Efficient Waste Collection via Edge Perception and Optimized Mobile Routing

We present a two-stage framework for robotic waste collection that combines global perception and local planning. Fixed edge cameras perform wide-area object detection, and a Multi-Layer Perceptron (MLP) corrects geometric projection errors for improved localization. Mobile robots then refine poses using depth sensing and execute pickup actions along routes optimized by a multi-objective Traveling Salesman Problem (TSP) planner. Experiments in a controlled lab demonstrate that the MLP correction significantly enhances positioning accuracy, while the optimized routing reduces path length and improves collection efficiency.