Is there a “first” autonomous car?
Several vehicles have a claim, and which one counts depends on the definition. If the test is a car guided without a human steering by a computer, the early candidates come from the 1970s and 1980s. If the test is a car that can handle ordinary public roads without anyone watching, no vehicle has clearly met that standard in the whole period covered here, and claims about current systems should be read carefully. This article follows the research vehicles that built the technology and the contests that accelerated it.
Early ideas and demonstrations
Self-driving was imagined long before computers could do it. In 1925, a radio-controlled car called the “American Wonder” drove through New York streets with no one at the wheel, directed by an operator in a following car. That is remote control rather than autonomy, since a person was still making the decisions. At the 1939 New York World’s Fair, a General Motors exhibit imagined automated highways, and in the 1950s, companies including RCA and GM tested ideas using cables or signals embedded in roads to guide cars.
The 1970s and 1980s
The first serious attempts to have a vehicle see for itself came with early computer vision. In Japan, a research group at the Tsukuba Mechanical Engineering Laboratory demonstrated a camera-guided car in 1977 that followed white road markers at slow speeds. In Germany, Ernst Dickmanns and his team at the Bundeswehr University Munich equipped a Mercedes-Benz van with cameras and computers. By around 1986 and 1987, their vehicle, known as VaMoRs, steered itself on empty roads, and speeds reached about 90 km/h on test streets by later stages of the work.
In the United States, the Defense Advanced Research Projects Agency funded the Autonomous Land Vehicle project in the 1980s, and Carnegie Mellon University began its Navlab series of vehicles in 1984. Europe launched a large collaborative program, called Eureka Prometheus, in 1987, bringing together carmakers and universities. All these projects depended on cameras and sometimes laser beams swept across the scene to measure distance, and on computers that were slow by modern standards.
The 1990s highway demonstrations
By the mid-1990s, the work had reached public roads under controlled conditions. In 1994, Prometheus-related cars drove on a Paris highway in traffic with human safety drivers able to take over. In 1995, a Mercedes-Benz car from Dickmanns’s group drove from Munich to Denmark and back, over 1,700 kilometers, with a computer handling most of the driving on motorways, though humans intervened in some situations. That same year, Carnegie Mellon’s “No Hands Across America” trip took a Navlab vehicle from Pittsburgh to San Diego, with its software steering for the large majority of the distance while a person controlled speed and braking. These feats show how much could be done in controlled settings and how much work remained for urban streets.
The DARPA challenges
A turning point came when the US defense research agency offered a cash prize for a vehicle to cross a stretch of desert without human control. The first Grand Challenge in March 2004 ended without a finisher, with the best entry covering only a few miles of the roughly 140-mile route. A year later, in October 2005, five vehicles completed a desert course of about 132 miles, and Stanford’s Stanley was the winner. Teams used a mixture of global positioning, laser scanners, cameras and radar, and they relied on advances in artificial intelligence for interpreting what the sensors saw.
In 2007 the Urban Challenge moved to a simulated town course, requiring vehicles to obey traffic rules, merge and park. Carnegie Mellon’s entry, Boss, won. The contests built a research community, and many participants later joined technology companies and car manufacturers.
From contests to products
In 2009, Google began a self-driving project that later became Waymo, using mapped roads and sensors, including a spinning laser scanner. Companies gathered large amounts of driving data, an application of big data methods, to train their software. Other firms and traditional automakers launched their own programs. Over the following years, limited services with no safety driver were introduced in a few cities, and many new cars gained driver assistance features such as lane centering and adaptive cruise control. Those features are not fully autonomous: the driver is expected to supervise, and the industry’s SAE levels from 0 to 5 are used to describe how much responsibility lies with the system.
Why hedging matters
Progress has been real, but timelines have often proven too optimistic. Handling rare events, bad weather and unpredictable people remains difficult, and crashes involving automated systems have led to investigations and debate about safety and regulation. Outcomes for any single technology remain uncertain. The better way to view the history is as a chain of research vehicles, each adding sensors, software or road experience, rather than a single first car.