Training a robot to sort recycling using deep learning
Supported by the Global Engagement Funds, a UCL and Istituto Italiano di Tecnologia collaboration leads to creation of a garbage-sorting robot
15 September 2021
Domestic garbage management is an important aspect of a sustainable environment, but with rapid urbanisation leading to the generation of more waste and excessive littering, waste collection and sorting are becoming increasingly challenging. Recycling facilities around the world rely on workers sorting items by hand, which is challenging and potentially hazardous.
To address this, Dr Dimitrios Kanoulas (UCL Computer Science) and his team have been collaborating with the Istituto Italiano di Tecnologia (IIT) in Genoa, Italy on the RECYCLE project to create an exciting and novel garbage classification and localisation system.
Supported by the UCL Global Engagement Funds' 2020/21 round of funding and the European Union’s Horizon 2020 SOPHIA Project, the collaboration has led to the production of a robot that can grasp and sort different recyclable types of garbage.
Dimitrios said: "The RECYCLE project, aimed at enabling autonomous mobile robots to navigate the environment, detect garbage and, based on their type, place them into the right recycling bin. Our RPL, UCL-CS lab worked closely with the HRI2 lab of Arash Ajoudani at IIT, Italy to achieve a real-world working version of this system in indoor environments. Towards a sustainable world, we aim to continue this work by enabling a fully autonomous garbage collection and recycling system, with our mobile robots navigating in real outdoor environments, such as parks and pavements. Special thanks to all people that worked on the project, including Pietro Balatti from IIT, Kirsty Ellis and Dennis Hadjivelichkov from UCL."
Using AI to tackle recycling
The mobile robot works by identifying the location and type of the item and adjusting its grasping pose in order to effectively grab the item before sorting it into its allocated bin.
To achieve this, the researchers trained a deep neural network called GarbageNet to detect and recognise different recyclable items through a visual sensor on the robot. Next, they created an algorithm that could identify the grasp pose needed to pick the item up from the ground. Finally, they designed a whole-body control framework, which allows the robot to grasp and place the item into the relevant bin.
Watch the video below to see the robot in action:
As the robot has the ability to collect and sort garbage simultaneously, the researchers hope to reduce the need for further classification in recycling centres, thereby minimising the burden on workers and their need to come into contact with potentially hazardous waste.
From software to hardware
Following the success of the initial stages of the project, the researchers hope to adapt the system and the hardware associated with it to the “wild” under a variety of different conditions, which would make the robot ready for real-life utilisation. Their focus is on ensuring the robot can find garbage in wide outdoor spaces and that the garbage collection and classification system is as time and energy efficient as possible.
The project is undoubtedly an optimistic step forward towards a more environmentally-conscious future.
The initial findings of the project are published in the paper Garbage Collection and Sorting with a Mobile Manipulator using Deep Learning and Whole-Body Control by Jingyi Liu, Kirsty Ellis, Denis Hadjivelichkov, Danail Stoyanov and Dimitrios Kanoulas from UCL Computer Science and Pietro Balatti and Arash Ajoudani from IIT.
The Global Engagement Funds 2021/22 round of funding is now open for applications. Find out more and apply here.
- The RECYCLE project's website
- Garbage Collection and Sorting with a Mobile Manipulator using Deep Learning and Whole-Body Control paper
- Global Engagement Funds 2020/21 recipients
- SOPHIA Project
- Istituto Italiano di Tecnologia
For the latest news about UCL’s international activity, partnerships and opportunities, subscribe to the bimonthly Global Update newsletter.