Computer scientists have been at the self-driving vehicle problem for longer than you might think. Early research into the automated logic required for autonomous cars was published in the mid-70s, while the first fully robotic van came around in the early-80s courtesy of Ernst Dickmanns and his team at Bundeswehr University Munich. Efforts at Carnegie Mellon, meanwhile, were pushing the technology on the other side of that Atlantic. First came Terregator, in 1983 (via IEEE Spectrum):
Piloting city streets wasn't yet part of the plan. Then came NavLab, in 1986.
Yeah, it's pretty quaint, but machine vision algorithms, in particular, were still young. In 1989, Carnegie Mellon pioneered a neural network called ALVINN that could be employed in road following tasks in certain field conditions. It used machine learning to watch human drivers and adapt its own strategies. ALVINN would become the precursor to the self-driving technology of today.