Navlab: a van drives a road without stopping
In May 1988 Thorpe, Hebert, Kanade and Shafer of Carnegie Mellon described in IEEE Transactions on Pattern Analysis and Machine Intelligence (volume 10, issue 3, pages 362-373) the perception and navigation system of the Navlab van: a distributed architecture around the CODGER database, colour vision for road following and 3-D vision for detecting and avoiding obstacles; the vehicle drives continuously on roads in a real outdoor environment while avoiding obstacles. The CMU year-end report for 1987 lists among its milestones a Navlab demonstration on the Warp processor at 50 cm/s.
Why it matters
For the first time a vehicle with a camera drove a road continuously rather than in bursts between computations. Navlab became the test-bed on which the ALVINN neural network would steer the following year in place of hand-written algorithms, and the ancestor of the CMU road vehicles that would cross America in 1995.
The paper's abstract names no speed; the 50 cm/s stands in the 1987 milestone table of the 1987 Year End Report for Road Following at Carnegie Mellon (CMU-RI-TR-88-4, August 1988), which was found in the CMU archive under the title of the paper itself, is a different document, and is used as a source in its own right. The same report says that one vision cycle of the PARK I system took about 10 seconds on a Sun 3/160 and 4 seconds on the prototype Warp, and that the Warp gave a fivefold increase in vehicle speed over the 1986 algorithms. The full text of the PAMI paper was not read: it is paywalled. What the record does not claim: distances, run durations, lane-keeping accuracy or any figure beyond those quoted from the report.