Why the military led
Driverless ground vehicles have long appealed to armed forces for a simple reason: they can go into places where it is dangerous to send people, such as minefields, contaminated areas and supply routes under attack. Military research agencies also have the money and patience to fund long projects with uncertain payoffs. As a result, many of the basic tools of today’s driver assistance, including machine vision borrowed from industrial robots, laser ranging and path-planning software, were first developed with defense funding.
Early military projects
Remote-controlled ground vehicles appeared in the Second World War, when the German Goliath tracked mine and similar small machines were guided by cable. They carried explosives and were operated by people watching, so they were remote-controlled rather than autonomous.
True autonomy needed computers. In the early 1980s, the US Defense Advanced Research Projects Agency began its Strategic Computing program, which included the Autonomous Land Vehicle, a large truck with cameras and a scanner using laser beams built to follow roads on its own. At around the same time, universities such as Carnegie Mellon developed experimental vans. Speeds were low and conditions were limited to clear weather and well-marked roads, but the work showed that software could interpret road scenes. The Army later pursued a program called Demo III in the late 1990s for off-road robotic scouts.
Challenge contests
In 2004, 2005 and 2007, DARPA held contests that offered prize money to teams who could build vehicles to navigate without a driver, first across desert tracks and later in a mock urban environment. The contests brought university teams, companies and hobbyists together, and many participants later joined companies working on commercial self-driving systems. For this story the key point is that they linked military funding to a pool of talent that moved into the civilian industry.
Military ground robots remain in use for bomb disposal, and many are remotely operated rather than autonomous. Fully autonomous military vehicles are still mostly under development or limited to specific tasks, such as convoy-following systems.
Driver assistance arrives in consumer cars
While researchers pursued full autonomy, carmakers added smaller automated features that kept the driver in charge. These can be seen as a ladder of steps.
Cruise control came first. Ralph Teetor, an American engineer, patented a speed-control device in 1950, and Chrysler offered it commercially in 1958. In the late 1990s, adaptive cruise control that uses radar to keep a gap to the car ahead appeared on some luxury models, including Mercedes-Benz in 1998, and similar systems had been tried in Japan a few years earlier. Anti-lock brakes and electronic stability control added safety by automatically intervening in skids.
In the 2000s and 2010s, additional features followed: lane departure warning, lane keeping assist, automatic emergency braking and parking assistance. They depend on cameras, radar and sensors, along with growing computing power from integrated circuits. Studies of such features have generally found benefits in certain crash types, but they have limits, such as difficulty in poor weather or with unusual road markings, and they do not replace attentive driving.
Levels of automation
To describe these differences, engineers use a scale defined by SAE International, with levels from 0 to 5. At Level 0 there is no automation beyond warnings. At Levels 1 and 2, the car can control steering or speed, or both, but the driver must watch the road and be ready to take over at any time; most consumer “autopilot” or “hands-on” systems on sale are in this group, whatever their marketing names suggest. At Level 3, the system drives in defined conditions and asks the driver to take over when needed. At Level 4, the system drives without a human in a limited area or set of conditions, and at Level 5 it would drive anywhere a human could. No Level 5 vehicle exists as of this writing.
From driver assistance to driverless services
Limited driverless taxi services have operated in a few cities without a safety driver, using detailed maps, many sensors and remote support. These are Level 4 systems operating in limited areas rather than anywhere. They are the exception, and personal cars sold to consumers are almost all Level 2 systems that require supervision. Safety investigations after some crashes involving driver-assistance systems have led regulators to examine how such features are named and monitored.
What to take away
The path from military research to consumer cars was gradual and uneven. Technology developed for defense, from lasers to machine vision, informed civilian projects, which then moved on a separate track once companies saw commercial value. Advances in artificial intelligence and global positioning have made the technology more capable, but the label “autonomous” is still used loosely, and the history shows why it matters to ask which tasks the machine handles and what the human must still do. Prospects for fully driverless consumer cars remain uncertain, and projections have often been too optimistic.