How the AI Works

Our idea uses artificial intelligence (AI), advanced computer processing, to identify rubbish in the users hand. It can determine what the waste is and then tell the user in real time the most appropriate bin for disposal.

We interviewed Tuomas Lukka, former chief scientist at Zen robotics which invented a robot sorter for industrial waste.

In the extended interview video below we ask the following questions:

1, Can you summarise how a machine learning algorithm can categorise items?

2, When approaching the bins how long would it take to categorise an item of waste?

3, Is there enough power on a donated second hand tablet?

4, Would an Al system get confused by different coloured gloves?

5, Is one camera enough to be able to tell apart different objects?

Interview with Tuomas Lukka
- AI Expert

What is Machine Learning?

Machine learning is the area of AI where a computer independently learns how to complete a task, usually involving a set of data, without any human working to find the solution (1). For example machine learning could be used to identify spam emails from genuine emails. There are two general types of machine learning: unsupervised (used for finding patterns in unlabelled data) and supervised (learning how to predict the class of unlabelled data based on previous labelled examples) (2). Machine learning algorithms can take different approaches based on the type of data that is being analysed with different suitability for the task.

How can Supervised Machine Learning be Used to Identify Waste?

We are interested in the supervised model since our algorithm will be trained by providing it with images of different types of medical waste labelled with their name and correct bin of disposal. The algorithm will then come up with a set of complex parameters. These parameters will then be able to predict the type of item and which bin to dispose of it in on data sets which have previously not been seen (3).

What Existing Solutions are there?

AI algorithms have been used by Zen Robotics to sort industrial waste and rubble to reclaim resources (4). This shows object identification using machine learning in the context of sorting waste is feasible. We need to now train new algorithms to be able to recognise the different types of waste used in healthcare.

How does Our Idea Build Upon and Improve This?

Our idea focuses specifically on healthcare waste for the reasons highlighted. The strategy we have chosen focuses on educating people to sort waste at the site of disposal. This is because secondary sorting with robotics is inappropriate in the healthcare setting due to issues of cross contamination. In addition robotic sorting is more expensive (5) and therefore less suited as a global solution. The algorithm we have develop will be more appropriate for global implementation as highlighted here.

Reference List:

1. Brynjolfsson E, Mcafee A. Artificial intelligence, for real. Harvard business review. 2017;1:1-31.

2. Badillo S, Banfai B, Birzele F, Davydov II, Hutchinson L, Kam-Thong T, et al. An Introduction to Machine Learning. Clinical Pharmacology & Therapeutics. 2020;107(4):871-85.

3. Black JE, Kueper JK, Williamson TS. An introduction to machine learning for classification and prediction. Family Practice. 2022;40(1):200-4.

4. Lukka TJ, Tossavainen T, Kujala JV, Raiko T, editors. Zenrobotics recycler–robotic sorting using machine learning. Proceedings of the International Conference on Sensor-Based Sorting (SBS); 2014: Citeseer.