Towards a Low-Cost Object Collecting and Organizing Household Robot using Deep Learning

Abstract

Recent advances in deep learning algorithms, and the availability of low-cost sensors, processors, and actuators are opening up a new opportunity for making household robots with a limited budget. Household robots can help automate the monotonous tasks of daily life such as putting dishes in the dishwasher, folding the laundry after washing, cleaning the floor, and organizing kids’ toys on the shelf at the end of the day. In this paper, a robot has been developed that can detect and recognize an object using deep learning from images, move toward the object scoops and lifts the object, and then puts the object to its assigned level on a shelf. This robot can be used to collect and organize objects such as toys, in the kid’s room. A deep learning model is trained with a custom dataset and the mean average precision (mAP) of the object detector is 79.1%. A prototype of the robot - with mechanical structure, camera, motors, controller, and image processing algorithms - is developed and tested successfully.