FloorScan ML – Robotic Floor Mapping & Cleanliness AI
Overview
Developed a machine learning system for a small, handheld mapping robot that scans floor textures, identifies debris hotspots via computer vision, and outputs an interactive cleanliness heatmap.
Deep Dive
FloorScan ML bridges computer vision and embedded robotics to digitize spatial cleanliness tracking. Powered by a Raspberry Pi, the system collects video feeds from a multi-camera array and processes them through a custom PyTorch convolutional neural network (CNN) trained on semantic segmentation to identify dirt, dust, and debris. Simultaneously, it uses SLAM (Simultaneous Localization and Mapping) to track the robot's relative grid coordinates. These spatial and vision data points are dynamically combined in a ROS-managed stream to compile a detailed 2D heatmap showing cleanliness hotspots, marking high-traffic dirt concentrations to optimize facility cleaning routes.