Why Robots Still Fail in the Real World
Automated Podcast 42:57
485 views · 16 likes Watch on YouTube ↗
“Simulation is an approximation of the real world.”
Robotics is advancing fast, but deploying robots outside controlled environments is still incredibly difficult.
In this episode of Automated, Ranjay Krishna explains why robots struggle with real-world tasks, from learning by watching humans to adapting to unpredictable environments.
Ranjay breaks down the core challenges in robotics today, including the sim-to-real gap, the lack of diverse training data, and why errors compound over time when robots interact with the physical world.
Brian and Ranjay also discuss why simulation may still be the most scalable path forward, how large-scale synthetic environments could unlock new capabilities, and what an ImageNet moment for robotics might look like.
KEY MOMENTS
(00:00) Why robotics struggles in the real world
(02:16) The data collection problem in robotics
(03:29) Why robots need unstructured environments to learn
(05:15) The sim to real gap explained
(07:47) Why simulation lacks real-world diversity
(09:24) Training general-purpose models in robotics
(11:17) Why robotics is harder than AI models
(13:20) Moving from navigation to full task execution
(21:48) The ImageNet moment for robotics
(25:25) Can simulation-trained robots work in the real world
(30:16) Why deployment is still difficult today
(32:06) Why robots can’t learn like humans yet
(35:12) The role of simulation and real-world learning together
(38:08) Why embodiment matters in robotics
Connect with Ranjay
https://www.linkedin.com/in/ranjay-krishna-1a344444/
Thoughts? Guest suggestions? Email us at podcast@automate.org
Also, join us at MassRobotics for a happy hour with Brian Heater from A3. Wednesday, April 8, 4:30 PM - 6:00 PM EDT.
https://luma.com/on8uvgte
We’d love to hear from you. Have thoughts or guest suggestions? Reach 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 and leave a review on YouTube, Apple Podcasts, and Spotify.
Subscribe to the Automated Newsletter:
https://www.automate.org/automation/newsletter-automation-roundup
You can also find us on:
LinkedIn https://www.linkedin.com/showcase/automated-podcast-by-a3/
Instagram https://www.instagram.com/automatedpod/
Robotics is advancing fast, but deploying robots outside controlled environments is still incredibly difficult.
In this episode of Automated, Ranjay Krishna explains why robots struggle with real-world tasks, from learning by watching humans to adapting to unpredictable environments.
Ranjay breaks down the core challenges in robotics today, including the sim-to-real gap, the lack of diverse training data, and why errors compound over time when robots interact with the physical world.
Brian and Ranjay also discuss why simulation may still be the most scalable path forward, how large-scale synthetic environments could unlock new capabilities, and what an ImageNet moment for robotics might look like.
KEY MOMENTS
(00:00) Why robotics struggles in the real world
(02:16) The data collection problem in robotics
(03:29) Why robots need unstructured environments to learn
(05:15) The sim to real gap explained
(07:47) Why simulation lacks real-world diversity
(09:24) Training general-purpose models in robotics
(11:17) Why robotics is harder than AI models
(13:20) Moving from navigation to full task execution
(21:48) The ImageNet moment for robotics
(25:25) Can simulation-trained robots work in the real world
(30:16) Why deployment is still difficult today
(32:06) Why robots can’t learn like humans yet
(35:12) The role of simulation and real-world learning together
(38:08) Why embodiment matters in robotics
Connect with Ranjay
https://www.linkedin.com/in/ranjay-krishna-1a344444/
Thoughts? Guest suggestions? Email us at podcast@automate.org
Also, join us at MassRobotics for a happy hour with Brian Heater from A3. Wednesday, April 8, 4:30 PM - 6:00 PM EDT.
https://luma.com/on8uvgte
We’d love to hear from you. Have thoughts or guest suggestions? Reach 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 and leave a review on YouTube, Apple Podcasts, and Spotify.
Subscribe to the Automated Newsletter:
https://www.automate.org/automation/newsletter-automation-roundup
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.