Computer Vision for Safe Unloading in Pharma Production – Cilyx

Challenge

Nature of collaboration

To solve this, Cilyx partnered with Sagacify to develop an AI solution capable of determining a safe extraction sequence based on real-time visual input. The teams co-developed a model tailored to Cilyx’s specific packaging format and deployed a custom labeling environment to support high-precision training data.

What we built

Sagacify built an AI model that analyzes 2D images captured by a camera mounted on the robotic arm. The system identifies all visible boxes in the crate, detects which ones have all four corners free of overlap, scores them based on pickability, and transmits precise (x, y, z) coordinates to guide extraction.

The process is executed in sequential steps: initial detection of areas of interest using reduced-resolution images, high-resolution corner detection, and scoring of free boxes. Once a removable box is identified, the robotic arm is instructed to proceed. The operation repeats until the crate is empty. The model runs entirely on the edge, embedded in a compute unit on the robotic arm.

Impact