Article

Team fully switches to frontier LLMs for robot control in customer deployments

A team fully switched to using frontier LLMs to control robots in customer deployments over the past month. The architecture resembles a System 1/System 2 division of labor: high-level reasoning is handled by agents based on Opus and Astra, while the fast execution layer uses computer vision code written by the models themselves, which they describe as fast, editable, and reliable. LLMs also bring in-context learning and continual learning; when new situations arise, the LLM steps in and adjusts the process.

Part of the event: Team switches to frontier large models for robot control in customer deployment

This page provides a summary and source link only.