ALV: the Strategic Computing land vehicle drives itself
In May 1985 the Autonomous Land Vehicle, which Martin Marietta was building for DARPA's Strategic Computing programme, drove a road by itself in public for the first time, about 1 km. In June 1986 it covered the whole test track, 4.2 km, at up to 10 km/h, and in October 1986 it first steered around obstacles while staying on the road and reached 20 km/h on a straight road without obstacles.
Why it matters
The first of the three demonstrations promised by the 1983 Strategic Computing plan reached the road: vision, reasoning and control worked together on board a vehicle. The 1986 report gives the programme's yearly goals as plans, while DARPA's budget of February 1987 already lists road following over 4 km at up to 10 km/h among its accomplishments.
Martin Marietta Denver Aerospace won the integrating contract in August 1984; the University of Maryland supplied the vision algorithms for the first demonstration. On board were a colour video camera and an ERIM laser range scanner, with images processed on a VICOM processor. Leighty's report from the US Army Engineer Topographic Laboratories (March 1986) says that later in 1985 the vehicle travelled 2 km at 10 km/h, needing 1.75 seconds a frame, and that the VICOM was to be replaced by Carnegie Mellon's WARP systolic computer. DARPA's budget of February 1986 lists among the accomplishments of fiscal 1985 that the vehicle was engineered and the initial road following demonstrations carried out, in the agency's words with an improvement factor of 100 over prior capability. Road following at up to 20 km/h with obstacles on the WARP and Butterfly architectures reappears in its programmes for fiscal 1986 and 1987 as a plan, not a dated demonstration. Roland and Shiman (2002) add the terms. In late 1984 DARPA added a demonstration for the following May: a slow kilometre on a generally straight paved road, with up to a quarter of the way allowed blind on dead reckoning. They give the course as 1,016 m run in 1,060 s, 870 m of it under visual control, which DARPA called a hundred times faster than any autonomous vehicle before. The next demonstration, six months later, asked for 10 km/h on a straight, 3 km/h through a sharp curve, a stop at a T-intersection and a turn-round, and they say the vehicle passed it in November 1985. The demonstration of June 1986 went so well, they write, that DARPA's Clinton Kelly repeated it for the press a few weeks later; obstacle avoidance was due at the less formal review of October 1986. Discrepancies. Leighty's report describes the first demonstration of May 1985 as 1 km at 5 km/h, and the same sentences stand in an earlier paper by Leighty and Lane; the Martin Marietta engineers' paper in IEEE PAMI (May 1988), the abstract of the company's final report on the first phase (May 1988) and an abstract by its employee Robert Douglass (June 1988) give 1 km at 3 km/h. So 5 km/h stands only in the sponsor's papers and 3 km/h in every text of the contractor. Popular Science of October 1985 and a Lockheed Martin page of 2020 write three miles an hour (about 4.8 km/h), while Roland and Shiman's course and time divide to about 3.5 km/h. No document read says whether the contractor's 3 km/h is a slip of units; the record does not decide. The second demonstration is dated June 1986 by the paper (4.2 km), May 1986 by the final report's abstract and a quarterly report (4 km), and Douglass writes of 4.5 km. Roland and Shiman also have June 1986. The first phase closed with the demonstration of 5 November 1987: by the final report's abstract, 4.2 km at up to 20 km/h, 14 km/h on average, while avoiding obstacles on the road; by the quarterly reports the Warp had been installed on the vehicle before it. Roland and Shiman, from the seventh quarterly report, give the same 20 and 14 km/h. By their account Takeo Kanade of Carnegie Mellon faulted the programme as too driven by demonstrations at a review in Denver on 6 November; DARPA quietly dropped the milestones, and in April 1988 cancelled the programme; DARPA's amended budget of February 1988 still lists the ALV among the major milestones of fiscal 1987, without figures. Martin Marietta's reports were not found on any open host; only their abstracts, in the Engineer Topographic Laboratories' bibliographies. The annual report ETL-0413 covers August 1984 to August 1985 by the catalogue of US government publications, is marked for official use only and was printed in July 1986; Roland and Shiman date it December 1985. Of the programme's annual reports only the first (February 1985) was found, with nothing yet about driving; the second to the fourth (February 1986 to November 1988), which Roland and Shiman cite, were not. The IEEE PAMI paper was read only in snippets from the Internet Archive's full-text search. The National Research Council's report on Army unmanned ground vehicles (2002, Appendix D, pp. 152-153), written outside the programme, lists 1 km at 3 km/h in 1985, 4 km at 10 km/h in 1986, 20 km/h in 1987 and, in August 1987, the first autonomous cross-country traverse, at 3.5 to 5 km/h with about 75 per cent of traverses completed. It dates the 20 km/h to 1987, not October 1986. What the record does not claim: the programme's goals (5 km with obstacles in 1986, desert in 1987 and so on) as achieved, since Leighty's report gives them as plans. The improvement factor of 100 as a measurement: it is the sponsor's figure without a method. The day of the first demonstration: a Lockheed Martin page of 2020 gives 31 May 1985, but no document of the time read here names a day. When and why the project ended as its own documents tell it: the review and the cancellation come from Roland and Shiman, citing papers not seen here.