Dual Mode Controlled Water Surface Garbage Collecting Robot by using Embedded Deep Learning

Abstract

Plastic waste has become one of the most prevailing environmental concerns in modern society. Without proper environmental control policies and regulations in place, we have witnessed plastic garbage being irresponsibly disposed of into rivers and other water reservoirs, causing blockage of water passage as well as toxic contamination. Since plastic materials are unable to degrade in nature so quickly as biomaterials, human workers are typically required to collect plastic garbage, either manually or mechanically assisted. In some conditions, the maneuvers could cause personal injuries and infections. To resolve this issue, this paper proposes the design and development of a prototype water surface garbage collecting robot. The prototype was implemented on a Raspberry Pi board and controllable from an Android device via wireless (WiFi™) channel. Controlling robot movement, speed, and conveyor belt could be done either manually or automatically. The former was enabled by intuitive user interface, while the latter relied on deep learning of local scene acquired within active range. Experimental results of automatic operation, driven by deep learning models demonstrated garbage collecting performance at 83.25% accuracy.