That was an awful period, at the beginning of the AI Winter. The expert systems people were in charge, but expert systems couldn't do very much. I made the unpopular comment at the time that expert systems will turn out to be more important than syntax-directed parsing (once a hot topic in computer science - YACC, Bison, etc.) but less important than relational databases.<p>I visited CMU and was inside the ALV once. Didn't get to see it move, though. It had a really slow 2D LIDAR scanner, with two moving mirrors, and three racks of SUN workstations. It wasn't a fundamentally bad idea. It just needed about four more orders of magnitude of compute power. They made it work by making it really slow, using a hydraulic transmission to allow creeping speed.
By the DARPA Grand Challenge of 2005, it was finally possible to do off-road navigation. It's taken another two decades to deal with complex traffic successfully.<p>Neural nets were around, but nobody had enough compute power to use them on anything useful. I once had to write a neural net program to recognize digits 0-9 as a class assignment. That was about what you could do before 2000 or so.