Listen to this episode on Apple Podcasts, YouTube, Spotify, and Substack. Transcripts are also available by clicking on the “transcripts” button in the top right corner of this post on the app/web version. You can also access them from the email version by clicking here.
In today’s episode, I’m speaking with Matt Cragun, Director of Product for Autonomous Vehicles at NVIDIA. Matt breaks down NVIDIA’s strategy for autonomous vehicle development and how the company is approaching both the hardware and software sides of the AV stack. We discuss how NVIDIA works with AV developers to accelerate development, the role of its Hyperion platform, and how NVIDIA is making its technology easier to integrate across different vehicle platforms.
We also dive deep into Alpamayo, NVIDIA’s open-source family of AI models for autonomous driving. Matt explains how Alpamayo works, how it was trained, and how developers can adapt the models to their own use cases. We discuss the limitations of the technology, how it approaches edge cases, and how Alpamayo combines end-to-end AI with more classical autonomous driving approaches rather than relying entirely on one or the other.
Matt also discusses how Level 4 autonomy will eventually make its way from robotaxis into consumer vehicles, where the AV industry stands today, and what still needs to happen before AV technology can scale more broadly. Finally, he shares what excites him most about the next few years of autonomous driving and where he expects the technology to make the biggest advances.
Chapters
(00:00) Introduction to Matt Cragun
(02:10) What is Alpamayo?
(03:50) Is NVIDIA more focused on building the hardware (chips) or software (models) for AVs?
(05:29) NVIDIA’s strategy for AV development
(07:03) How NVIDIA helps new AV developers accelerate development
(08:44) How open-source is Alpamayo, and can it be run on non-NVIDIA chips?
(09:28) Hyperion and how NVIDIA democratizes access to AV hardware
(10:40) How Hyperion is vehicle-agnostic and easy to integrate into different vehicle platforms
(11:20) How Level 4 autonomy will make its way from robotaxis to consumer vehicles in the coming years
(15:10) Digging deep into Alpamayo: How it works and how it was trained
(17:16) The limitations of Alpamayo and how AV developers can tailor it to their use cases
(19:20) The different use cases for Alpamayo, and its approach
(22:15) How Alpamayo combines the end-to-end AI and classical AV approaches
(27:23) How their AV models handle edge cases
(30:46) The current state of AV development
(32:52) What excites Matt most about the next few years for AVs
(33:50) Conclusions and final thoughts
Notes/Links:
You can find Matt Cragun on LinkedIn.
An article about Alpamayo 2 Super on Hugging face, mentioned at the 04:43 timestamp (link).
NVIDIA’s article about Alpamayo 2 Super (link).
NVIDIA Alpamayo landing page (link).
-Harry
