Title

AWS re:Invent 2022 - Mobileye: Smarter, safer autonomous solutions with Habana & Granulate (PRT279)

Summary

• Intel's VP and General Manager of Data Center and AI Sales, Silberto Vargas, introduced the session, emphasizing Intel's commitment to innovative technical solutions for urgent global problems.
• Vargas highlighted the importance of semiconductors and outlined Intel CEO Pat Gelsinger's "five superpowers": ubiquitous compute, pervasive connectivity, cloud to edge infrastructure, artificial intelligence, and sensing.
• Ohad Chitri, Senior Director of Engineering and Algorithms at Mobileye, discussed scaling AI training for autonomous driving using Amazon EC2 DL1 instances powered by Habana Gaudi processors.
• Mobileye's approach to autonomous driving is based on three pillars: sensing with camera and radar/LiDAR systems, mapping through crowdsourcing, and decision-making with Mobileye's Formal Driving Policy (RSS).
• Chitri explained the engineering challenges of developing a shared backbone for computer vision tasks and the use of knowledge distillation to optimize performance.
• Mobileye leverages DL1 instances for cost-effective training, leading to more experiments and better models for safer driving.
• Pini Reisman, Director of REM Cloud Application at Mobileye, detailed the challenges of high-definition mapping for autonomous driving and how Granulate's optimization software significantly reduced CPU hours and costs for REM's map creation process.
• The session concluded with Vargas reiterating Intel's vision of enabling technology to improve lives and inviting attendees to learn more at Intel's booth and sessions.

Insights

• Intel and AWS are collaborating to address significant challenges in the technology industry, particularly in the areas of autonomous driving and AI.
• The "five superpowers" concept reflects the areas Intel believes are critical for technological advancement and innovation.
• Mobileye, an Intel company, is at the forefront of autonomous driving technology, with a focus on AI training scalability, sensing, mapping, and decision-making.
• The use of Amazon EC2 DL1 instances with Habana Gaudi processors demonstrates a commitment to cost-effective and efficient AI training for autonomous driving solutions.
• Granulate's optimization software plays a crucial role in reducing computational costs and improving efficiency in the creation of high-definition maps for autonomous driving.
• The session showcased the synergy between Intel's hardware and software capabilities, its portfolio companies like Mobileye and Granulate, and AWS's cloud infrastructure to drive innovation in autonomous driving technology.