Skip to content
Microsoft Industry Blogs

 The advent of autonomous driving is shaping up to be one of the most impactful and disruptive innovations to hit the automotive industry. While the opportunity for those ready to capitalize is immense, so too is the risk for those that fail to adapt. Automotive companies need to put it in the work to get there first to seize upon a once in a generation opportunity to transform transportation into mobility-as-a-service.

In a recent survey conducted and published by KPMG, automotive executives weighed in on the future of the automotive industry, and discussed how the emergence of autonomous driving is poised to profoundly transform the criteria used by customers in selecting an automobile—before consumer automobile ownership falls off all together. As autonomous driving technology becomes the norm, 2 out of 3 executives surveyed at least partially agree that the current criteria used by consumers will soon be rendered either irrelevant or commoditized. The entire automotive industry is poised for transformation as consumer car ownership becomes a thing of the past—in a shift to mobility-as-a-service.

As automotive companies look to the future, it is becoming more and more apparent that autonomous driving will transform the way consumers think about cars. Soon, customers will care less about what they can do with a vehicle and more about what they can do in a vehicle they utilize as a service, like Uber or Lyft. The popular cars of the future will be equipped to facilitate productivity while providing cutting edge in-vehicle infotainment services leaning on a new generation of IoT products, and they will be owned in large fleets by companies instead of individually by consumers. Mobility-as-a-service represents a new opportunity for monetization, leading to new and diverse revenue opportunities for the companies that can grab a share of the market. With so much opportunity on the line, automotive companies first need to get there—and to get there first—to position themselves to capitalize.

 

 

The long road ahead—leveraging deep learning to provide safety and peace of mind to a generation of passengers

While the technology behind autonomous vehicles has advanced by leaps and bounds as companies continue to invest, safety and trust remain the two obstacles to achieve NHTSA level 4 classification. To achieve this level of safety, massive amounts of training and testing is required.

In a recent study by non-profit research organization Randcorp entitled, Driving to Safety: How Many Miles of Driving Would It Take to Demonstrate Autonomous Vehicle Reliability, researchers concluded that 10 billion miles of testing is required to reach the NHTSA level 4 classification.

 

To reach this goal, automotive companies need to leverage deep learning and digitally recreate the physical world to train and validate autonomous vehicles by leveraging hyper-cloud infrastructure to hasten the process—all while ingesting and processing massive amounts of data from test fleets almost daily.

Training and testing autonomous vehicles requires data ingestion of LIDAR, video, and other sensor data from geographically dispersed test fleets. Sampling and data-wrangling to abstract the data for training is a highly labor-intensive process. Moreover, even though there is a significant reduction in the size of the data, the image training jobs are still run on very large datasets that require the ability to scale as high as 128 GPUs as increasingly higher camera resolution is adopted. This makes it increasingly difficult to secure a turnaround time before the next ingestion cycle commences as models must be trained and ground truth must be updated before new data can be utilized.

Furthermore, once an algorithm is trained, the logical controls and vehicle design spec must be digitally validated. Essentially, it is one giant validation control loop exercise where once the algorithm is updated, the vehicle performance must also be confirmed. In sum, you have massive ingestion, storage, and compute needs—all of which are very difficult to accommodate via on-premise infrastructure in the race to be first to hit level 4 certification.

High-powered GPUs such as those found in Azure’s datacenters are essential for enabling the processes necessary for deep learning. These GPUs can train large data sets at scale, cutting the time necessary to train neural networks dramatically.

A gating factor for many organizations is they are either unwilling or unable to expend the upfront capital required to build compute, storage, and networking assets at the edge where their test fleets are. By leveraging Microsoft Azure, manufacturers can take advantage of the same high-powered GPUs without investing in costly infrastructure that also rapidly becomes obsolete.

Microsoft Azure is the platform best positioned to help automotive companies win the race to fully autonomous vehicles. In addition to possessing large scale capacity of GPUs on the cloud, Microsoft offers support for all major deep learning frameworks, support for specialized hardware including Infiniband-equipped clusters for multi-node scaling and PaaS-type capabilities that provide this scaling as a managed service. By offering scaling compute power as a managed   service, customers can save time and reduce costs, as compute resources automatically scale as needed preventing unnecessary spend on provisioning time. Moreover, Microsoft has access to ingestion tools, broad, open, and transparent support for container standards and a proven track record managing global, multisite, distributed big data implementations.

 

 

The end of the road—the future of smart and connected cities

The Microsoft Azure platform is not only the development and testing platform but also the runtime platform at scale required for the successful deployment of fully autonomous vehicles—a crucial step with broad implications for the future of intelligent transportation and mobility-as-a-service which Microsoft is uniquely situated to service. Vehicle-to-everything (V2X) initiatives promise a future where these connected and autonomous vehicles become vital sources of data in a connected, urban setting. These vehicles will communicate information with other vehicles, infrastructure, and public services regarding accidents, road conditions, and other key data—allowing others to intelligently avoid adverse road conditions, high-traffic areas, and crash sites, preventing traffic blockages, shortening commutes, and improving safety.

An autonomous vehicle would never be constrained to park near its passenger’s destination, enabling it to vacate heavily trafficked downtown areas to relieve congestion and at least partially eliminate the need to construct large parking centers in densely populated areas. Further, local events with large attendances can be more easily accessible, reducing the need for police to monitor traffic and ward against accidents involving thousands of vehicles and pedestrians entering and exiting large parking areas. As a source of data, autonomous vehicles can become the listening mechanism by which city officials monitor and improve the infrastructure we interact with every day, optimizing the way in which traffic flows through the city, providing valuable feedback on the construction of new roadways, and keeping the city up to date and maintained. Autonomous vehicles and mobility-as-a-service also stand to revolutionize access in urban spaces, providing safe and affordable transportation to passengers who otherwise couldn’t operate a vehicle including elderly and disabled citizens and children – who may no longer need to be shuttled from place to place by their busy parents.

The impact of autonomous vehicles and mobility-as-a-service will ultimately be felt across multiple industries. Electric vehicles could communicate with charging stations throughout the city, charging themselves while their owners are at work. Renewable energies can be more efficiently siphoned into these charging stations, reducing the need for excess energy to be stored in batteries. Ultimately, all of these initiatives are intertwined, and stand to have an impact spanning far beyond the automotive industry, helping to revolutionize the energy and public-sector industries to name only a few.

Get started now

44 companies have announced significant autonomous vehicle investments—a number that should continue to grow. Microsoft Azure has the hyper-scale cloud and hyper-scale network to cover the end-to-end workflow needed to train autonomous vehicles at scale, with the performance to meet required turnaround times in a secure way without an upfront investment in infrastructure. Let us help you take the future of intelligent products out of your imaginations and onto the road.

Are you ready to unlock the full value of Big Compute for your business? Learn more about our Big Compute solution portfolio today.