In 2003, Kurt Dresner showed up late to a meeting he had with his advisor, Peter Stone, the director of the Learning Agents Research Group at the University of Texas at Austin. Dresner had been sitting idly at a red light waiting for the light to turn. No cars came through. He could have crossed the road safely and been punctual. It was a complete waste of time. “I can do better,” Stone remembers the tardy Dresner saying.
Dresner, then in his second year of graduate school, went about making good on that boast. He switched his focus from working on Stone’s Robot Soccer project to developing distributed systems that coordinate how autonomous cars move through intersections. Stone dove in as well. “People would look at us like we were crazy,” says Stone.
A decade later, and the idea doesn’t seem so nuts.
Google, where Dresner now works, has been developing autonomous cars for years and recently Nissan said they wanted to bring the Googley technology to market by 2020. In September, Japan issued a license to an autonomous electric Nissan Leaf. These are signs that the dawn of robotic cars, or “driverless horseless buggies” as Stone calls them, is upon us.
But it won’t be as simple as just telling your robocar where to drop you off. This coming shift in transportation technology isn’t just about the vehicles, it’s also about the roads the cars will travel. That will require a new paradigm of traffic control–one that is designed for machines rather than humans. Precisely what a handful of researchers, Stone chief among them, are tackling.
Since that fortuitous red light, much of Stone’s research has focused on creating autonomous transit systems that can take advantage of the precision sensors computers have at their disposal. For example, instead of using stop signs and traffic lights to manage traffic, the robotic intersections of the future might rely on a distributed grid system that’s in constant communication with autonomous vehicles.
If a car wants to move through a particular section, it communicates with a nearby server, requests a reservation and only moves forward when its reservation has been confirmed, preventing collisions. “Each car is deciding for itself,” Stone says.
This reservation system could also help optimize the use of space at intersections to allow more cars to move through in less time, reducing congestion and fuel usage. Knowing cars’ precise location down to the minute, Stone says, could also be used to incentivize drivers to clear high-trafficked roads by applying tolls and suggesting longer alternate, but free, routes. He imagines a whole new revenue systems for cities based on microtolling. Imagine that at rush hour in a grid-locked city like New York – actual method rather than madness.
Stone says he and his team have tested out their open-source model transit system in computer simulations with good results. They’ve also taken to the streets of Austin, to see how “Marvin”, their autonomous vehicle, performed among human drivers (it was able to safely make a left turn) and in a mixed reality simulation of Stone’s autonomous intersection management system.
To cross an intersection, Marvin had to communicate with a server that granted or denied his grid reservations. It was the only autonomous car in the physical world whose actions were guided by the model, but it still had to coordinate its movements with a number of additional digital cars added to the traffic mix. Watching the video detailing the project, Marvin seems to have succeeded.
The results are encouraging, but there are still many safety, security and legal concerns that need to be ironed out. What happens if the communication system with the managing server fails? What encryption methods need to be put in place to keep hackers out? How does a system like this work on roads where both robocars and drivers coexist? If a collision does happen, who is liable?
“Industry isn’t ready for this [yet],” says Stone. Still, he’s confident that robotic cars are the way forward. It’s just a matter of time. “If the bar is to be perfect, autonomous cars aren’t going to get us there. If the bar is to be better drivers than people, that’s not that hard a bar to clear.”