Voice-controlled robotic arm
A Dobot Magician that takes instructions in plain language. Speech-to-text feeds a locally hosted LLM with tool-calling; those tools drive OpenCV detection, OCR, and a homography that maps camera pixels into arm coordinates. Nothing leaves the machine.
- Runs fully offline on local hardware — no cloud APIs anywhere in the loop.
- Letter detection with rotation-safe pick-and-place.
- Built and defended as my bachelor's thesis, June 2026.