S C R A P S O R T
The Problem
Waste disposal is becoming an ever greater concern in our modern world. While more waste is being generated year over year, waste processing is not keeping pace.
In an effort to combat the environmental harm of waste, many elect to use products and packaging that are recyclable. However, less than a third of recyclable waste is actually recycled. In the US, millions of tons of recyclables are discarded each year: 6.87 million tons of glass, 13.8 million tons of metals, and 18.35 million tons of paper.
Increasing efficiency in the waste management sector and capturing a greater share of recyclables would yield a sizable reduction in landfill usage.
There are several issues with the current trend of centralized waste processing plants. The problem of classification is compounded when a plant can receive nearly all possible waste. Reliable classification of such a vast and diverse set of items is exceedingly difficult. The high cost of hardware and software for complicated sorting systems is prohibitive outside of large metropolitan areas. Another challenge faced by centralized plants is contamination of recycling streams. Trash from an office building, for example, which may consist mostly of paper, may be mixed with waste from, say, a restaurant or factory. Contaminated recycling streams are often sent straight to landfills because they can no longer be salvaged.
The Solution
In order to alleviate these issues, we propose an autonomous system that can sort trash on-site immediately after it is discarded. Such a system would recognize various kinds of trash, sort the trash into bins for further processing, and operate at the edge self-sufficiently with limited space, power consumption, and cost.
The system, known as Scrapsort, will use a robotic arm to serialize trash on a conveyor belt. A computer vision algorithm will recognize the trash and categorize it. The computer vision task will be accomplished with an inexpensive, ultra-low-power processor with a hardware accelerator for convolutional neural networks, allowing the system to perform this operation cheaply and efficiently.
The conveyor belt will be lined with pistons, which will move trash items into appropriate bins, each of which may hold, for example, paper, plastic, metal, etc.
Diagram of basic system operation
Scrapsort within a broader waste management system
In a real-world context, Scrapsort would merely constitute the first stage of a waste management system. Trash would be sorted inexpensively on-site by Scrapsort.
Different Scrapsort systems would use different computer vision models trained for a particular context. For example, waste from a school may comprise several different kinds of paper waste that need to be processed separately, whereas a factory may dispose of different kinds of metal.
Sorting on-site prevents cross contamination and allows for highly specialized and tuned computer vision systems which may only need to recognize a handful of waste categories. Once Scrapsort has sorted the trash into separate bins, the waste can be sent to specialized facilities for processing. This obviates the need for an expensive central facility to handle all possible waste inefficiently.