‘I hope it will inspire others to build on our work’

Thijs Jansen, a Master's student in the Smart Systems Engineering (SSE) programme, has been researching an innovative project that has the potential to transform healthcare. The project focuses on the use of robots in buildings where privacy is an issue and cameras are not allowed, such as hospitals. Thijs and his fellow students came up with a creative solution using LiDAR technology, which allows them to detect objects without cameras.

‘Our project was about using robots in privacy-sensitive spaces. The problem is that in some buildings you are not allowed to use cameras for privacy reasons, but robots usually need cameras to navigate. So we had to find a way to detect objects without cameras. We decided to use LiDAR: a rotating laser that scans the environment. We used a 2D LiDAR because it is cheaper and requires less computing power than the 3D variant.

The challenge was to apply AI to this 2D LiDAR data. We first tried to write an algorithm ourselves, but it was difficult. A fellow student experimented with automatic machine learning, but that didn't work either. It proved difficult to make something recognisable with just 2D points.

Then I had an idea. I saw a YouTube video about AI and image recognition and thought: why don't we just take a picture of the points that the LiDAR sees? Instead of writing a complex algorithm for point coordinates, we could use existing image recognition on these 'photos'. It might be an extra step and require a bit more computing power, but the technology already existed.

We decided to go down this route. We created our own dataset of images of different objects at different distances and resolutions. The results were surprisingly good. We were able to extend the dataset using techniques such as data augmentation, which we had learnt during our studies. Building the dataset took a lot of time. We spent hours at the computer manually selecting and labelling recordings. But the results were worth it. We were even able to run the model on the robot itself, a Raspberry Pi. There were some technical challenges with compatibility, but eventually the model was able to run once per second. This was fast enough for a slow moving robot.

The robot was aware of its surroundings and could distinguish between a wall, a door and a table. It was great to see that our idea actually worked in practice. Our supervisor, Filip, suggested that we turn it into a publication. We submitted our research to a conference and are now waiting to see if it will be accepted.

It's exciting and a great honour. I didn't expect our idea to be so innovative. When our research is published, I hope it will inspire others to build on our work. In doing so, I hope to help fill the gaps in research and provide better solutions for small-scale automation in environments where privacy is important.

The theories, procedures and programming skills I have learnt have been incredibly valuable. School cannot teach you everything, but it can give you the building blocks to build on. That way you can make the most of your own creativity and intellect to create something innovative.'