Robotic Waste Sorter With Agile Manipulation and Quickly Trainable Detector

Owing to human labor shortages, the automation of labor-intensive manual waste-sorting is needed. The goal automating replace role robust detection and agile manipulation waste items with robots. To achieve this, we propose three methods. First, provide a combined method using graspless push-and-drop pick-and-release manipulation. Second, robotic system that can automatically collect object images quickly train deep neural-network model. Third, mitigate differences in appearance target objects from two scenes: one for dataset collection other sorting recycling factory. If exist, performance trained detector may decrease. We address illumination background by applying scaling, histogram matching equalization, synthesis source target-object images. Via experiments an indoor experimental workplace waste-sorting, confirm proposed methods enable quick training image sets classes (i.e., aluminum can, glass bottle, plastic bottle) higher than do not consider differences. also enables robot manipulate objects.