IEEE Robotics & Automation Magazine - June 2021 - 51

deployed in a waste processing plant, where it is successfully
assessed in recyclable sorting under difficult and demanding
industrial conditions.
The Need for Automation
The current trend toward an environment-friendly " circular
economy " changes the way products are made and consumed.
The circular economy is based on the assumption
that materials will not end up in landfills but will be recovered
and reused. Traditionally, the recovery of valuable
material from waste has been performed by humans, a solution
that suffers from low productivity and increased health
risks. During the past two decades, optical sorters have been
become popular in industrial material recovery facilities
(MRFs). The devices employ a combination of lights and
sensors to illuminate and capture images of objects. Image
processing reveals the qualitative characteristics of materials,
which determine whether objects should be accepted or
rejected. Optical sorters use compressed air to move rejected
objects to different bins. This technology has difficulty dealing
with heavy items, such as semifull containers, which are
not rare in waste flows. As a means to overcome the limitations
of optical sorters, a new trend involves the use of
robotic technology. This approach assumes that there is a
low integration cost and that equipment can be easily
installed in existing MRFs. Robotic waste sorting systems
rely on high-cost imaging technology that exploits recent
artificial intelligence (AI) advancements to identify and categorize
waste. However, many computer vision solutions are
built on proprietary data sets that are not publicly available.
This article supports the development of low-cost, computer
vision-based waste categorization modules that can be
directly applied in the industry. To this end, we exploit the
industrial research setup implemented in the MRF on the
island of Crete, Greece (Figure 1), to collect images of different
waste types. The images are further
processed in a lab to develop a
rich and well-documented recyclable
waste data set that is used to train an
AI-powered module for recyclable
detection and categorization. The data
set as well as the entire collection of
image processing tools is available for
public use. The implemented module
is integrated with our robotic separator
[1], [2] to recover recyclables from
urban waste streams. In addition to
the integration of the vision-based
waste sorting module, the present
work discusses the partial advancement
of the robot to reduce the time
to complete the task. The composite
system is tested in demanding industrial
operating conditions, with very
promising results (an average success
rate of 91.8%).
Related Work
Today, there is mounting pressure for more, and more
effective, recycling; in the European Union, for example,
European Commission Waste Directive 2018/850 stipulates
that, by 2035, the amount of municipal waste that
is landfilled should be reduced to 10% or less. Industrial
waste treatment has, so far, been mostly based on the manual
sorting of recyclables for the recovery of valuable, reusable
materials. However, this
is a laborious and potentially
dangerous job that
humans should avoid.
Recently, many automated
systems have been
developed that can separate
and recover materials,
such as metal, paper,
glass, and plastic, from
waste streams using optical
sorters, magnets, eddy
currents, and inductive
and near-infrared sensing;
[3] summarizes
advances in the mechanical
processes, sensors,
and actuators that are
employed. Robotic technology
is coming to support (and even replace) existing
installations by offering a more effective and autonomous
alternative to the established procedure [4], [5]. Currently,
the most well-known commercial robots used in the
waste management industry come from Sadako [6], SamurAI
[7], AMP Robotics [8], and ZenRobotics (which also
deals with the management of construction waste) [9]. The
majority of the existing systems capitalize on the agility of
Robot Cabin
Urban Waste
Installation
Controlled
Waste Mix
Robots need sophisticated
visual and manipulation
skills to be able to
work in the extremely
heterogeneous, complex, and
unpredictable waste sorting
industrial environment.
Residue Bin
Bins for Recyclables
Figure 1. A depiction of the industrial research installation at the Crete MRF. Starting from
the right, two feeders place waste on a conveyor belt. The feeders are located above the
belt to provide a controlled waste mixture or inside the processing installation to feed
real urban waste. The belt carries waste to the robot cabin for sorting. Under the cabin,
there are separate bins for each material. The residue of the sorting process ends up in
a dedicated bin.
JUNE 2021 * IEEE ROBOTICS & AUTOMATION MAGAZINE *
51

IEEE Robotics & Automation Magazine - June 2021

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2021