Inside NVIDIA’s Race to Build a Robot Brain: Interview with Deepu Talla
Automated Podcast 51:13
2,583 views · 21 likes Watch on YouTube ↗
Just over a decade ago, NVIDIA got out of mobile processing and into robotics. Deepu Talla helped lead the risky push that is now paying off in the form of the chipmaker’s Jetson platform. The industry veteran explains how a company best known for its gaming graphics cards became a ubiquitous fixture on next-gen robotics.
Key Moments
00:00 – Why the world is racing to build a general-purpose robot brain
00:38 – Welcome + Deepu’s rockstar status at Automate
01:02 – Jumping from mobile to robotics at NVIDIA
03:19 – How mobile tech paved the way for robotics
06:01 – Why the best minds are now focused on physical AI
07:14 – The two breakthroughs enabling robotics: LLMs + simulation
10:12 – Inside NVIDIA’s robotics platform (training, sim, deployment)
14:57 – The massive data gap in robotics
16:45 – Synthetic data and preventing AI hallucinations
18:58 – How robots learn: teleop, imitation, video, gen AI
20:19 – Building a generalist robot brain like ChatGPT
22:22 – Using LLMs and multimodal models for robots
26:01 – Step-function breakthroughs change everything overnight
27:56 – Why ease of use is as critical as accuracy
29:52 – Lowering barriers for building robotics tech
31:20 – NVIDIA’s open platform and open source approach
34:06 – Tailoring robotics tech for each industry use case
36:30 – Baking safety into chips, software, and systems
42:09 – Are we underestimating the difficulty of robotics?
44:40 – Robotics is a multi-decade journey
45:06 – Why humanoids are the testbed for a general-purpose brain
48:08 – Biggest surprise from 10 years in robotics: the power of simulation
Thoughts? Guest suggestions? Email us at podcast@automate.org.
You can find the transcript and more episodes of Automated at automate.org/podcast.
Subscribe to the Automated Newsletter: https://www.automate.org/automation/newsletter-automation-roundup
Unlock full access to Automated and explore everything automation. Subscribe today on YouTube, Apple Podcasts, and Spotify.
You can also find us on:
LinkedIn https://www.linkedin.com/showcase/automated-podcast-by-a3
Key Moments
00:00 – Why the world is racing to build a general-purpose robot brain
00:38 – Welcome + Deepu’s rockstar status at Automate
01:02 – Jumping from mobile to robotics at NVIDIA
03:19 – How mobile tech paved the way for robotics
06:01 – Why the best minds are now focused on physical AI
07:14 – The two breakthroughs enabling robotics: LLMs + simulation
10:12 – Inside NVIDIA’s robotics platform (training, sim, deployment)
14:57 – The massive data gap in robotics
16:45 – Synthetic data and preventing AI hallucinations
18:58 – How robots learn: teleop, imitation, video, gen AI
20:19 – Building a generalist robot brain like ChatGPT
22:22 – Using LLMs and multimodal models for robots
26:01 – Step-function breakthroughs change everything overnight
27:56 – Why ease of use is as critical as accuracy
29:52 – Lowering barriers for building robotics tech
31:20 – NVIDIA’s open platform and open source approach
34:06 – Tailoring robotics tech for each industry use case
36:30 – Baking safety into chips, software, and systems
42:09 – Are we underestimating the difficulty of robotics?
44:40 – Robotics is a multi-decade journey
45:06 – Why humanoids are the testbed for a general-purpose brain
48:08 – Biggest surprise from 10 years in robotics: the power of simulation
Thoughts? Guest suggestions? Email us at podcast@automate.org.
You can find the transcript and more episodes of Automated at automate.org/podcast.
Subscribe to the Automated Newsletter: https://www.automate.org/automation/newsletter-automation-roundup
Unlock full access to Automated and explore everything automation. Subscribe today on YouTube, Apple Podcasts, and Spotify.
You can also find us on:
LinkedIn https://www.linkedin.com/showcase/automated-podcast-by-a3
Playback is via YouTube's official embedded player. Data from YouTube; Exumo is not affiliated with YouTube.