第一辆后装无人驾驶车辆加州上路测试 First Aftermarket Autonomous Cars Hit the Road in California

去年,在知乎上有网友问我:“Cruise 无人驾驶系统会对 Google 无人驾驶汽车产生冲击吗?” 这是我第一次了解到Cruise这个初创公司。借着回答相关问题,也对这个团队以及他们当时在研发的产品RP-1进行了了解。我和来自德国大陆集团的研究员Kevin Zhong针对网友的提问,给出了一致的结论:RP-1是一个不错的产品,但10000美元的价格和仅在高速路段的应用决定了竞争力有限。但Cruise这个团队聪明在于选择了后装市场,这就避免了和博世(Bosch)、德尔福(Delphi)等一线供应商产生直接的竞争。

IEEE Spectrum最近公布了Cruise的最新消息:Cruise宣布其两辆无人驾驶车获得了加州交管局的上路测试执照。两辆测试车所使用的即是Cruise正在不断优化的后装设备RP-1,不过售价依然是早期公布的10000美元,并没有在成本和价格控制上有所调整。

RP-1面向高速路环境,实现在高速路段的自主驾驶,包括车道保持和车辆检测及避障功能。RP-1安装在车顶,内置了毫米波雷达、立体视觉、GPS和惯性测量单元;分别有控制器对刹车、油门和方向盘实现控制,处理单元则安装在车辆后备箱;而激活RP-1的功能仅需要按中控面板的对应按钮即可。

Cruise除了正在不断优化RP-1的算法,以让系统更佳地适应光照阴影变化之外,也希望能将系统扩展到城区环境下(自然这是我们课题组一直在研究和解决的问题=.=,“面向城区环境下的无人驾驶技术集成验证平台”)。但目前RP-1所有的测试均在奥迪A4和S4s上进行,Cruise尚未公布RP-1何时能够适配更多车型,尽管Cruise一直宣称RP-1可以安装在“几乎所有车辆上”。

Cruiseblog-1435759755617

有关去年我在知乎上对Cruise的分析可以点击此链接,或继续阅读,我已全文重新发布于此。


 

Google’s driverless car project has been getting plenty of attention for the kinds of changes it could bring to transportation, but buying an entirely new, self-driving car is out of most people’s budgets and many years away. Cruise Automation is a San Francisco-based startup that thinks it can get the technology to market sooner with something far more simple: a $10,000 accessory you can strap to the roof of your car and plug into your footwell.
via: Driverless Cars For $10,000? This Startup Is Challenging Google With A Simple Sensor
以10000美元的价格安装一套无人驾驶系统会对Google 的无人驾驶汽车产生什么影响吗?

崔迪潇无人驾驶汽车研究员 半吊子摇滚混子乐手

写在前面:

今早查邮箱时 收到了刘尧老师几个不同的关于Cruise的问题 为节约时间 统一在这个问题下做简短回复 借通过对比google现行技术和Cruise的方案 给感兴趣的朋友 也使自己进行一些思考~其余相关问题的内容 也争取在这里涉及到 烦请刘老师协助编辑下~关于Cruise的介绍:
1)总部位于San Francisco的创业公司 去年刚成立 目前刚7个月
2)其创始人为Kyle Vogt 毕业于MIT 其LinkedIn可以查看Kyle Vogt | LinkedIn

也是一个在创业从业方面经验丰富的主 有过多个公司(或许下一个钢铁侠)
Scoialcam 6000万美元卖给了Autodesk
视频研究相关公司 Twitch Google现在考虑以10亿美元收购
Cruise中的很多视频图像处理技术来自于他及团队这些年在视频领域创业的积累
3)从业人员数目多少 不知

关于Cruise发布的无人驾驶系统RP-1的介绍

1)10000美元 需要在车顶安装视觉传感 驾驶室内安装执行部件 (可能还需安装雷达 相关文献提到了他们也在尝试进行障碍物信息监测)处理单元 及在中控面板安装一键切换(见我前面的相关回答)

2)目前只适用于奥迪2012年及更新的A4或S4车型 成本约为奥迪A4成本的29% 公司正计划以后适配更多的不同的车型
3)初期只提供50套预售版本 且全在加州 在正式交付之前 目前还有大约6个月至9个月的测试要做 包括第三方测试
4)所有初期的测试版本只在加州Bay Area的101高速路和280高速路上进行 每次测试的无人驾驶时间(驾驶员无方向盘干预)为30秒到10分钟
5)该系统被称为Highway AutoPilot 即现阶段系统仅在高速路段上使用 在雨天 雾天 夜晚无法工作(部分环境google也无法解决)
6)Cruise在San Francisco的Alameda Point海军基地做了一次基于奥迪A4的公开演示 演示内容显示车道保持较为稳定 但测试中没有行人车辆等障碍物 参与测试的两辆A4也无法避让对方 即尚不具备障碍物检测功能 所行驶路段车道较普通3.5m-3.75m的标准车道宽约1m 即接近5m道宽

