(Think Delphi’s cross-country voyage in an autonomous Audi SQ5 is impressive? Try doing it with 90s tech, without GPS navigation, and in a salvaged Pontiac minivan. That’s what Carnegie Mellon research scientist Dean Pomerleau and then-Ph.D. student Todd Jochem did in 1995. Here’s the story of their journey as it appeared on Robotics Trends. — PG)
For the past several years, self-driving cars have been prominently featured in mainstream media outlets. Great technology and future plans from organizations such as Stanford University, Google, various car manufacturers, and more recently Uber and Delphi, have been showcased.
It is with great intellectual interest, pride, perspective, and a fair bit of humor that I have read about these recent “firsts” for autonomous vehicles.
Why? Because July 23, 2015, will be the 20th anniversary of “No Hands Across America,” the first long-duration field test of a self-driving car. I was fortunate to be part of the ragtag team from Carnegie Mellon’s Robotics Institute that built the car and was a passenger on the cross-country trip from Washington, D.C., to San Diego, Calif.
After two decades of technology development and societal acclimatization in the area of robotics and self-driving cars, it’s amazing how much has changed—and, really, how much has stayed the same.
I thought it might be interesting to share some comments on that time, our trip across the country, and what has and hasn’t changed between then and now.
Today’s self-driving cars are so stylish
I’m jealous of the stylishness and integration of the most recent self-driving cars—luxury brands and cool little special-purpose cars. We used a minivan that had plastic side panels and cloth seats. But it was better than nothing.
We owe a huge thanks to Ashok Ramaswamy—a visionary engineering manager at Delco (predecessor to Delphi)—who cut through monumental red tape and salvaged a Pontiac Transport minivan from the junk heap and “donated” it to us to use as we saw fit.
We built the vehicle and software over about a four-month time frame for under $20,000. We had one computer, the equivalent of a 486DX2 (look that one up), a 640x480 color camera, a GPS receiver, and a fiber-optic gyro.
It’s funny to think that we didn’t use the GPS for position, but rather to determine speed.
In those days, GPS Selective Availability was still on, meaning you couldn’t get high-accuracy positioning cheaply. And if you could, there were no maps to use it with! But, GPS speed was better than nothing, and it meant we didn’t have to wire anything to the car hardware, so we used it.
Software challenges
In late 1994, Dean Pomerleau had pushed his ALVINN neural network lane tracking software about as far as it could go, but there were limitations with training speed and performance that he felt prevented it from getting to the next level—superior performance across all road types in all weather and lighting conditions.
As the vehicle moves along, a video camera mounted just below the rearview mirror reads the roadway, imaging information including lane markings, oil spots, curbs, and even ruts made in snow by car wheels. The camera sends the image to a portable computer between the car’s front seats that processes the data and instructs an electric motor on the steering wheel to turn right or left.
The driving system runs on the PANS (Portable Advanced Navigation Support) hardware platform. The platform provides a computing base and input/output functions for the system, as well as position estimation, steering wheel control, and safety monitoring. It’s powered from the vehicle’s cigarette lighter and is completely portable.
But inspiration hit Pomerleau on how to go forward while he and Chuck Thorpe were driving down the Rocky Mountains in a snowstorm after a DARPA “meeting.” The insight, which is still proprietary, was enough to make him junk ALVINN immediately and start over.
From around January 1995 to April or May 1995, he built a new system called RALPH (Rapidly Adapting Lateral Position Handler) that quickly equaled ALVINN’s performance—at least on local roads.
But to truly test the system, more roads were needed. And that was when the plan to drive across the U.S. was hatched. Quickly dubbed “No Hands Across America”—mainly because it was a nice play on the “Hands Across America” movement to combat hunger and poverty—the plan was to drive I-70 from Washington, D.C., to I-15 in Utah, then south to San Diego.
From May until when we left on July 23, 1995, time was spent refining the technology, planning the stops along the route, getting approval (I think) from CMU’s board of trustees, and getting sponsorships and fundraising to pay the tab.
Since there was no real sponsor for the trip (no one in their right mind would pay for something this crazy), we had to supplement the little money we got from CMU’s discretionary accounts with free equipment.
As noted above, Delco provided the car, while the computer, GPS, and gyro were all donated to us in exchange for a sticker on the side of the minivan. For gas and spending money, we sold trip T-shirts. I’m not kidding. They were $10 apiece and helped pay for food and hotels. Seriously.
Not sure if the NASCAR model of fundraising (car decals and T-shirts) has been used since then for robotics.
You can read the trip journal for more details, but suffice it to say that we learned more on that seven day trip than the entire research community may have learned in seven years. We also had a ton of fun. From renewing marriage vows in Las Vegas in a self-driving car, to seeing a six-legged cow at Prairie Dog Town, to driving across Hoover Dam autonomously, to meeting Jay Leno, it was a trip for the ages.
And perhaps the highlight was when Otis Port, a writer for Business Week who was doing a story on the trip (read the story here), was pulled over by a Kansas State Trooper—as we sped by with our hands off the wheel.
While I’ll admit to a strong bias, I think Carnegie Mellon University in Pittsburgh was the center of technical excellence in self-driving cars. Beginning in the 1980s and 90s with the NavLab project, CMU has led the development or trained the people who have been at the forefront of this technology.
Initially, it was Kanade, Whitaker, Thorpe, and Pomerleau. Under those pioneers in the field, the next generation of technology and thought leaders like Sebastian Thrun (Stanford), J.O. Urmson, and Astro Teller (Google) were trained.
And it’s clear that CMU remains at the center of the self-driving car universe even now, with Uber’s decision to locate self-driving car research in Pittsburgh and to essentially in-house CMU’s brainpower, and finally, Delphi’s drive across the country powered by—you guessed it—a CMU spin-out company called Ottomatika that is based in Pittsburgh.
I was lucky to be at the right place, at the right time, with the right people. It’s not often in life that the right circumstances are in place to do something that no one had ever done before. When we did the trip, the field was about discovery and expanding technical frontiers.
I think it still is now, but unfortunately, it’s now also about patent fights, liability concerns, and state laws. (If you’re ever in doubt about your patent, just assume it was done at CMU between 1985-1997 – you’ll save a lot of money!)
Those were the good old days, I guess!
This article first appeared on Robotics Trends, and its text and photos have been republished here with permission.
Click here for more information about “No Hands Across America,” including the trip journal, pictures, and description of the vehicle.
Dr. Jochem is a robotics, unmanned systems, and technology professional interested in entrepreneurship and technology creation in robotics and related business sectors. He is currently a consultant to small technology businesses. Before his business career, Jochem was a systems scientist at Carnegie Mellon University’s Robotics Institute and a 1997 recipient of CMU’s Allan Newell Award for Research Excellence.