We are going to have to wait a few years, perhaps even a decade, to see self-driving cars on our roads. But in the meantime, the sector is developing rapidly, thanks not least to the researchers who are studying the implications of this new mobility model for the current transport system. One of them, Marco Pavone who, as head of the Autonomous Systems Lab and co-director of the Center for Automotive Research, both at Stanford University, is dealing with various critical issues regarding this technology, from how algorithms learn to the debate on self-driving vehicle safety.
At what stage is the development of self-driving vehicles?
This very complex technological system is made up, in turn, of many different technologies that must work together. There are – it’s true –various applications at the advanced stages, starting with the assisted driving software and sensors already in use by the automotive industry. We must not forget, then, that there are some vehicles already operative in special contexts: construction sites, seaports and industrial plants for example, where the complexity of the interaction between humans and machinery is much more limited than it would be on a roadway and so the learning process for automated vehicles is simplified.
In general, I’m quite optimistic, and I believe that in a few years’ time, the first self-driving vehicles could become available on a large scale. We’re going to have to wait at least a decade though for this technology to become pervasive.
We have to look at the development of these vehicles as a continuum. Take aviation, for example: when the Wright brothers flew the first airplane, they had no idea that one day a plane would easily cover the distance from Europe to the United States. The self-driving field is the same thing, although the process is much speedier.
What are some of the problematic aspects of this technology?
The problematics are both technological and systemic in nature; that is, regarding the vehicles’ integration into the mobility system. Fundamental to the technological aspects is how the machines interact with human beings. Self-driving vehicles have to be able to use probability to predict the behavior of another human being – a driver, pedestrian, cyclist – and to use that data to make decisions and anticipate the actions of others. When we are driving, we are constantly negotiating our actions with the other drivers on the road. For a robot, this is a very complex task that requires advanced artificial intelligence decision making systems.
If by intelligence we mean the ability to generalize our behavior even in unknown situations, we have to say that machines are still very short on intelligence. We are able to assess a person with a driving test of just a few minutes, but a self-driving car needs millions of kilometers of experience. This is because machines are not good at generalizing their behavior. We have been working at my Stanford laboratory on algorithms capable of predicting human behavior using probability and learning through experience.
When will it be possible to consider self-driving vehicles safe?
The concept of the safety of autonomous vehicles is a topic of debate and we are a long way from consensus. There are many issues to be confronted by every part of the mobility ecosystem. Can we consider an autonomous care safe if it has an accident rate lower than a car driven by a human? How many kilometers does it need in order to learn, and under what traffic and weather conditions?
And let’s not forget that self-driving vehicles change in each local setting and that, therefore, they need to be tested in every environment and driven thousands of kilometers in every city where they will be in use. It is a very complicated process. The advantage is that, as opposed to humans, algorithms are capable of immediately sharing what they have learned; thus, the knowledge of one can easily be extended to all the others in circulation.
Do self-driving vehicles have a future in Italy, a country with a great automotive history?
The opportunities are enormous and I believe that Italy can meet the research challenge. I am working with the Milan Politecnico on designing a course analogous to the one I teach at Stanford on autonomous technologies. I think that once the first young people are trained that could trigger a virtuous circle.
If we look at what’s happening in Silicon Valley, where the number of firms associated with the automotive industry has grown exponentially over the past five years, we see that there is a lot of concentration on software. With the exception of Tesla, most firms are developing the technological part, while production of the actual vehicle is left to traditional automakers. This does not mean, however, that the traditional automotive industry will not have to undergo some deep changes.
In a world moving increasingly toward vehicle sharing, it remains possible that demand will drop below the current level. Another issue is competition from countries where assembly is cheaper due to lower labor costs. I think, for this as well, that advanced economies such as Italy must stay decidedly concentrated on developing software. The transportation sector, which has remained substantially stable for nearly a century, is in search of a new, high yield business model capable of making all the technologies work together: from shared mobility, to self-driving to electric vehicles.
What will be the fate of sports cars in a world governed by algorithms?
Those who can afford a “supercar” will continue to buy them and want to drive them, with the possible assistance of evolved technological systems. Tin fact, the economic data indicate that the number of wealthy people in the world wanting to purchase luxury autos is on the rise. In the end what happens will be what happened to horses, a means of transportation for everyone that became a sport for enthusiasts only.
Marco Pavone is Associate Professor of Aeronautics and Astronautics at Stanford University, where he is head of the Autonomous Systems Lab. Also at Stanford he do-directs the Center for Automotive Research. Before becoming a professor at Stanford in 2012, he worked on a robotics team as a Research Technologist at the NASA Jet Propulsion Laboratory.