How to make sustainable food systems with AI

Startups use artificial intelligence (AI) to make food systems more sustainable. For example, Orbisk reduces food waste with a smart monitor. Professor Ioannis Athanasiadis wants to support these types of companies with a European testing and demonstration facility. They both present their views at F&A Next.

The startup Orbisk has developed a food waste monitoring system that catering companies and restaurants can place around their waste bin. A smart camera, in combination with a weighing scale, provides insight into their food waste. As a result, these companies can optimize procurement and processes in their kitchens and doing so increase sustainability and profitability.

‘AI can take over labour-intensive tasks of us’, says Orbisk founder Olaf van der Veen. ‘We use image recognition equipment above the waste bin to determine what goes wasted. Research shows that an average restaurant throws away 10,000 kilos of edible waste every year and that is also what we find in our data. On average, 70 percent of our food waste is overproduction and over-stocking and only 30 percent comes from oversized portions. Restaurants and caterers usually purchase too much because they are afraid of coming up short. Our monitor gives them insight into the surplus and allows them to purchase much more efficiently.’

AI is one of the themes of F&A Next, an agrifoodtech event taking place on May 22 and 23 in Omnia at the Wageningen Campus, where start-ups, scale-ups, investors, businesses, and researchers come together.

Neuraal netwerk

Orbisks waste scanner works with a neural network that has learned what food looks like, for instance sandwiches, peppers and pasta remains. Orbisks main work was training the neural network to recognise all kinds of food, says Van der Veen. The company also uses other data sources. For example, they predict the number of guests, but also do research on what they eat. ‘What people eat depends enormously on the weather, so the weather forecast is important. These are very complex datasets, so we need AI to recognise patterns in consumer behaviour.

Orbisk is growing like crazy. The startup, founded in 2018, has now sold six hundred food monitors in 33 countries and is growing by over 10% per month, says Van der Veen. About sixty people work there now.

I want businesses to share their datasets and algorithms, so they can learn from each other.
Ioannis Athanasiadis

70% savings

The food waste scanner can prevent 30% to 70% of food waste, says Orbisk. Its oldest customer, the Leiden University Medical Center, has had the food scanner for more than four years and structurally achieved 70% savings. Van der Veen also thinks he can make big savings in the catering industry. ‘A lot of food is offered for free. This degrades the value of food, causing a lot of food to be thrown away. There is a lot to be gained in such an environment.’ The waste scanner has recently also been installed on a cruise ship. ‘There is a lot to be gained there. The food on cruise ships is plentiful, but food waste is also becoming an issue there because of the negative perception and the costs.’


Wageningen AI professor Ioannis Athanasiadis thinks he can support startups like Orbisk in the development of their business. ‘Developments in AI are moving very quickly. Startups need support and facilities to test their prototypes and products in different conditions, so that they can develop and scale up their innovations. The EU has recognised the importance of this and has funded one of our programmes, called ‘Testing and Experimentation Facilities for Agrifood Innovation’. This is a European network of facilities to test AI solutions.’

When considering AI solutions for agriculture and horticulture, Athanasiadis is thinking of weeding and picking robots that do their work under different conditions in, for instance, The Netherlands, Austria and Spain, but he also thinks about datasets and digital products that allow these facilities to better perform their tests. ‘We aim to set up a digital infrastructure with other knowledge institutes in which we can pre-competitively test AI solutions, so that these innovations will become ready for practice more quickly and can scale up faster.’

Weeding robots

Companies that apply AI can learn a lot from each other, the professor thinks. ‘For instance, there are already various systems in use in the field of weeding robots. Most are focussed on a specific crop and specific environmental conditions. They use different data streams, algorithms and protocols that are often not interoperable. I would like companies to share their datasets and algorithms so that they can learn from each other. Sharing data in such a public facility for startups and small companies can accelerate their development process.’ This facility, AgrifoodTEF, is now developed.