In 1989, the Berlin Wall began to fall, the World Wide Web made its debut, Madonna’s “Like a Prayer” topped the charts, and in Pittsburgh, a retrofitted Army ambulance called ALVINN was driving around Carnegie Mellon University without any human intervention.
Self-driving cars may seem like a very recent technological phenomenon, but researchers and engineers have been building vehicles that can drive themselves for over three decades. Research on computer controlled vehicles began at Carnegie Mellon in 1984 and production of the first vehicle, Navlab 1, began in 1986. ALVINN, which stands for Autonomous Land Vehicle In a Neural Network, was used as a test vehicle well into the 1990s.
This proto-driverless vehicle came up recently in a Twitter discussion between two engineers: Oliver Cameron, who heads an open-source self-driving car project at Udacity, and Dean Pomerleau, a CMU professor who ran the self-driving car project that gave birth to ALVINN. Cameron tweeted a video shared by some of his students of a car steering itself autonomously using only a camera.
This prompted Pomerleau to ask a few questions about deep learning and neural networks. After some back and forth, Pomerleau brought up ALVINN, which had an operating system of 100 million floating point-operations per second, or about one-tenth the processing power of the Apple Watch. The vehicle’s CPU was the size of a refrigerator and was powered by a 5,000 watt generator, he added. Nonetheless, ALVINN was able to hit 70 mph by the early 1990s.
The result of eight years of military-funded research at CMU’s robotics institute, ALVINN could be considered the forefather of today’s self-driving cars, Cameron told me in an email. “Why? The approach ALVINN took was using a neural network to drive the car, which was absolutely novel for the time and is quickly becoming an increasingly popular approach with self-driving car efforts,” Cameron said.
While Google’s self-driving cars rely on 3D maps to situate itself in its environment, ALVINN’s use of a neural network meant the vehicle was “narrowly intelligent” and make decisions without the need for a map, he added. “You can drop a neural network powered car in locations it has never seen before, and have it gracefully perform, using techniques human use [from] past experience,” Cameron said.
By using a neural network to teach the vehicle to drive, CMU’s Pomerleau hoped to build an autonomous driving system that was more adaptable across a variety of conditions. The prototype was designed to control the NAVLAB, Carnegie Mellon’s autonomous vehicle program which started in 1984. ALVINN’s neural network was “beautifully implemented, but constrained very much so by the hardware,” Cameron wrote in a subsequent post on Medium.
It’s also worth noting that Chris Urmson, the one-time head engineer of Google's self-driving car project, was a CMU colleague who worked with Pomerleau. Urmson built on the technology pioneered by Navlab, to lead the team from CMU to victory in DARPA Urban Grand Challenge in 2007, before getting snapped up by Google to lead their autonomous car efforts. Meanwhile, Uber poached dozens of CMU robotics experts to work on its self-driving car technology. So many of the autonomous vehicle projects operating on the road today can trace their origins to Pomerleau and the NAVLAB project.
“It's really quite gratifying to see, and it's been fun interacting [via Twitter] with the Udacity students learning about and developing their own self-driving car technology,” Pomerleau told The Verge in an email, “sharing with them ideas from way-back-when and lesson learned.”
This local TV news spot on ALVINN from 1997 perfectly encapsulates the awe and skepticism that still colors most mainstream media reports on self-driving cars. It also features some familiar arguments made by supporters about the impact self-driving cars could have on traffic accidents and fatalities — arguments still being made today by Google, Uber, and all the rest.