7)测试路段需要提前人工驾驶 构建地图信息 并不是纯粹的依靠实时感知 Cruise也在计划进行大范围的地图描绘(同时采集图像 GPS IMU信息) 同时借助于车辆移动平台进行地图数据的共享和高频率的更新 前期任务是覆盖住要进行测试的Bay Area

与google无人驾驶车的主要区别

1)Cruise定位于人类驾驶行为的部分取代 而google的self-driving car是基本具备完全自主驾驶功能的
2)无法实现点到点的无人自主驾驶 仅仅是在google的自主驾驶功能和市面上的ACC之间的产品 具有车道保持 障碍物检测及减速(后续)等功能 无法完全解放驾驶员
3)相对于google高成本激光为主的技术路线 Cruise坚持走低成本产品(对我这等穷苦百姓依然是天价)

综上:

1)RP-1依然是一个处在研发中的半成品 距离实际使用依然有大量的工作要做 现在谈不上称之为革命性的产品
2)类似低成本的系统已经有很多原型 我所在单位的系统 从相关文字表现对比来说 要优于目前的RP-1
3)低成本的思路是正确也是趋势 同时不以替代人的驾驶功能为出发点 也值得肯定 也是相较于全自主无人驾驶来说 更易于实现的 无论从技术角度 还是从产品商业化角度
4)RP-1的定位决定了产品的应用范围有限 也许正式产品上线后 在结构化场景如高速路段能够搅一搅市场的局 但整体市场的长远竞争还需要在本身的感知控制方面下工夫

题外话:

不能以无人驾驶的潜在风险和责任归属作为借口 而在产品本身的性能上退而求其次 没有人会讨厌一台聪明的电脑 我们需要一套100分的智能系统 而只是需要我们选择在哪些地方在哪些场所让它发挥到什么程度 用牛刀来杀鸡虽然浪费 但我们至少可以百分之百肯定 鸡能够被杀死


英文报道见如下:

The number of self-driving cars on California’s public roads has nearly doubled in the last month. As of mid-June, 77 vehicles from eight manufacturers have been issued autonomous testing permits by the California Department of Motor Vehicles (DMV). That’s up from 48 in May, and includes the first two vehicles fitted with self-driving technology that its maker claims “can be installed on almost any car.”

Cruise Automation is a Silicon Valley start-up developing a US $10,000 system called the RP-1 Highway Autopilot. As the name suggests, the RP-1 is designed to take over driving on freeways, keeping the car within lanes, and sensing and avoiding other vehicles. Its most obvious component is a pod mounted above the windshield, containing millimeter-wave radar, stereo video cameras, GPS and inertial sensors—although not the expensive laser-ranging lidar systems favored by Google.

Inside the vehicle, actuators control the steering, gas and brakes, while the brains of the whole thing are a computer housed in the trunk. To activate the RP-1, a human driver simply maneuvers into a lane on the highway and pushes a button on the dashboard. Although Cruise only received its DMV permits for two cars in June, the company’s job ads say that it has already “logged thousands of autonomous miles on California highways.”

An aftermarket robot chauffeur could accelerate the arrival of self-driving vehicles. Although regulations covering autonomous technology for public use have yet to be finalized in California (or elsewhere), not having to crash test or get an entire vehicle approved is bound to save time.

And even at $10,000, fitting an RP-1 will probably work out cheaper at first than buying a new production car with the technology built in. (Tesla is promising a similar highway autopilot for its $70,000 Model S sedan this summer.) Cruise is doing its best to keep the price down further, using commodity hardware, low-cost sensors and affordable but high-powered graphics processing units (GPUs) to process images of the road ahead.

But like other companies working on autonomous vehicles, Cruise is finding that the self-driving devil is in the details. An advertisement for computer vision engineers note that the company is working on “making our vision systems more robust to changes in lighting conditions or shadows” and developing “new ways to use image features or tracking to improve mapping and localization.”

Re/code reported last year that Cruise intended to start installing its first 50 Highway Autopilots, for late-model Audi A4 and S4s, in “early 2015.” Cruise would not comment on this story, or say when the first installations would actually begin. There is also no word on when the RP-1 might be available for other models or makes of car.

Ultimately, Cruise would like to offer a kit that could also handle city streets, where we do most of our driving. But to do that, it needs its brace of experimental cars to provide far more real-world experience. “Over time, we’ll use the data we collect to make smarter, more automated products,” says Cruise’s website.

The start-up has some catching up to do. Google now has 48 self-driving vehicles on the road in California that have driven about three million kilometers in total. Cruise had better put its robotic foot on the gas or it risks being left in the slow lane.

English source is from IEEE Spectrum, by Mark Harris.

关于dixiaocui

无人驾驶汽车研究员 半吊子摇滚混子乐手 Researcher on Intelligent Vehicle Guitarist/Singer/Rocker
此条目发表在Autonomous Driving分类目录,贴了, , , 标签。将固定链接加入收藏夹。

发表评论

电子邮件地址不会被公开。

此站点使用Akismet来减少垃圾评论。了解我们如何处理您的评论数据