Cornea: Image Segmentation Skills from the Telekinesis Agentic Skill Library
This new library gives robots pixel-perfect vision for complex manipulation tasks.
Deep Dive
The Telekinesis Agentic Skill Library has released Cornea, a new module providing over 10 different image segmentation skills for robotics applications. It combines classical computer vision techniques (RGB/HSV/LAB segmentation, Watershed, GrabCut) with deep learning models like BiRefNet and Meta's Segment Anything Model (SAM). Developers can use it for pixel-level image understanding in tasks such as vision-guided pick-and-place, palletizing, obstacle detection, and grasp planning for Physical AI agents.
Why It Matters
It significantly lowers the barrier for building advanced robotic vision systems, accelerating development of practical Physical AI applications.