Between Frontiers - Robotics and AI
Meet Salah Missri
Computer Vision Technical Lead @ Pix4D
Canton of Vaud, Switzerland
Salah Missri studied in a French school in Morocco before joining Switzerland after high school to study robotics at EPFL.
Here Salah GitHub Page called SyrianSpock with some of his projects
How you describe your first encounter with Robotics?
My first time it was reading Asimov's books the cycle of robots. A fun book with short stories that I highly recommended.
Then I used to play a lot of Lego and I received a lot to make a robot. So that got me thinking to study engineering.
At age 16, I got the Mindstorms kit and that made me fall in love with robots.
How did you become a Robotics software engineer?
I studied at EPFL Switzerland from 2011 to 2017 and specialized in robotics.
I joined a robotics club CVRA.ch in 2014 because my studies were still very theoretical IMO.
During my master's I took a 9-month break to work on software as a consultant.
Then I joined a computer vision startup for my master's thesis. I stayed there since 2016.
I specialize in computer vision currently. In robotics, it's often called the perception problem and it's one of the hardest.
It's challenging but kind of "solved" to make complex shapes move despite all the ongoing research on soft robots and batteries.
A lot of work remains in reading sensory input like cameras and interpret what is around in order to act and move accordingly.
My love of mathematics and fascination for perception led me into computer vision.
Software engineering is still about problem-solving so the engineering dimension is still important IMO.
I think a lot of software engineering is learned outside school (test-driven development, continuous integration, refactoring practices).
How do you see Robotics could help society?
A lot of the focus is on helping elders because we have a growing old population in Europe and Japan and our health system won't be able to handle all sadly.
Currently, it's not a threat, and I hope humanity is good enough to stay away from dangerous applications of robotics.
Killer robots are a threat but so far we hopefully stay away from those applications.
Mention any computer vision projects you have been involved in?
Reading of energy counter in real-time was my first exposure to computer vision.
We used a Convolutional neural network trained on a PC and deployed on the microcontroller to process images taken by the device in real-time of the energy meter and detect the digits on it and send the results to the cloud for monitoring.
I currently work on Structure from motion: the science of extracting 3D maps from 2D images.
Essentially, our software currently is making 3D maps you can use for measurement and survey out of 2D images taken by phone or drone.
How do you see the future of Robotics in 2020?
Bright. To sum it up.
I think automation in manufacturing will increase as we realize with this virus outbreak how much we save while keeping factories close to us, instead of going to China or other cheap labor countries.
I'm not against globalization but I'm keener on the idea of a distributed system.
It has shown its robustness in software. So more local production. More autocracy. Federal states are more resilient. With robotics, the cost of local production becomes cheap and with low manual labor that you can produce anywhere.
I think with the Coronavirus outbreak we'll think more and more about local production.
What role models been your inspiration to get into AI studies?
AI is just a solution, it's still software. So you solve software problems with that.
I'm also a big fan of sci-fi and some inspiration comes from Asimov characters.
How do you see AI could help society?
I always see AI as a software solution, it a tool. Since new discoveries in the field make many problems solvable we can solve many problems in real life.
Like we can count crops, detect if they're good or bad and thus improve our yield over time.
I see a lot of applications in medicine as well where books or expert knowledge can go into AI software and the doctor can act as horizontal expert asking the disease AI for diagnosis then the medication AI about the medication then through his knowledge determine if the suggested medication is appropriate in light of the patient case and history.
What we lack IMO it's the expression of uncertainty in our results, so our decisions are binary when they should be nuanced.