AI IN AV

Download the Solution Brief Start Developing

AUTONOMOUS DRIVING IS ON THE VERGE OF TRANSFORMING THE TRANSPORTATION INDUSTRY.

Driven by a vision of a safer and more efficient transportation system, the global autonomous vehicle (AV) market is projected to be valued at nearly $1.5 billion by 2026. ^[1]

The autonomous vehicle industry relies on artificial intelligence – especially deep learning solutions – to operate components of not only autonomous driving systems, but also advanced driver-assistance systems (ADAS) and industry-related fields like car maintenance, supply chain, and marketing.

Habana offers hardware and software solutions that shorten the time of AI training cycles, accelerating the pace of innovation and time to market. In a competitive industry like autonomous driving, Habana solutions reduce development and validation costs, allowing researchers to explore more within their budget.

Applications of AI in AV

AI-based algorithms are present in all levels of automation.

Utilizing data from a network of sensors and cameras, AI-based and deep learning algorithms execute short- and long-term decisions within AVs. Computer vision systems help perform lane departure warnings, and emergency brake assistance in driver assistance and partial automation systems. Large semantic segmentation tasks running over multiple cameras and point clouds help AVs detect and recognize road lanes, other vehicles on the road, pedestrians, traffic signs, lights, etc. Deep learning is also used for tracking those objects and predicting their behavior, which helps derive control decisions. Deep reinforcement learning algorithms are also widely used in planning and controlling components of an AV. Fully automated driving, where the vehicle independently controls complete journeys on the highway as well as in city traffic, would not be possible without these advancements in AI.

Challenges

The main challenge in autonomous driving is hyperscaling. Streets are wild environments and the variety of objects that can appear on the road is limitless. Core perception algorithms need to evolve to detect these objects in geographically diverse locations with various lighting and weather conditions, which is possible only by continuously adding interesting and informative samples to the train data. This means lots of training iterations on enormous datasets, which then require thousands of compute nodes for development and validation. This raises the need for instances that can handle large distributed machine learning training jobs.

Why is gaudi a good fit for AV Use Cases?

To adapt to ever-changing road conditions, the autonomous driving industry needs scalable compute infrastructures that can handle many large distributed ML training jobs. There is an increasing need for frequent retraining and updating of models to reach generalizability across different location conditions.

The two primary considerations that come into play when employing AI processing are time- and cost-to-train. Habana’s Gaudi Training Processors are expressly designed—in both hardware and software—to deliver high-efficiency cost- and time-to-train, making AI training more accessible to more organizations and for more applications.

First-generation Gaudi delivers up to 40% better price-performance than comparable GPU-based solutions—for both the EC2 DL2 instance and on-premise systems. Training with Gaudi clusters is available both in the cloud with AWS EC2 DL1 instances consisting of 8 Gaudis and on-premises with the Supermicro X12 Gaudi Training Server also consisting of 8 Gaudis. In addition, Gaudi2, which launched in May 2022, offers substantial performance advances that enable significantly faster training of models while preserving cost-efficiency. Gaudi2 systems will be available from Supermicro in 2H 2022 for on-premises implementation.

News and customer testimonials

“As a world leader in automotive and driving assistance systems, training cutting edge Deep Learning models is mission-critical to Mobileye business and vision. As training such models is time consuming and costly, multiple teams across Mobileye have chosen to use Gaudi-accelerated training machines, either on Amazon EC2 DL1 instances or on-prem; Those teams constantly see significant cost-savings relative to existing GPU-based instances across model types, enabling them to achieve much better Time-To-Market for existing models or training much larger and complex models aimed at exploiting the advantages of the Gaudi architecture. We’re excited to see Gaudi2’s leap in performance, as our industry depends on the ability to push the boundaries with large-scale high performance deep learning training accelerators.”

Gaby Hayon
Executive Vice President of R&D
Mobileye

“On our own models the increase in price performance met and even exceeded the published 40% mark.”

Chaim Rand
Mobileye

“We are consistently seeing cost-savings compared to existing GPU-based instances across model types, enabling us to achieve much better time-to-market for existing models or training much larger and complex models.”

David Peer
DevOps Tech Lead & Specialist
Mobileye