Why Real-World Data Will Decide the Future of Robotics
Automated Podcast 46:28
672 views · 25 likes Watch on YouTube ↗
Robotics is getting better fast.
But Sergey Levine says the real breakthrough will not come from better demos alone. It will come from real-world data, better learning loops, and systems that can improve through deployment.
In this episode of Automated, Brian Heater speaks with Sergey Levine of Physical Intelligence about why robotics has hit an inflection point, and why the future of physical AI depends on far more than just clever models in a lab.
They discuss why useful robot learning looks more like guided practice than repetition, why common sense is still one of the hardest problems in robotics, and why real-world data may be far more practical to collect than most people assume.
Brian and Sergey also explore why the future of robotics is not just humanoids, what a general robotics intelligence layer could unlock, and why Physical Intelligence is built more like a research lab than a traditional startup.
If you want a sober, practical look at what is really happening in robotics right now, this is the conversation.
KEY MOMENTS
(00:00) Why today’s deployments are still experiments
(01:57) Why Sergey Levine left Google DeepMind
(03:16) Why robotics needs more than a lab
(05:35) What real-world robot deployment looks like now
(08:53) What counts as a useful robot experience
(11:04) Why common sense is now the real bottleneck
(13:44) Why physical AI still has major open challenges
(15:23) The upside of truly general robotic foundation models
(17:03) Why robots should not be “metal humans”
(18:19) Why embodiment mattered less than expected
(20:10) Building a robotics team with very different skill sets
(21:43) The 33-lab experiment that changed his thinking
(24:54) Why industry data now dwarfs academia
(27:43) Why real-world data is more practical than people think
(30:02) The boring but pragmatic path to better robots
(32:17) What it means to supervise a robot’s thoughts
(34:13) Why sober capability assessment matters
(37:28) Why Physical Intelligence is structured like a lab
(40:09) Why the company has almost no titles
(41:22) Why academia can drift toward the wrong problems
(43:56) How this era of physical AI may be remembered
Connect with Sergey Levine
https://www.linkedin.com/in/sergey-levine-5a31a24
Learn more about Physical Intelligence
https://www.physicalintelligence.company/
We’d love to hear from you.
Have thoughts or guest suggestions?
Reach us at podcast@automate.org.
You can find the transcript and more episodes of Automated at automated.fm.
Unlock full access to Automated and explore everything automation.
Subscribe today and leave a review on YouTube, Apple Podcasts, and Spotify.
Subscribe to the Automated Newsletter:
https://www.automate.org/automation/automated-newsletter
You can also find us on:
LinkedIn https://www.linkedin.com/showcase/automated-podcast-by-a3/
Instagram https://www.instagram.com/automatedpod/
But Sergey Levine says the real breakthrough will not come from better demos alone. It will come from real-world data, better learning loops, and systems that can improve through deployment.
In this episode of Automated, Brian Heater speaks with Sergey Levine of Physical Intelligence about why robotics has hit an inflection point, and why the future of physical AI depends on far more than just clever models in a lab.
They discuss why useful robot learning looks more like guided practice than repetition, why common sense is still one of the hardest problems in robotics, and why real-world data may be far more practical to collect than most people assume.
Brian and Sergey also explore why the future of robotics is not just humanoids, what a general robotics intelligence layer could unlock, and why Physical Intelligence is built more like a research lab than a traditional startup.
If you want a sober, practical look at what is really happening in robotics right now, this is the conversation.
KEY MOMENTS
(00:00) Why today’s deployments are still experiments
(01:57) Why Sergey Levine left Google DeepMind
(03:16) Why robotics needs more than a lab
(05:35) What real-world robot deployment looks like now
(08:53) What counts as a useful robot experience
(11:04) Why common sense is now the real bottleneck
(13:44) Why physical AI still has major open challenges
(15:23) The upside of truly general robotic foundation models
(17:03) Why robots should not be “metal humans”
(18:19) Why embodiment mattered less than expected
(20:10) Building a robotics team with very different skill sets
(21:43) The 33-lab experiment that changed his thinking
(24:54) Why industry data now dwarfs academia
(27:43) Why real-world data is more practical than people think
(30:02) The boring but pragmatic path to better robots
(32:17) What it means to supervise a robot’s thoughts
(34:13) Why sober capability assessment matters
(37:28) Why Physical Intelligence is structured like a lab
(40:09) Why the company has almost no titles
(41:22) Why academia can drift toward the wrong problems
(43:56) How this era of physical AI may be remembered
Connect with Sergey Levine
https://www.linkedin.com/in/sergey-levine-5a31a24
Learn more about Physical Intelligence
https://www.physicalintelligence.company/
We’d love to hear from you.
Have thoughts or guest suggestions?
Reach us at podcast@automate.org.
You can find the transcript and more episodes of Automated at automated.fm.
Unlock full access to Automated and explore everything automation.
Subscribe today and leave a review on YouTube, Apple Podcasts, and Spotify.
Subscribe to the Automated Newsletter:
https://www.automate.org/automation/automated-newsletter
You can also find us on:
LinkedIn https://www.linkedin.com/showcase/automated-podcast-by-a3/
Instagram https://www.instagram.com/automatedpod/
Playback is via YouTube's official embedded player. Data from YouTube; Exumo is not affiliated with YouTube.