LocoBot

For my Computer Vision and Cognitive Systems exam, I built a system capable of recognizing hand gestures that command a robot to track or stop.

What I did:

• Selected YOLOv5, pre-trained on Microsoft COCO, as the base network.
• Fine-tuned it on the HaGRID dataset — roughly 700 GB of gesture images with varying resolutions and quality.
• Modified the network to retain the knowledge from its original training while extending it to the new gesture-recognition task, so the robot could still map objects encountered while tracking.