Robovision offers a path to extra revenue streams out of your data. We achieve this with an elastically scalable software stack on top of our AI engine (RVAI, request a demo). This enables us to process hundreds of HD streams in parallel. At the other extreme we are developing embedded fpga based deep learning solutions, for customers that require a low energy and low cost solution in their machinery.

Robovision develops fully scalable deep learning solutions, for big (for instance city camera networks) and small data (IOT). We  excel at reducing training times of large datasets, and industrializing academic state of the art knowledge, while designing the right deep architecture. Our product is a computationally optimized setup, running RVAI.

An example case: the parallel processing of TV streams

We generate metadata on top of media streams. Which T-shirt is the main character wearing, and where can we buy this item?

On the one hand databases of Zalando / Amazon need to be scraped and tagged, so we can search for the right item and generate overlays, and one can buy that particular item. On the other hand we need to detect objects and clothing in media streams, and link it with the offerings available in our database.

In most current use cases the data are visual. But we can apply our deep learning technology to other data streams like pure text (product descriptions, cyber security) or even a customer database just as well. Do not hesitate to contact us if you are looking for a deep learning solution.


Deep learning enables companies and society to achieve interpretation of complex videos and pictures by using vast amounts of virtual neurons stacked in layers on top of each other (check this video). The essence of this technology is that the optimal feature selection is no longer done by people (vision engineers). The deep neural architecture groups and aggregates the right information clusters, to generate an optimal decision process. In doing so, it creates its own abstractions, similar to the abstractions our brain is making for the visual processing of information (see figure below).

We at Robovision design such architectures, and use state of the art knowledge (sometimes even a deep learning Arxiv paper that is only a couple weeks old) to get the right results. Check our blog for the latest in deep learning!

Why choose Robovision for deep learning applications?

▪     We have an extensive track record. Often success is about pragmatic choices, having the right people on the team and a strong passion for the target. We have the ideal team size to share expertise and focus.

▪     We develop the right power tools to process your data: user friendly data labelling, cloud deployment with user and session management, parallelisation on a large series of GPUs.

▪     We have tagging teams that can start tagging your data today, and thanks to our developed pipeline you will see tangible results very quickly.

▪     Robovision is a pioneer in applying deep learning for the processing of large amounts of (visual) data.


Together we can turn your data into gold.



In some cases an opportunity proves so interesting, that we don’t want to be limited by nifty algorithms, silicon and GPUs. Than we decide to build AI robots.


Because the manipulation of natural objects (the main subject of our AI Robotics) is an art. A fresh Chrysanthemum stem is very fragile. If you squeeze it too hard, it will die. Our AI Robots (over 160 working worldwide already) plant 4 million flowers a day, generating terabytes of data. Each optimisation in this chain is causing big profits. Nifty mechanical and electrical tricks can go hand in hand with AI techniques. If a robot makes a failure, it can learn from its mistake with no human interaction, and pass on to the next stem as quickly as possible (see the slow motion video below).

Together with Robovision Integrated Solutions (R.I.S. nv) we can efficiently prototype and combine advanced AI techniques with robots and small conveyers. R.I.S. is a partner company focusing on the mechanical side of AI robots.

Together with Robovision Integrated Solutions (R.I.S. nv) we can efficiently prototype and combine advanced AI techniques with robots and (small conveyers). R.I.S. is a partner company focusing on the mechanical side of AI robots.

Since 2016 Walt has been working at the Audi factory in Brussels. He helps the human factory workers to assemble the Audi A1 car bodies.

Walt is surrounded by hundreds of other robots. The big difference is that they all operate in cages, while Walt sits, free as a bird, amid his human co workers. He is what is called a compliant collaborative robot, or cobot in short.

He recognises his human colleagues when he sees them, greets them with their name, and looks at their hands so they can give him gesture commands. Besides his friendly face, Walt can also display icons or short animations, to clearly state his intentions.

The face is the single most important instrument for direct communication between persons. Persons like in humans, now also like in robots.

While the human face has a lot of very practical functions, most of its muscular capacity is focused on communication with other humans. The human face is capable of many forms of direct, indirect, or even involuntary communication. That is why Walt was equipped with a versatile face.

To get to this result, Robovision worked together with character designer Jan De Coster (Slightly Overdone,). He has many years of experience designing social robot characters, who are on display in exhibitions all over the world.  The chromed ring around Walt‘s face is an obvious reference to the Audi logo.