Emilio Frazzoli is fascinated by the unwritten rules of the road. His accomplished career is a testament to this: in both academia and industry he has focused on autonomous vehicles, transportation networks, and the theory and algorithms behind them — winning many awards for his work along the way, including the IEEE Kiyo Tomiyasu Award and the IEEE George S. Axelby Award — prestigious honors in the field.
In the latest of his successes, the startup Emilio co-founded in 2013, nuTonomy, was acquired last year by industry powerhouse Aptiv (formerly Delphi Automotive). Like Uber, Tesla, and Google, nuTonomy has been working to create driverless cars that are safe, economical, environmentally friendly, and designed for use in urban areas. Once an underdog in the race to develop this technology, nuTonomy has since emerged as a frontrunner, partnering with companies such as Lyft and running test programs for driverless taxis in Singapore and Boston.
Behind it all is Emilio, contemplating the subtleties of how humans operate vehicles. As he points out, “There are a lot of nuances in the way that we want cars to behave that are not completely captured by what we know as the rules of the road.”
An MIT alumnus who graduated in 2001 with a Ph.D. from the Department of Aeronautics and Astronautics, Emilio continued in academia, teaching first at University of Illinois at Urbana-Champaign, then at the University of California, Los Angeles, returning to MIT as a professor at LIDS from 2006 until 2017. Although he now spends about 80 percent of his time at nuTonomy as CTO and chief scientist and 20 percent of his time as a professor at the Swiss Federal Institute of Technology in Zurich, he’s also a visiting professor at MIT, supervising several graduate students. It was during his time as a LIDS professor that Emilio began the work out of which nuTonomy grew — work on how to manage large groups of robots that he conducted with his students. The central question of that work, he explains, was “how do you control a robot, that is actually a car, when it’s driving in an environment that’s shared with other cars, pedestrians, bikes….you have a potentially very dangerous machine that has to share the space with humans.” Ensuring the safety of such interactions became the basis of nuTonomy’s mission, which he and Karl Iagnemma, another MIT alumnus, launched together. Emilio says that Delphi’s goals “are very much aligned with our goals,” so one of the many positives of the acquisition has been that it allows Emilio to continue research and development that is meaningful to him, but now with expanded resources.
More specifically, Emilio’s current work involves building the technology needed to create and run large groups of “robot taxis” in urban locations across the globe — and, importantly, understanding the societal impacts of such a project. “How many of these taxis would you need to serve the entire city of Boston, or Singapore?” Emilio asks. “What would it mean in terms of demand for parking? What would it mean for personal mobility? What does it mean in terms of cost and accessibility?” In addition to these questions, he’s considering thornier, less clear-cut issues that speak to the social responsibilities that autonomous car companies must contend with — perhaps the thorniest of these being issues of public safety and liability. Although he believes there is “tolerance for some accidents” among the general public as driverless cars transition from an idea to a reality, he also shares some of the concerns that people have raised. He cites the recent case of Uber’s autonomous car in Arizona that struck a woman who was crossing the street outside of a crosswalk with her bicycle: “The police chief in Tempe, Arizona, stated that ‘Uber would likely not be at fault in this incident,’ because the woman was crossing the street where she shouldn’t have. But are you really happy with autonomous cars that do not attempt to avoid collisions with jaywalkers? Of course not!” The problem is that expectations of appropriate behavior on the road don’t necessarily match up with what robot cars are told to do if they are programmed to follow the letter of the law. The “good behavior” that humans naturally tend to practice when driving is rooted not only in the primary goals of trying not to hurt others while following the rules, but also in things like safeguarding their own lives and protecting their cars from damage. Of course, robots don’t have a natural inclination to do any of this, which is why it’s so crucial that the people creating the cars keep such factors in mind when designing the software for autonomous vehicles.
The best way to do this, Emilio suggests, is by bringing in the entire community on the decision of how autonomous cars should behave when it comes to these types of ambiguous situations, where the written rules of the road don’t completely account for the unwritten ones. He is developing the framework to write the logic that makes the decisions for the car, but he wants the public, the authorities, and industry stakeholders to weigh in and contribute to customizing it to the particular needs of each community. He sees that as the most equitable and ethical way to move forward, and as one that makes good business sense. Referring again to the recent Uber accident, he talks about the fact that Uber settled with the woman’s family, before the completion of any official investigation on the case. He says, “Uncertainty about the interpretation of the rules of the road is a risk I don’t want” as a business, partly because the benchmarks can change. The current rules of the road have many “gray areas,” which require a human’s discretion to resolve. A judge can give one ruling, and later another judge can contradict that ruling, and in the meantime if a company has deployed thousands of vehicles, they would have to recall them and rewrite software to correspond with the new decision. “For me,” he concludes, “the key point is, we have to identify all these gray areas and what to do if [a car] is not behaving [the way the community wants it to].” This is his biggest challenge right now: navigating the best way to handle the increasingly complex situations that will arise as autonomous cars become more integrated into society.
For Emilio, the rollout of autonomous cars is at the moment a question of scope. With a very well-known situation, in which the road conditions, traffic, and weather are specified and understood, it’s easier to tackle the technological challenges of driverless cars and offer a service that people are interested in using, whether because it’s better for the environment, cheaper, or more efficient. Once some success is had in those set circumstances, nuTonomy can widen their plans to include other situations and services. Some of these developments are likely in the near future, others are not. “If you’re asking me when will you be able to go to a car dealership and buy a new autonomous car, probably [it will be] another 15 to 20 years. If you’re asking me when you will be able to go somewhere in the world and actually be able to summon an autonomous car that takes you to someplace specific that you want to go, closer to two years,” he says.
In the meantime, nuTonomy has already piloted autonomous cars in Boston and is continuing its work in Singapore with its fleets of driverless taxi. As Emilio splits his time between the business and academic worlds, travel, and family life, he’s happy to continue working with students at MIT. “Everybody was way smarter and more clever than I am, but that’s a good thing,” he jokes when remembering his own days as an MIT student. “It’s a very supportive environment.” Thanks in part to that support, and his time at LIDS, he’s at the forefront of his field today.