EXIMO teaches a robotics model to imitate its control setup, a post says

A post describes EXIMO as using Gemini to guide Gemini Robotics through an external control setup that supports fast iteration. It says imitation training transfers that orchestrated behavior into the robotics model itself. The author argues that absorbing this scaffolding lets the model undergo further reinforcement learning—training through rewards—and potentially go beyond what the hand-designed setup could achieve, setting up another cycle of external control design.