Solving for Urban Air Travel: A Q&A with François Sillion, Director of Uber ATCP

Bolstering Uber’s position at the leading edge of transportation technology, the new Advanced Technologies Center in Paris (ATCP) supports the development of Uber Air, our effort to add a third dimension to our platform using flying vehicles. 

Leading ATCP is François Sillion, who recently served as director of artificial intelligence at Inria, France’s national research institute for the digital sciences. François earned his Ph.D. in Physics and Computer Graphics from the Ecole Normale Supérieure and has enjoyed a long career as a researcher in computer graphics and machine learning.

In the future, Uber Air will offer aerial ridesharing using a network of all-electric, vertical takeoff and landing (VTOL) aircraft powered by distributed electric propulsion. This vision requires building artificial intelligence along with optimization and airspace management systems to support Uber Air at scale. 

ATCP joins a portfolio of research offices Uber uses to explore technologies that will let us offer the most efficient transportation solutions at a global scale. In San Francisco, Uber AI pushes the limits of artificial intelligence and connects the latest advances in machine learning to all aspects of our business, while Advanced Technologies Group offices in Pittsburgh, Toronto, and San Francisco invent the future of safe, reliable, and affordable self-driving transportation. 

To learn more about our vision behind ATCP, we sat down with François:

What is the role of the Advanced Technology Center in Paris (ATCP)?

Our primary goal involves maintaining strong relationships with academic research institutes in Europe while engaging in our own research on topics relevant for Uber’s applications, with a particular focus on Uber Air, the initiative to develop urban air transport.

Currently, ATCP is hiring research scientists to produce new knowledge in this space and cultivate strong relationships with our academic and industry partners. 

The topics we research are fairly broad but have initially focused on software, such as artificial intelligence for decision-making and optimization. One big topic for us is the safety of software systems, which is particularly crucial for anything related to aviation. Safety, of course, applies to the aircraft itself, but also to the broader picture of urban aviation. You see, when you start operating aircraft in large numbers there will be a need for new algorithms and systems to scale air traffic management. 

What lead you to join Uber as the head of ATCP?

I’ve worked in research for 30 years, including 20 years at Inria, the French research institute for digital sciences. I started out researching computer graphics and vision and then moved on to machine learning. Over the last 10 years, I gradually moved into research management. I managed about 500 people at Inria, then became its scientific director, covering all of our work in computer science and mathematics. 

I’m always excited by research, and I really want to see it have an impact. Last year I was ready for something different and wanted to work at a company that was advancing the state of the art while making a difference in people’s lives. When I looked at Uber, I found this really exciting project called Uber Air, and there were a few aspects that interested me. First of all, it’s a game-changer in that it’s something that hasn’t existed before and could have a huge impact on transportation. Another aspect is that when I talked to the people at Uber, I realized they had a 360-degree vision of the challenges. This wasn’t a naive initiative of just building a flying car; the people involved are identifying all of the tough problems, from what type of batteries the aircraft will need down to whether it would be too noisy for public spaces. I liked Uber’s realistic approach and am excited about the possibilities. 

How will the ATCP staff divide its time between doing original research and collaborating with academic institutions?

There won’t be such a clear division, as the research we will be performing has to be connected to universities and institutes. Nowadays, research is seldom done by just a couple of people. Connections are important because we’re building on each other’s work, especially in terms of software. The open source movement is a great example of how software depends on a community rather than singular developers.

With research, it’s difficult to differentiate between the internal and collaborative, but I think maybe half of our effort would be on producing new results. And I don’t like to suggest a distinction between fundamental and applied research. I think all good research should find applications in the real world. We won’t just be devising algorithms in front of a whiteboard– we need to show that what we create can run at scale using real-world data. Any research we do must be meaningful.

Since ATCP will be focused on research areas important to Uber Air, all of our work will be about enabling the future of urban air transport, and our office will be looking for work arising from universities that could have an impact down the road. It will also be important for members of the Uber Air team, who have a clear idea of the product roadmap, to spend time with the research teams at ATCP.

What are the main technical challenges that ATCP will be solving?

One of the first things ATCP will be looking at is optimization. Of course, many teams are already working to optimize the Uber platform, but we will be looking at problems specific to Uber Air. One of them, for instance, is infrastructure. For instance, where should we place the skyports from which these new aircraft will launch so as to serve the greatest number of people and also create the safest and most seamless operational conditions?

There are many geographical and geometrical constraints to consider. For example, although our aircraft will be electric, they will still generate some degree of noise. When finding an optimal place for a skyport in a city, we will need to take noise, among other factors, into account, to provide the best utility and comfort for people who use the service and those who work or live in the area. Then there are a city’s traffic patterns to consider. These are some of the fairly static factors we need to understand so that we can place skyports in an optimal way.

Then there are more changeable factors, such as the weather, which plays a large role for anything involving air transportation. For the kind of aircraft we are proposing, which run on batteries, the weather will directly affect energy consumption. If a strong wind is coming from the West, for example, aircraft traveling against that wind will use more energy, so we may limit the number of passengers in each aircraft, reducing weight and saving energy. Things such as wind make air travel a large scale multifactorial optimization problem that is not needed for Uber’s current products.

To operate efficiently and at scale, we will likely want these aircraft to follow each other fairly closely. That will require high-quality sensors and some degree of automation. Machine learning could be crucial in helping these systems make informed decisions to ensure safe travel. The decisions made by any machine learning systems we use must be transparent, so we know why these decisions are being made and can trust them.

How will AI play a role in Uber Air?

One example involves predicting ETAs, which is already something we do at Uber for the ridesharing business. When we start dealing with aircraft, we have a multimodal trip, where the first and last mile will involve a rider taking a car to the skyport, then getting picked up by a car after they land and being taken to their final destination. For that model, we need to calculate ETAs for when the rider will be dropped off at the skyport, coordinate with an aircraft flight plan to ensure that the skyport pick-up and drop-off logistics function seamlessly, then have an ETA calculated so a car will be waiting for that rider at the landing skyport.

As another example, in optimizing the location of skyports, we are analyzing things like population movement patterns and weather in a particular city or area. That’s a time-consuming and complicated process, especially if we are thinking of scaling Uber Air to hundreds of cities. Using AI, we can build models to help extrapolate to other cities, using their local data to get predictions of optimal skyport locations. As we explore skyport locations, we will probably find there are a number of city types, such as dense, European-style cities and sprawling cities common in the U.S. This type of categorization can help when we use AI to suggest potential skyport locations.

Finally, Uber Air vehicles will be piloted initially, but we will prepare the evolution towards remote supervision and self-piloted vehicles. This research has similarities with Uber ATG’s self-driving car work, as deep learning for sensor acquisition and fusion are relevant in both domains. 

ATCP currently looking for top engineering and research talent. If developing innovative, urban air transportation for cities excites you, apply for a role on our team!