Giving vision to robots with Pedro F. Proença
Updated: May 12, 2021
Pedro, originally from Portugal, moved to Guildford four years ago to pursue a PhD in Computer vision at the University of Surrey. Almost finishing his degree, he shares his journey in research and his opinions on the future of artificial intelligence.
Pedro may have his doctorate degree now but he has not always been a good student. As a child, he struggled with paying attention in class and focusing on homework. He said, “my grades were terrible when I was a child, but as I grew up I started to understand that I was good at maths and Science”. Eventually, he became a high-achieving student in high school and ended up pursuing Computer Science and Telecommunications at university.
As an undergraduate student, he soon realised that he was more interested in computer science and programming than in telecommunications. Later, during an artificial intelligence (AI) module, he discovered computer vision and AI. He told me, “I had an instant attraction to these areas because there is so much potential”. Therefore, Pedro decided to specialise in Computer vision.
Computer vision is a field that deals with how computers take information from videos or digital images. Applications for computer vision are endless. It can be used to give robots awareness of their surroundings, for surveillance drones that recognise people, to build a software that automatically identifies cancer cells in a microscopy image, for augmented reality glasses that point you to the right direction, etc. If you imagine a technological future, computer vision will undoubtedly be a big part of it.
Pedro’s master thesis was in object recognition, developing methods that allow computers to identify and distinguish objects. For us, it is simple and straightforward to identify a chair, for example, but how do you explain what a chair is to a computer? Chairs can have different colours, materials, and designs. If a computer is taught that chairs have four legs, then this will affect its ability to identify 3-legged-chairs. Thus, although for us it is easy to identify an object, giving a computer the ability to properly recognise and classify an object is a challenging task. Pedro worked to solve this type of problem, using machine learning and 3D cameras.
Machine learning uses pattern recognition and statistics to teach computers. If you give a computer enough pictures of chairs and tell it to prioritise having legs over being yellow, for example, then the computer eventually learns what a chair is. Using this approach, Pedro taught a computer how to identify an apple, a cereal box, a cap, and other everyday objects.
After the master's, Pedro realised he wanted to stay in academia to continue his research, “I found out that what I wanted to do is research, so a PhD is the next step”. Thus, he then applied for a PhD in robotics, using artificial intelligence to solve problems in the nuclear industry: “I was always more focused in something between theory and practice I found that this PhD would be a good compromise”.
Initially, his PhD project, funded by Sellafield Ltd (a nuclear plant), aimed to assist workers at the nuclear plant using augmented reality. Augmented reality is an interactive experience where virtual content is added to our real-world environment. Pedro’s initial goal was to use this technology and combine it with his background in computer vision, in order to help people navigate through space. For example, imagine there is a spill in a nuclear plant. A nuclear worker needs to rapidly leave the building but does not know the way. If wearing smart glasses, the glasses could direct the worker to the exit by drawing virtual arrows that point the right way. This can be very useful not only in hazardous environments or dangerous situation but is also applicable to our everyday life.
However, as with all PhD projects Pedro’s project did not go as planned. He told me, “It drifted a little bit more to the theory side, it became more mathematical than I originally thought”. To compensate for knowledge gaps in the field, he developed new methods of self-localization and navigation. Pedro explained, “you need to know where you are to know where you’re going. My work is on how to use moving cameras to estimate their location using only visual information. If a camera moves in an environment, my task is to estimate where the camera moved to”. His project turned out to have a broader application than he originally thought: “The main goal of this project is to support mobile robotics”, he says. His work is mainly a contribution to the field which is building a future where robots are able to move and be aware of their location.
Pedro intends to keep doing research in the field, since “AI can be very helpful if used wisely”. Pedro believes that AI technology is developing too fast and that we need to know the impacts of AI in society and develop a code of ethics. “It depends on the hands-on which you put it, but it definitely has the potential to help people. There are certain tasks that are too dangerous or too dull to give it to people”, he explained. His research can certainly contribute to a future where robots will help society.
With many scientific papers already published, Pedro admits the beginning of his PhD was not great and that he often felt lost. But now he realises that it is rewarding to see how much he learned and achieved in the last three years.
To future PhD students, Pedro advises: “Choose your area and supervisor wisely. Try to understand if the area is promising, and get someone that is supportive but also good in the field. Do not rush things. Think about it before you dive into a PhD.”
You can follow Pedro’s research on twitter (@pe_proenza) and on his website.
This piece was published in the online version of The Stag, University of Surrey's student lead magazine.