Can you introduce yourself in a few words?
My name is Raoul de Charette, and I am researching computer vision, also known as vision algorithms. This is to say that I analyze images in order to get a sense of what is behind animation, object recognition, and getting an understanding of the scene, particularly for mobile robots like autonomous cars. The idea is when you have a mobile robot like a car, there are a number of sensors inside, like output color,texture or geometry, that can help us understand which category of scene the machine is “seeing,” for example, if the environment urban or rural, rainy or clear. This has been the overarching theme of my research over the years.
However, the focus of my research was not always in automatic driving. I have a bit of an unusual background. Initially, my field of study had nothing to do with engineering, which is where I am now. I studied image design in a graphic arts school where I learned 3D animation and film making. The common point between all my projects, both then and now, is image--how to create, analyze, or design images for practical applications.
At the time I was interviewed for Knock Knock Doc, I was studying the detection of blurry drops by intelligent machines, so essentially, how raindrops appear to machines on a lense or windshield when it is focused on the the environment beyond the glass. On this same line, I also researched how traffic lights can be detected—how machines can identify a particular color, shape, and position in enough specificity to be able to differentiate it from other lights or colors on the road.
What have you been working on since you completed your thesis?
Once I finished my thesis in France, I worked abroad in the US at Carnegie Mellon University, and eventually I returned to Paris to join the same robotics lab at MINES ParisTech only to go abroad again. While in Paris, I continued to researching computer vision. I’m currently working again on new vision algorithms to use in intelligent cars, and part of my work still focus on degraded weather, so blurry drops and beyond.
The key to detecting something like a blurry drop is understanding what vision truly is. What you are seeing, therefore what a machine sees, is being modified by the atmosphere; you have a distortion of the world you’re working in. So what exactly is vision, and how can we translate it in a way digestible by machines? This question is still completely open in research, and we can see it in how advancements in autonomous cars are being described in the news. Everything that you see and are going to see about mobile robots will be in ideal weather conditions. In other words, they consider the atmosphere to be transparent. As soon as they’re faced with larger molecules, there are visual artefacts such as semi-transparent streaks with raindrops, or like bright opaque spots with snowflakes, the vehicles no longer function properly. There is very few ongoing work about how to solve computer vision in the case of degraded weather, and that’s one of the subjects we work on in my lab.
What do you like to do in your free time?
I have traveled a lot since my thesis; I love to go backpacking. Going to southeast Asia was my first long distance travels, and it continues to be one of my favorite experiences. The reason is because the culture is incredibly different. The way of living is completely opposite to France and other Western cultures like that of the US, South America, and even Australia for example. It’s a really cool thing—you don’t have any Western people there, so you become completely lost in another culture with different meanings for everything you do.
What was one of the greatest skills you learned from your PhD experience excluding academic knowledge?
Right now, I’m supervising a team of students and learned a great deal in terms of pedagogy. I have found that there are two ways of supervising: 1. to impose your own way of doing onto those in your team, or 2. to adjust to people. The latter is more positively inclusive. What I think is really important in research is to remain open to completely new approaches, completely new ways of seeing a problem, which is very difficult. Sometimes we have one vision of seeing the problem and get stuck in that direction which, ultimately can keep your research from moving forwards.
I recommend to my students to attend presentations in physics, optics, chemistry, anything, to remain open to other fields. Travelling is also a good option to remain open. It’s a running joke at the office that this is how I justify all my traveling to my colleagues. But in all seriousness, even looking for and reading research of these other fields can bring you some very interesting ideas. In fact, the most interesting ideas are often those that are unusual and completely unexpected.
Watch Raoul's Knock Knock Doc video from 5 years ago here!