"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis

E12: The Robot Revolution with Keerthana Gopalakrishnan of Google Robotics

Mar 30, 2023
Keerthana Gopalakrishnan, a Robotics and machine learning engineer at Google AI, dives into the future of robotics and AI. She discusses training robots through supervised learning, contextualizing commands with language models, and the challenges of data scarcity. Gopalakrishnan highlights the ethical implications of human-robot interactions, emphasizing the balance of safety and efficiency in advanced robotics. With insights into using robotic technology to better understand ourselves, she shares her hopes and fears about AI's societal impact.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

Robot Control System

  • Robots use a two-level control system: high-level for task sequencing and low-level for execution.
  • The SayCan language model handles high-level planning, while RT1 handles specific actions.
INSIGHT

Scaling Robot Data

  • Keerthana is working on scaling autonomous data collection and mining internet data, like human videos.
  • This aims to overcome the bottleneck of limited robot-specific action data.
INSIGHT

Tokenizing Actions

  • Foundation models in robotics rely on data interoperability, converting images, language, and actions into tokens.
  • RT1's innovation is tokenizing actions, treating them like language tokens for prediction.
Get the Snipd Podcast app to discover more snips from this episode
Get the app