To cope with challenges influenced by population aging and workforce insufficiency, artificial intelligence (AI) and robotics have been considered to improve productivity in the recycling industry. To understand the challenges in recycling better, we analyzed five mixed constructional waste recycling factories in Tokyo areas and nine kinds of sorting robots implemented worldwide. Research showed that sorting waste by materials is the hardest part of automated recycling; this activity still relies on the worker’s ability in Japan. Also, today’s waste sorting technology is still insufficient to replace manual sorting due to sorting accuracy problems and the inability to sort waste under unclear conditions. This research tries to figure out the ideal characteristic of waste recognition AI’s training data to improve the efficiency and effectiveness of the AI training process. Three characteristics tested in this research are “quantity of the targeted object,” “presence of shadows,” and “whole/partial appearance.”