sensors generate a huge amount of data, making communications and data management an important factor in overall designs.
“3D Lidar sensing will create a significant amount of data, but similar to radar and camera, there are software techniques to help minimize the amount of data, eliminate useless or unimportant data and extract the detail from the data of concern,” Jefferson said. “Furthermore, the techniques used to filter data, group/cluster data, identify objects, etc. also determine the amount of data that needs to be processed, which is the real concern in terms of managing data volume.”
Curiously, no real hurry
Though there’s plenty of development, the market isn’t expected to see much activity for some time. Many engineers say Lidar can develop slowly while waiting for autonomous vehicle designs to solidify. For now, system designers can create prototypes using mechanical components while they wait for next-generation modules.
“Solid-state Lidar will be in production later this year, but for pilots and software development, you don’t need solid-state,” Eldada said. “Though we plan to ship solid state products in Sept., we won’t have automotive-grade parts ready until a year later.”
The rollout of Lidar-equipped vehicles is as murky as the emergence of autonomous cars. Corporate fleet programs like Uber’s autonomous current tests in Pittsburgh may expand into market opportunities before mainstream OEMs start ordering Lidar sensors.
“We’re looking at series production in the 2021 timeframe, but it may happen faster in different segments,” McConnell said. “Some fleet-service companies are aggressive about getting vehicles out with automated driving in a geomapped area.”
Once Lidar is in use, many developers don’t expect it to displace many other sensors. A range of technologies is needed to provide the capability and redundancy needed to drive autonomously in all weather conditions.
“We do not see 3D Lidar as a sensor replacement, but rather as an innovation that can enables the high resolution sensing needed to realize SAE Level 4-plus automated driving,” Jefferson said. “3D solid-state Lidar, camera, radar, ultrasonic sensing and other technologies will continue to play a role—a combination of these will be necessary to properly sense the vehicle environment in 360 degrees, in real time.”
That’s not a universal conclusion, however.
“Ultrasonics will go away,” Eldada countered. “Video is needed for color, things like seeing traffic lights. Fusing Lidar and cameras 'colorizes' our data so it’s more valuable. Radar is needed for redundancy; you need another sensor before deciding to steer or hit the brakes.”
Author: Terry Costlow
Source: SAE Automotive Engineering Magazine
圖為高分辨率的Quanergy激光雷達圖像：中間部分為街道，藍色部分為樹木。
大陸集團的固態激光雷達模塊（如圖所示）和其他類似模塊，預計將為大多數自動駕駛車輛所采用（圖像：Continental）。
