Follow us:

Panorama Newsletter


The TinyML Summit 2023 Research Symposium (March 27, 2023, San Francisco) is the primary annual gathering of senior-level technical experts and decision-makers representing the fast-growing global tinyML community.


Tecniplast has developed a tiny, embedded camera project for laboratory animal welfare management that uses local monitoring and end-to-end system application. This project has several benefits, such as scaling monitoring operations, improving sterility, reducing costs, and introducing a new way of machine learning services. Tecniplast has partnered with STMicroelectronics to develop machine learning for hardware design and deployed tinyAI intelligence onto embedded boards.

Marco Garzola, the Digital Innovation Manager of Tecniplast, spoke at the 2023 TinyML Summit in San Francisco, discussing Tecniplast's interest in TinyML and providing details about the co-development work done with STMicroelectronics.

Garzola has a degree in Telecommunication Engineering with Signal Processing specialization from Politecnico of Milan and technical responsibility for new products and PoCs (Sw/Fw/Hw) focusing on digital transformation (from Fw to cloud computing) at Tecniplast. He has published two scientific papers in relevant IEEE conferences and holds two patents.

Marco, why were you at the TinyML Summit?

The opportunity I had to address the TinyML summit at San Francisco Stemmed principally from our collaboration with STMicroelectronics: Tecniplast was invited as a speaker following research carried out with their AI team. Tecniplast’s participation in an event of this caliber, unique at world level, was much appreciated as an example of concrete use and impact on the market.

During the course of 2020 we began as an Innovation Team to develop POC (Proof of Concept) based on vision systems (cameras) to complement the data of our current DVC® system.

The original purpose was to carry out technological scouting in order to understand any eventual pros and cons of a widely diffused solution based on this particular technology. At the time the idea was to forward the data from the cameras to the cloud and thereafter to analyse them. We very soon realized that such a solution scaled neither as regard costs nor performance.

I remember that at a conference I had chanced to meet Danilo Pau (IEEE and ST Fellow) who was presenting on behalf of ST certain AI-based solutions. He stated that such solutions could be achieved on micro controls that were much like what we ourselves were developing.

My curiousity aroused, I sought to understand more and thanks to a series of opportunities we managed to show them our particular use. Danilo and his team of students were attracted to it as a challenge. And so it was that Danilo decided to help us in developing certain neural networks on micro controls that equipped the DVC®.

In short order we managed to get exciting results, which we decided to corroborate through a number of scientific articles published at IEEE conferences.

As of today, our innovative Tecniplast team is almost completely dedicated to AI and ML issues. As our dealings with technology partners intensify, I am confident that we will be able in the near future to come out with and validate scalable innovative solutions and thus fully meet our customers’ needs.

Can you briefly describe the application of machine learning in the DVC®?

The prototype we are currently testing involves the use of two low-resolution cameras which monitor 24/7 home cage activities.

We are now, thanks to machine learning, able to pick up bottle presence, food level and cage presence with great precision – and all this at local level without having to forward great quantities of data to the cloud.

We are also testing new types of algorhythms more linked to animal activity with very interesting results.

Our research doesn’t stop just with this type of sensor: we’ve got a lot more up our sleeves. But let this much suffice for the moment.

Very interesting! How do you see the machine learning application in the LAS?

The world is evolving very rapidly and when you take part in such conferences as tinyML Summit you realise we’re on the threshold of a new technological evolution. Over these last few years there’s been a lot of talk about ML and AI, but much of the development in this field has taken place in the cloud. This entails an outlay of energy/capital that not all markets can accept, a typical case being LAS.

This market, still highly paper-addicted, is generally loth to take on great innovations or very slow to do so.

Problems connected to costs, data privacy are surely issues that have held back certain types of technological evolution over the years. However, in my view, the new ML solutions, which are ever more embedded in devices, will help this area to evolve more rapidly.