Why Self-Driving Cars Took So Long (And What Everyone Got Wrong)
Automated Podcast 47:53
531 views · 11 likes Watch on YouTube ↗
“The basic technology was ready quite some time ago.”
So why are self-driving cars not everywhere?
In this episode of Automated, Martial Hebert, Dean of Carnegie Mellon’s School of Computer Science, explains why the biggest challenge in AI is not building the technology.
It is proving it works in the real world.
While most people focus on breakthroughs in AI models, the real bottleneck is validation, safety, and reliability in unpredictable environments.
Martial breaks down why physical AI is fundamentally harder than software-based AI, why simulation cannot replace real-world testing, and why robotics still faces a major data problem.
He also explains why timelines in AI are so often wrong and why the answer to “when will this work” is almost always: it depends.
If you want to understand why progress in AI feels slower than expected, this is the conversation.
KEY MOMENTS
(00:00) Why self-driving cars took so long
(00:40) The role of talent, culture, and expertise
(01:16) Why AI timelines are always wrong
(03:25) Building the Robotics Innovation Center
(04:46) Why scale matters in robotics
(05:30) Physical AI vs software AI
(06:00) Why simulation fails in the real world
(07:52) The first self-driving car project (DARPA)
(09:55) Why DARPA solves “impossible” problems
(12:15) How long-term research actually wins
(14:20) Startups vs research
(15:54) How CMU became a hub for autonomy
(16:23) Why culture beats technology
(18:55) The problem with narrow thinking in AI
(23:22) Why AI breakthroughs don’t solve everything
(24:42) Why theory doesn’t work in reality
(26:47) “It depends,” the real answer in AI
(28:20) The real bottleneck: validation and safety
(30:07) The robotics data problem
(31:07) Simulation vs real-world data
(36:12) Why collaboration drives breakthroughs
(40:08) Robotics in disaster response
Connect with Martial Hebert
https://www.linkedin.com/in/martial-hebert-76448756/
Learn more about Carnegie Mellon Robotics
https://www.ri.cmu.edu/
Thoughts? Guest suggestions? Email us at podcast@automate.org
Find the transcript and more episodes at automated.fm
Unlock full access to Automated and explore everything automation.
Subscribe today on YouTube, Apple Podcasts, and Spotify.
So why are self-driving cars not everywhere?
In this episode of Automated, Martial Hebert, Dean of Carnegie Mellon’s School of Computer Science, explains why the biggest challenge in AI is not building the technology.
It is proving it works in the real world.
While most people focus on breakthroughs in AI models, the real bottleneck is validation, safety, and reliability in unpredictable environments.
Martial breaks down why physical AI is fundamentally harder than software-based AI, why simulation cannot replace real-world testing, and why robotics still faces a major data problem.
He also explains why timelines in AI are so often wrong and why the answer to “when will this work” is almost always: it depends.
If you want to understand why progress in AI feels slower than expected, this is the conversation.
KEY MOMENTS
(00:00) Why self-driving cars took so long
(00:40) The role of talent, culture, and expertise
(01:16) Why AI timelines are always wrong
(03:25) Building the Robotics Innovation Center
(04:46) Why scale matters in robotics
(05:30) Physical AI vs software AI
(06:00) Why simulation fails in the real world
(07:52) The first self-driving car project (DARPA)
(09:55) Why DARPA solves “impossible” problems
(12:15) How long-term research actually wins
(14:20) Startups vs research
(15:54) How CMU became a hub for autonomy
(16:23) Why culture beats technology
(18:55) The problem with narrow thinking in AI
(23:22) Why AI breakthroughs don’t solve everything
(24:42) Why theory doesn’t work in reality
(26:47) “It depends,” the real answer in AI
(28:20) The real bottleneck: validation and safety
(30:07) The robotics data problem
(31:07) Simulation vs real-world data
(36:12) Why collaboration drives breakthroughs
(40:08) Robotics in disaster response
Connect with Martial Hebert
https://www.linkedin.com/in/martial-hebert-76448756/
Learn more about Carnegie Mellon Robotics
https://www.ri.cmu.edu/
Thoughts? Guest suggestions? Email us at podcast@automate.org
Find the transcript and more episodes at automated.fm
Unlock full access to Automated and explore everything automation.
Subscribe today on YouTube, Apple Podcasts, and Spotify.
Playback is via YouTube's official embedded player. Data from YouTube; Exumo is not affiliated with YouTube.