Ranjay Krishna on Why Robots Still Fail in the Real World and the Data Problem Holding Them Back

Robotics is advancing quickly, but real-world deployment is still far more difficult than most people expect.

In this episode of Automated, Brian Heater speaks with Ranjay Krishna about the fundamental challenges preventing robots from working reliably outside controlled environments, and why solving the data problem is key to unlocking the next wave of robotics.

Ranjay explains why today’s robots struggle with tasks that humans find intuitive, from learning by observation to understanding perspective and adapting to new environments. While AI models have made massive progress in language and vision, robotics introduces a new layer of complexity where actions change the world in real time and small errors compound over time.

The conversation explores the limitations of current approaches, including why training robots in simulation often fails to translate to the real world, and how the lack of diverse environments creates major gaps in performance. Ranjay shares how his team at the Allen Institute is addressing this by building large-scale simulated environments designed to better reflect the variability of real-world spaces.

They also discuss the concept of an ImageNet moment for robotics, and what it would take to create the kind of large, diverse datasets that transformed AI. By generating hundreds of thousands of simulated environments and scaling data collection, his team is exploring whether robots can learn more effectively in simulation and generalize those skills into the physical world.

The conversation also covers why robotics requires more than just better models, including challenges in hardware, sensing, and real-world interaction. From embodiment and perception to reasoning and adaptation, it is a grounded look at why robotics remains one of the hardest problems in AI and what needs to happen next for the industry to move forward.

We’d love to hear from you. Have thoughts or guest suggestions? Reach us at [e-mail].

You can find the transcript and more episodes of Automated at automated.fm.

Also, join us at MassRobotics for a happy hour with Brian Heater from A3. Wednesday, April 8 - 4:30 PM - 6:00 PM EDT

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/

Hosted on Acast. See acast.com/privacy for more information.