Home > News content

What are the layouts of Microsoft that did not test cars in autopilot?

via:雷锋网     time:2018/4/17 7:01:34     readed:372


Microsoft China official store

What did Microsoft do?

Microsoft's most direct automatic driving program is not officially established. In January 2017, several employees within Microsoft's internal deep learning and robotics community got together because of an automated driving online course. After hitting them, they formed a team called the Road Runner automated driving project.

However, this team is more like an interest group. They don't have the funds to buy a car for experiments. In the beginning, they used a remote control car and added the automatic driving function to the remote control car. However, the remote control car is too small to stand up to the real road toss. In the end, they decided to put an end to this kind of poverty-stricken test and use a simulated environment to test their autopilot algorithms.

The simulator they use is Microsoft's open source platform AirSim. AirSim was initially used only for drone research, but after working with Road Runner, in November 2017, AirSim announced a platform upgrade that covered the car simulator.

AirSim has a very detailed 3D urban environment that includes a variety of dynamic scenarios such as traffic lights, parking lots, lakes, construction sites, and more. Users can test the system in different environments in the community, such as downtown, suburban areas, vegetation coverage and industrial areas. Developers can use the Microsoft Cognitive Toolkit (CNTK, Microsoft Cognitive Toolkit) with AirSim help for in-depth learning.

Road Rnner is characterized by making virtual collisions that allow AI to constantly collide and learn to drive safely. Their system has acquired a large amount of data that is difficult to test in reality.

However, because it is an experimental project, Road Runner should not become a commercial project, but Microsoft may promote their systems in the future, making it an important player in automotive software.

Playing an auxiliary route

In 2017, Microsoft’s head of business development, Peggy Johnson, publicly stated that he will not develop an autonomous vehicle, but will continue to develop automotive software. “We will not develop our own self-driving vehicles, but we hope to help autopilot and assist driving."This is Microsoft's position, but on this line, will Microsoft be able to give Microsoft a share of the autopilot?

Let us first look at the data, according to Statista's data, Microsoft Windows car operating system in 2017 has a market share of 15.8%, ranking third in the world, but the future has a downward trend, and may be eroded by QNX and Linux. This trend is very unfavorable for Microsoft's "assisted" route. In addition, the study of deep learning is mainly built on the Linux operating system, and Microsoft's CNTK deep learning framework is not as popular as TensorFlow, torch, and Chainer.

In addition to the on-board operating system, Microsoft also has a car networking platform called Connected Vehicle that provides services including communications, car entertainment, real-time traffic conditions, and advanced navigation. Toyota and Ola, a taxi company in India, have already reached cooperation with Microsoft in car networking. But in addition, Microsoft did not find more partners.

In fact, despite Microsoft's efforts in software, due to historical reasons, Microsoft has a shortage of talent in this area. When deep learning is not as hot as it is today, the Microsoft Research Institute is a company where talents compete to enter. Therefore, Microsoft recruits more people who have rich working experience and fewer people who study deep learning (because at the time The study of learning is mainly led by some doctoral students.

Perhaps aware of the lack of internal driving force, in February 2018, Microsoft and Bosch jointly announced the accelerator OpenADx, which serves the development of autonomous driving and focuses on auto-driving software development tools. However, whether the accelerator can hatch out influential projects still requires time verification.

Cloud, or cloud

Compared to the pioneering technology of automatic driving, Microsoft may be more concerned about the back-office support. In other words, Microsoft prefers autopilot companies to use theircloud service.

Microsoft recently cooperated with ACM (American Centor for Mobility) in Michigan, USA, to provide cloud and network security services for this base. The base's CEO, John Madox, said that Microsoft's Azure cloud service can handle terabytes of information for companies that test at ACM and ensure a secure network environment.

Last year, Microsoft became a partner of the Baidu Apollo program and will provide Microsoft SmartCloud Azure services to Baidu's Apollo open platform in regions other than China. Due to the accelerating development of domestic auto-driving speed, various companies also have a number of "sea-going" plans. With more than 90 partners, Apollo will be able to expand its market share of cloud services in the future.

According to Microsoft’s fiscal second quarter financial report, Microsoft’s second-quarter smart cloud revenue was US$7.795 billion, an increase of 15% from US$6.758 billion in the same period of last year, and operating profit was US$2.832 billion. Cloud service is a large profit-making business of Microsoft. Instead of competing with Waymo and traditional car companies, Microsoft knows how to consolidate its own market.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments