Do not be alarmed if you see a black Ford Focus sedan driving along Route 1 with no one holding the wheel. It’s not out of control — it’s a prototype self-driving car developed by a group of Princeton University researchers, and the engineers who made it hope to demonstrate its full capabilities at the upcoming Princeton Reunion weekend.
Alain Kornhauser, a Princeton professor who has been researching self-driving vehicles since 1971, says the car, the third generation of a vehicle under development since 2005, is almost ready to hit the road under its own guidance. Kornhauser is currently leading the Princeton Autonomous Vehicle Engineering (PAVE) team, which is composed of graduate students who are building a robocar.
Alain Kornhauser’s PAVE team will demonstrate the vehicle Friday, May 29, from 3 to 5 p.m. in front of McCosh 10. For more information and a full schedule of the reunion, visit alumni.princeton.edu.
Kornhauser says the car is currently rigged with controls that allow it to be driven remotely by a human operator. The next step is handing those controls over to a computer (albeit with a biological, non-artificial intelligence ready to take over at a moment’s notice).
To get the car to that point, the PAVE team is giving the car a kind of virtual driving lesson. Teaching the computer to drive is a process called deep learning neural networking. Unlike some self-driving cars, which are equipped with radar and other kinds of sensors, the PAVE vehicle relies completely on cameras to see its surroundings. The key is teaching the car’s artificial intelligence to correctly interpret the images the camera takes.
“It’s an approach to doing a preprocessing of an image to get the basic characteristics of that image, and take those characteristics through a standard neural network that relates those characteristics to some representation of an image,” Kornhauser says. “So, in face recognition, it would be recognizing faces. What we’re doing is trying to determine where the car is relative to the lane it’s currently operating in, how far ahead the car in front of it is, if there are obstacles in the lane, and where they are relative to the position of the car. Also, what the speed of the car is.”
Kornhauser says the PAVE team has “trained” the AI in the virtual world, and taken it out in the car, not hooked up to controls, to test its reactions. “We’ve driven it along Route 1 with the camera and looked at the values it would be putting into the steering and brake and throttle controls,” he says. “Those values look good. We hope we can do it. Hopefully by the reunion, but I don’t know if we’re going to make it.”
The PAVE project is purely an academic exercise. It also won’t be the first time one of its vehicles has taken to the highway under its own guidance. As early as 2007, the PAVE team let its prototype car loose on the highway with a human only as backup.
The work of the PAVE team and others has pushed forward the technology of self-driving cars rapidly in the last decade, and automakers and tech companies have been aggressively developing the technology. Many new cars now come with crash-avoidance features that prevent a human driver from rear-ending cars in front of it, or from sideswiping another driver during a lane change.
Google is building an entirely self-driving car that the company hopes to put on the market in five years. Tesla, the electric car company has announced it plans to release a software update this year for its existing cars that includes some self-driving capability for its newer Model S sedans. Tesla CEO Elon Musk told reporters the update would allow the cars to steer themselves when driving on the highway, and would allow the car to leave a parking space and drive to its owner as long as it was in a parking lot.
Kornhauser says the self-driving car is an idea whose time has arrived.
Kornhauser was born in France and grew up in Pittsburgh. He majored in aerospace engineering at Penn State, graduating in 1965 and going on to Princeton where he earned a doctorate. He has stayed there ever since. In addition to his academic endeavors, he co-founded ALK Technologies, an early developer of GPS mapping technology that is based on Harrison Street. ALK Technologies was bought by Trimble, based in Sunnyvale, California, in 2013 and is now a division of its parent company. Alain’s son, Michael, leads ALK as general manager.
Alain Kornhauser has recently gotten into the self-driving car business himself as the co-founder of Soterea, a company based on Farr View Drive that is developing a system called “Autonobox” that is an automatic, secondary braking system for trucks and heavy vehicles that can be retrofitted onto existing vehicles. (U.S. 1, March 12, 2014.)
Kornhauser has also proposed turning the vacant military base at Fort Monmouth into a center for autonomous vehicle research. “It’s the right place to do this kind of stuff,” he says. “It’s a protected environment. You could do an enormous amount of testing before you leak it onto Route 1. Why should everybody go to Silicon Valley, or Beijing, or Stuttgart, or Detroit? Why not New Jersey?”Another argument: the Garden State has rain and snow, conditions in which a self-driving car would need to be able to operate. Kornhauser has been pitching the idea for two years but has yet to gain legislative support for it.
It wouldn’t be the first time Kornhauser was ahead of his time. “I’ve been in automated vehicles for cities since 1971,” Kornhauser says. “It’s been a long road for me. The interesting piece is that in the past, we thought we needed to have a separate highway for these things. The idea of having these operating in a mixed environment with humans was at first thought to be totally crazy, but in the last 10 years, since Google has jumped on board, everybody says this is the way to go.”
There are two major development tracks for self-driving car technology. On one hand, Google is trying to develop an automated car in one fell swoop. Meanwhile, car companies are gradually adding crash avoidance features to their new models. Kornhauser applauds both trends. And as better safety systems become available, the case for putting them into service is becoming stronger. Investigators of the May 12 Amtrak derailment, which killed eight and injured hundreds, say it could have been prevented by a relatively simple automated speed control system.
To Kornhauser, there is no point in blaming human vehicle operators who cause accidents due to lapses in attention. “It’s unfair to ask someone to pay attention all the time,” he says. “That train engineer doesn’t get paid enough to pay attention absolutely every second and never get disoriented. Some of these guys are getting pelted by boulders as they’re coming down the track. Can you imagine? And we’re not going to have systems on board to help them? Whoa.”
Kornhauser is not a fan of warning systems. “I don’t need a backseat driver. If the system feels the vehicle is about to enter a situation that’s dicey, then just keep me out of there. We put stability control on all cars. If you’re taking a turn too fast, they use brakes and torque to redistribute force to keep you from losing your butt end going around a curve. You don’t see a warning to push your left rear brake two ounces and then jig and then jag. No way. Blind spot warning systems are very good, but if there is something there, don’t just warn me. Don’t let me change lanes.”
Kornhauser noted that a study by the National Highway Traffic Safety Administration last year showed motor vehicle crashes cost $871 billion in damage every year, not to mention the 30,000 lives lost. The immediate property damage and medical bills from injuries were only a third of that, with the rest coming from the traumatic aftermath of accidents that can disrupt lives and leave people out of work.
With the rapid development of technology, Kornhauser believes that cars like PAVE’s self-driving prototype will become the rule, rather than the exception. However, he does believe that widespread adoption is years away, although he would be happy to be proven wrong. He has gone on the record doubting Tesla’s planned “autopilot” rollout will actually take place in 2015.
“I have this big plate of crow that I’m ready to eat if they do it,” he says. “That would be neat. It just shows the power of all this.”