Li Gen's
report from the qubit of the concave temple | public account QbitAI
Just 4 hours ago, Wu Enda "showed off his wife" again.
Dressed in formal attire, wearing a cowboy hat, embracing his beautiful wife Carol Reiley, a familiar face and a familiar smile, revealing that he is in Texas, and spoilers to announce a big news.
Could it be... to torture dogs in Texas?
Now that the mystery is revealed, it is really big news, and it is really related to this pair of AI dependents:
Drive.ai, the self-driving company of which Andrew Ng is a director and Carol Reiley is a co-founder, is about to provide a driverless taxi service in Frisco, Texas.
Starting from July this year, Drive.ai will be deployed in cooperation with the local traffic management association for a period of 6 months, providing driverless (free) taxi services for more than 10,000 people in a specific area. You can get a driverless taxi.
There is no doubt that this is the first time in Texas history, and it is only behind Google's Waymo in the United States. For this unmanned car project that is related to Wu Enda, his wife and students, it is also worth noting.
Ng Enda excitedly declared that this means that "the driverless car has finally arrived", and it also shows the success of the three elements and three steps of the autonomous driving landing route.
But can this success be replicated?
Qubits travel thousands of miles across the ocean. Not only did they get an explanation from Wu Enda, but they also learned more exclusive details from Wang Tao, the co-founder of Drive.ai.
△ Drive.ai co-founders Song Wei, Wang Tao, Wu Enda
Hardware scheme updated
Different from all the previously exposed schemes, the car that Drive.ai put into trial operation this time has not only changed from a Lincoln MKZ to a Nissan NV200, but also the sensor scheme has been updated: equipped with 4 lidars, 10 camera, and 2 mmWave radars.
Compared with the previous solution of 9 high-definition cameras, 2 millimeter-wave radars, and 6 Velodyne Puck lidars, Wang Tao admitted that the cost has been greatly reduced, and it should be noted that the 4 lidars in the current solution are not all from Velodyne.
But Wang Tao did not explain more about why he "changed the car". He only said that the Drive.ai self-driving car conversion solution is fast and efficient, and they have recently completed a new model that can be used for passenger and cargo.
用户体验有创新
除了硬件方案不同，此次Drive.ai投入运营的无人车设计也相当醒目。
车身被喷漆成显眼的橙，还搭载了4块LCD显示屏。
王弢解释说，这是Drive.ai在落地自动驾驶中，对于“以人为本”理念的践行，醒目的车身颜色可以让行人更好认出无人车，而LCD则能向其他司机、行人有“交流”。
他们希望此举能成为无人车行业标准，就像当初转向灯等发明时一样。
当然，这算是一种用户体验上的创新，如果把无人车当成一款产品来看，此前更多被重视的是车内乘客的安全和舒适，而忽视了车外“用户”的交互。
吴恩达则补充说，更明显的识别度，可以让人们更快认出无人车，不仅利于大众接受，也能在运营中更安全。
同时，环绕车周的4块LCD显示屏，则会实现与外界的实时交互，比如无人车让路行人，LCD屏幕上就会显示：我在等您过马路。
运营方案有冗余
上述创新之外，Drive.ai还在试运营方案中设置了“冗余”。
实际上，此次完全参与Drive.ai德州试运营的无人车数量并不多，通常只有3辆，高峰需求期可能会增至4辆，也不会7*24小时运营，而只会在固定时间段（上下班、外出就餐）等供当地上班族免费代步。
运营路线上也会站到站，在HALL Park和The Star周围提供固定的上下客地点，之后还会扩展到弗里斯克车站，主要目的是通过无人车实现商业园区和生活服务区的连接，免除上班族开车、停车等困扰。
