In 1986, a blue Chevy van often cruised around the streets of Pittsburgh, Pennsylvania near Carnegie Mellon University. To the casual observer, nothing about it appeared out of the ordinary. Most people would pass by it without noticing the camcorder peeking out from its roof, or the fact that there were no hands on the steering wheel.
But if any passerby had stopped to inspect the van and peer into its interior, they would have realized it was no ordinary car. This was the world’s first self-driving automobile: A pioneering work of computer science and engineering somehow built in a world where fax machines were still the predominant way to send documents, and most phones still had cords. But despite being stuck in an era where technology hadn’t caught up to humanity’s imagination quite yet, the van — and the researchers crammed into it — helped to lay the groundwork for all the Teslas, Waymos, and self-driving Uber prototypes cruising around our streets in 2022.
The aforementioned van was designed and built by Carnegie Mellon’s Navigation Laboratory (Navlab) — long before the World Wide Web or Google existed, and with computers that were 10 times less powerful than the first-gen Apple Watch.
With funding from the U.S. defense department, Carnegie Mellon’s robotics division set up the Navlab in 1984 to explore autonomous navigation. The objective, Dr. Chuck Thorpe, the computer science professor who ran the project told Digital Trends, was to deal with “dull, dirty, and dangerous” situations.
The defense department, more specifically, was looking to build autonomous scouts. These scouts would go out on the field and map uncharted territories, where there’s usually a greater risk of hidden mines and enemies — a job humans would before risk their lives for. And thus, the Terragator was born in 1983.
The six-wheeled Terregator, which, at first glance, could be easily mistaken for the predecessor of the Mars Rover, was the world’s first autonomous outdoor driving robot, and for a time when mobile phones weighed 11 pounds, it was a remarkable engineering feat. It featured a range of sensors and computer vision technology to avoid obstacles, climb uneven terrain, track paths, and much more. Work on the Terregator helped researchers realize the potential of this technology, and three years later, the Navlab 1 — that blue Chevy van — hit the streets.
The Navlab 1 was as primitive as a self-driving car could get. It didn’t have the sleek touchscreens or the smartphone controls you’d find inside autonomous vehicles these days. What it did have was half a dozen racks of computer hardware the size of refrigerators, a full-size camcorder peeking out from above the windshield, a 20-kilowatt generator, and a few blocky monitors used to display the algorithm’s performance to a handful of grad students crammed into the back. The whole setup seemed more like an FBI surveillance van than a self-driving project.
The way Navlab 1 steered itself was fairly straightforward. Its lidar sensor — similar to the one found on the latest iPhones — would shoot lasers on objects to determine its distance from them. On top of that, with computer vision, it would break down the footage from the video camera to follow lane markings and figure out the edges of the road so that it doesn’t go off track. Results from these data points would ultimately help it send the final steering commands.
If that sounds like a lot of work for computers from the 1980s, that’s because it was. Since the hardware had not yet caught up to such advancements, it would take ages to churn out the calculations, and as a result, Navlab 1’s speed was limited to 20 mph.
What’s more, the heaps of hardware crammed in the back of the van suffered from limited ventilation, and therefore, it also frequently broke down and once even caught fire, according to Dr. Dean Pomerleu, who joined the Navlab team as a Ph.D. student.
While Navlab continued to refine its self-driving modules over the coming years, it wasn’t until 1989 when Dr. Pomerleu taught a camo-colored Army ambulance Humvee — Navlab 2 — to learn from its mistakes that the group achieved its next breakthrough.
Till 1989, Navlab students were hard-coding programs to patch the self-driving car’s shortcomings as it encountered unfamiliar situations. On the other hand, Dr. Pomerleu’s ALVINN (short for An Autonomous Land Vehicle in a Neural Neutral) algorithm allowed the vehicle to adapt to scenarios it wasn’t programmed for simply by watching how a human driver would react in that case. This meant that the next time Navlab 2 encountered that same scenario, it wouldn’t need human intervention. It’s what unlocked the next generation of self-driving cars, and even in today’s A.I.-based systems, one can find the hints of ALVINN.
Soon enough, Navlab 2 was cruising at 55mph on a 102-mile road trip from Pittsburgh to Erie, Pennsylvania. “That was the first really long trip it had done and convinced me that someday we’d see vehicles that could drive themselves on public roads,” added Dr. Pomerleu.
Since Navlab iterations depended on an adaptive neural network and not on 3D maps like Google’s self-driving car, they could be dropped in any location they have not seen before and perform well enough. That’s what ultimately powered the Navlab division’s victory lap: A nearly 3,000-mile road trip across the country from Pittsburgh to San Diego in 1995.
The Navlab 5 steered itself for over 98% of the trip, with Dr. Pomerleu and his grad student, Dr. Todd Jochem, taking turns to throttle and brake. And despite the vast variations in road types and terrains, the pair faced almost zero anomalies and journaled the whole experience throughout the trip, including the day when they demoed it for the former The Tonight Show’s host, Jay Leno, in what was one of the first travel online blogs.
“I think if you go back in time and get one of those cars now,” Dr. Jochem, who now drives a Tesla Model S, said in an emailed exchange with Digital Trends, “you’d be shocked at how identical it is in some situations to the performance you see on cars that are now commercial. Very proud of that.”
Members of the Navlab team went on to found and significantly contribute to today’s leading self-driving projects, such as Uber, Google, Tesla, and more. Yet, despite the progress the industry has made, Dr. Pomerleu believes an “A.I. winter,” a term used in academic circles to describe a period of low funding and growth in a field, might be looming around the corner for autonomous vehicles and Elon Musk could be to blame.
While Dr. Pomerlue agrees Musk has helped advance the self-driving age, his approach to autonomy, which relies far too heavily on camera sensors, and callous policies towards driver safety, is worrying. “Ultimately overpromising and underdelivering is unconscionable in my opinion, and threatens to contribute to another ‘A.V. winter’,” he added.
At the time of writing, U.S. National Highway Traffic Safety Administration announced it is investigating Tesla for letting drivers play video games on the dashboard screen while the car’s driving on autopilot.
The work for researchers like Dr. Thorpe has, therefore, yet to reach its finishing line. “Thirty years ago I predicted I would ride into retirement in a self-driving car,” he quipped, “I guess I can’t quite retire yet.”