ASTHRA Vision

The
Challenge

A welcome robot needs to know which visitor it is addressing and where that person is in the frame before it can turn or respond. Raw per-frame detections are noisy, and switching targets whenever another person appears breaks continuity.

The
Solution

ASTHRA Vision is the computer vision subsystem for the ASTHRA welcome robot. It runs YOLO person detection on webcam or ESP32-CAM frames, locks a target with IoU matching and a timeout, smooths position using an exponential moving average, and derives distance, screen zone, and a virtual neck yaw angle. The repository includes Raspberry Pi 4 deployment scripts and NCNN export guidance; physical servo control is not implemented yet.