Y Combinator: How General-Purpose LLMs Enable Robot-Use Agents
In this episode of Y Combinator's Decoded, the founders of robotics startups Waddle Labs and RoboCurve discuss the emerging paradigm of robot-use agents. Drawing on recent research and ideas highlighting how coding agents generalize beyond software, the conversation examines using general-purpose models to write policies and control distinct robotic hardware with minimal robot-specific training. The discussion explores code-as-policies, vision-language-action models, and the balance between in-context learning and model training. They also cover the engineering harnesses, evaluations, and skill-abstraction methods required to deploy these agent systems effectively in the physical world.
Original sources
YouTube Y Combinator
AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.