Back in 2020, while doing my PhD at EPFL I met a blind person who was doing a FaceTime to a friend. His friend was providing his navigation instructions like:
My Master was specialized in computer vision, my PhD in speech processing, I connected everything and I reached out to the ophthalmic hospital in Lausanne. Sat for 2 hours with them, and presented some ideas. We converged to the idea that the future of mobility in low vision should be camera-based.
In August 2020, I joined the hackathon ICC organized by my lab. Won 7k $ and bought the first hardware. Involved the first 20 users, met my co-founder, and we started building nights and weekends for a whole year.
Late 2021, we raised 600k in a round led by Serpentine Ventures and Venture Kick. The first feedback from testers was actually really good, but the path to market was still very long. We conducted a study at the ophthalmic hospital to benchmark our first results with SOTA, and went back to building:
Why a harness?
In 2022, middle of COVID, I dropped out of my PhD and went all-in on biped. My co-founder and I hired a team of 2 robotics engineers, 1 full stack engineer, and found partners for hardware engineering, industrial design and audio design.
The team focused on bringing the first feature to life: a robust obstacle detection system that leverages the ultra wide field of view of the multi-camera setup we created. First versions looked like this:
In 2023, after 250 tests in 10 countries, the picture became clearer. It should not only be an obstacle detection system. At the age of physical AI, it needs to do more. The device became NOA that day: Navigation, Obstacle detection and AI. The 3 features of the device, the 3 types of information that the person was providing on FaceTime. Simple but to get there, we really had to get back to work.
I also focused on clearing regulatory compliance for the software part and later for the hardware. We cleared all regulatory compliance as a class 1 medical device in Europe.
The device got to an MVP stage, ready to market. This is the demo video:
The hardware got there too, and started to really look cool:
The software platform also became a lot more advanced. The device runs a combination of Vision Language Models, local object detection, semantic segmentation, obstacle detection, GPS navigation, and generates speech output for the user.
We took this device to semi-industrial stage. It's now assembled by batches of 50 units in Switzerland with our partners.
And most importantly, used by people in 15+ countries!
Since then, we've opened 17 countries with over 25 distributors & helps users walk thousands of kilometers independently.
