“AI has the potential to revolutionize transportation"

ENAC is proud to welcome Alexandre Alahi, an Assistant Professor of Transportation Engineering. Alahi completed his PhD at EPFL and then spent five years at Stanford University as a Research Scientist and Postdoc before recently returning to Lausanne. His work on crowd behavior is appreciated by researchers around the world; for instance, he studied over 100 million people as they travelled through train stations – mainly in Switzerland – in order to identify their different walking patterns. Alahi’s research has played a key role in developing innovative methods that lay the groundwork for the future of transportation and mobility. These methods combine advanced technology like artificial intelligence, machine learning, computer vision and robotics with concepts borrowed from the social sciences.

What prompted you to become a researcher?

I always wanted to do something that was useful for society. And since I’ve also always been excited about innovation, research seemed like an obvious choice. When I was little I often wondered what new things we could invent to improve our quality of life. And then at university, I got interested in recent advancements in communication and IT systems. A few years later I became fascinated by the vast potential that artificial intelligence has to help us in our daily lives. After some experience working with companies, I decided to pursue my research in the academic world, where I have more freedom to take my work in more radical and ambitious directions.

Why did you decide to study transportation systems?

Throughout history, developments in transportation have always had a revolutionary impact, whether in terms of quality of life, human interaction, trade or the economy in general. We all use some form of transportation every day – when we go to work, buy groceries or visit a friend, for example. This field is hugely important for society and has been disrupted by scientific progress time and time again. We are now at the dawn of a new era where artificial intelligence (AI) stands to transform the industry. With AI, we can provide better and safer transportation systems and develop new services for the mobility impaired, such as blind people and the elderly.

What will your research focus on?

Our laboratory will develop a new kind of AI for use in transportation and mobility applications. Until now, AI has been used to solve clearly-defined problems with established rules, like playing chess or go, improving the results of online searches, filtering spam, suggesting movies and, more recently, recognizing images. But the problems we face in transportation are not so clearly defined, and the rules are often abstract and stem mostly from social conventions. The choices we make in travelling from one place to another are typically guided by common sense or the respect, consideration and empathy we have for others. So even though these rules are widely acknowledged in society, they are informal and can vary from one culture to the next. Our laboratory will seek to develop smart transportation systems that can understand and adapt to human interaction. For instance, our systems could be used to create a robot capable of assisting an elderly person at home while behaving in a way that is both socially acceptable and appropriate for cohabitation. Another application is self-driving vehicles, which could soon become a part of our urban landscape – as we innovate in this area, we shouldn’t forget ethical considerations, as they will be essential for designing autonomous transportation systems that will come into contact with human beings.

In other words, you will develop machines that understand human behavior.

When we walk through a crowd of people, our body language helps others predict what we will do next. It’s something we do naturally – most of the time we don’t even realize it. Today’s robots don’t have this awareness or the capacity to read others or foresee behavior. What’s more, they can’t follow the same social conventions we do. If two people are standing near each other talking, for example, and a robot calculates that there is enough space between them to pass through, it’ll do just that – interrupting their conversation. But a human would walk around them out of respect. We want to develop methods and algorithms that can teach machines such behaviors using empirical data, without having to manually program all of the rules.

How?

Instead of listing out the rules, we will let our machines automatically learn how raw input data is correlated with a given task or objective. That’s what’s called machine learning.We are in the process of developing autonomous platforms that can move around campus in a smart way, delivering objects and guiding visitors, but without disturbing students as they go to and from class. That entails collecting data on how people move in crowds and modeling their interactions so we can develop methods for predicting their behavior and improving our robots’ navigation systems.

How are you collecting the data?

Today, everything we do can be recorded digitally – by cameras, heat sensors or proximity sensors. But this data is in a very raw format and needs to be processed with complicated algorithms to obtain the exact information required for a given task. For example, scientists can use camera images to determine an individual’s coordinates in space and study how he interacts with his surroundings.

So in the end, it’s all about how we use space.

Exactly. That’s why our laboratory is part of EPFL’s Civil Engineering department and ENAC. The sensors and algorithms we use give intelligence and awareness to space, bringing a new aspect to the design and construction of the spaces of tomorrow.