Grasp Point Calculation and Food Waste Detection for Dish-recycling Robot

Abstract

The development of automatic dish-recycling robots is expected to make food industries more effective. However, grasp point calculation for the robot is a time-consuming process that disturbs real-time on-site dish recycling. Furthermore, usually, food waste exists which may dirty robots or restaurants by grasping the dish as an empty dish. This paper introduces an improved grasp points calculation method using image processing and proposes an anomaly detection model for realizing food waste detection. In detail, the target recycling dishes are detected by YOLO, a deep learning-based object detection model at first. Then, an optimized grasp points calculation method is applied to decide the grasp point from each detected dish for the robot hand grasp. We also adopt multi-class anomaly detection for food waste detection, based on the object name which is the output class of the YOLO. The experimental results show the grasp points are extracted well and multi-class waste food detection achieves an ACU score of 0.85, proving the proposal's effectiveness.