不过王弢也解释，不同于“公交路线”，这种站对站路线的中间，乘客也能指定系统要求“上下车”，只不过接驳时会倾向于建议站点。
更关键的运营“冗余”方案体现在部署步骤上，吴恩达对此进行了详细解释，分三阶段：
第一阶段，有安全员，可以在紧急情况下接管无人车，同时部署远程操控的tele-choice技术，二者具体如何分工，吴恩达举了个例子。
“我们的汽车想在交叉路口执行一项复杂的任务。如果它认为应该由人类来进一步确认安全性，就会首先停下车，然后让远程操作员来处理。假以时日，我们的深度学习系统就会学习这些案例，提升自动化程度。与直接让操作员控制汽车的“远程驾驶”不同，我们的tele-choice系统可以接受网络延迟和暂时的网络宕机，甚至考虑了一些极端情况，例如自动把延迟100毫秒的过期数据或请求认定为无效。”
第二阶段，去掉安全员，但配备监护员，同时tele-choice操作员依然是标配。但不同于安全员，监护员会坐在乘客座位，可以帮助乘客“监督”无人车运行情况，也不承担实时接管汽车控制权的职责。
最后一个阶段，车内只剩乘客，Drive.ai则通过tele-choice操作员来提供远程协助。而且一个tele-choice操作员可以监督多辆汽车，这也意味着可以更大规模地部署无人车。
不过吴恩达和王弢都未透露“大规模部署”可以多快，但量子位获悉，此次试运营的Drive.ai无人车，从加州来到德州弗里斯克之后的第5天，便已经完全适应“新路况”了。
Drive.ai对于自己深度学习为核心的自动驾驶系统，满怀信心。
但也不是已经完全没有了挑战。
吴恩达说，目前无人驾驶系统在一些特定的复杂场景下，还无法达到人类司机应有的判断力，比如理解养路工人的手势、交警的指令等，这些都需要更大量的数据，以及更多的反馈来实现进一步训练。
不过他也坚信，Drive.ai已经走出了一条落地可复制的无人驾驶之路——在限定性场景中解决挑战的前提下。
为此，他专门总结了无人驾驶出租车运营的三大关键因素：
一是技术：需要行业领先的人工智能和深度学习技术；
二是合作关系：需要与公共和私有合作伙伴共同部署；
最后也最关键的是安全：需要以人为本的安全体验及效果。
当然，吴恩达也重点夸赞了Drive.ai的“个人奋斗”，他说这支团队拥有很多深度学习专业人才，他们使用现代人工智能技术从头设计了无人驾驶基础架构。
Additionally, by deploying comprehensive software for autonomous driving—cognitive, motion planning, mapping, localization, fleet management software, mobile apps, communications, tele-choice remote assistance systems, and more—the Drive.ai team was able to move quickly, and resolve dependencies between systems.
Qubit also learned that Drive.ai has now expanded to 120 people, focusing on engineering research and development, based in Mountain View, California. Before the third quarter of this year, their team size will increase to about 170 people. At present, both software and hardware engineers are is where the demand is.
It is worth noting that there is also an indispensable Wu Enda in this team. The role he is playing is not just as simple as a teacher and a director.
Perhaps the dream is also driving, working in the future of unmanned vehicles that Google and Baidu have not realized, Wu Enda wants to realize it in Drive.ai.
As shown in this video:
-End- _ _
event registration
join the community
The qubit AI community 16 groups have begun to recruit. Students who are interested in AI are welcome to add a small assistant WeChat qbitbot6 to join the group;
In addition, qubit professional sub-groups ( autonomous driving, CV, NLP, machine learning , etc.) are recruiting for engineers and researchers who are engaged in related fields.
To join the group, please add a small assistant WeChat account qbitbot6, and be sure to note the keywords of the corresponding group~ After passing the review, we will invite you to join the group. (Professional group review is stricter, please understand)
Sincere recruitment
Qubit is recruiting editors/reporters to work in Zhongguancun, Beijing. We look forward to having talented and enthusiastic students to join us! For relevant details, please reply to the word "recruitment" on the QbitAI dialogue interface.
Qubit QbitAI · Toutiao signed author
վ'ᴗ' ի Track new developments in AI technology and products