Self-driving cars have been a science fiction staple for years, featuring in movies like “Demolition Man” and “Total Recall”. Given that 1.2 million people a year die in accidents on the roads, it is hardly surprising that people have been trying to make this particular piece of fiction into reality for quite some time. The first attempts at building a self-driving vehicle were made as far back as the 1920s, with a radio-controlled car tested on New York streets by a company called Houdina Radio Control in 1925. In 1977, the Japanese Tsukuba Mechanical Engineering Laboratory designed a car equipped with cameras that could operate without any rails or wires under the road. Progress really started to be made in the 1980s with various projects and competitions aimed at spurring on the design of such vehicles. The Robotics Institute of Carnegie-Mellon University built working autonomous vehicles, and many other universities and car companies started down this path. The Italian company VisLab demonstrated an autonomous car called Braive in 2013, capable of driving on a mixed traffic route that was open to other vehicles. Navlab of Carnegie Mellon pioneered the use of neural networks to control autonomous vehicles as far back as 1989.
The US Defence Department sponsored a series of competitions, offering $1 million in 2004 for anyone capable of producing a car that could complete a 150-mile course in the Mojave Desert. None succeeded, but by October 2005 five vehicles had completed the course. The first successful vehicles mostly used a technology called LiDAR (Light Detection and Ranging) that used a pulsed laser to measure ranges, with the lasers being reflected off obstacles. By 2007 the competition had evolved to an urban environment, and in 2008, the mining company Rio Tinto Alcan deployed the world’s first autonomous mining haulage system. In 2010 the research vehicle Leonie was licensed for street driving in Germany. Progress continued, with trials in various US states in 2012. A trial in the UK was initiated in Milton Keynes in 2015, and a test in China began in 2016 by Volkswagen.
A gruesome milestone was passed in 2016 with the first fatality from such a vehicle, when a Tesla S in Autopilot mode crashed into a truck. By 2019, no less than 29 US states had passed laws permitting driverless cars. The Chinese company Baidu also has commercial robotaxis in several cities, including Wuhan and Beijing. In mid-2025, Waymo was running autonomous taxis in San Francisco, Phoenix, Los Angeles and Austin. The company claims that its vehicles are much safer than human drivers, with 78% fewer crashes. However, broadly speaking, self-driving vehicles seem to have slightly more accidents than human drivers per passenger mile so far. Different reports show different findings, and this is an area that will develop over time as more and more data is gathered.
Artificial intelligence is the driving force behind autonomous vehicles. Machine learning, computer vision and neural networks are used for route planning, perception and decision making. Deep learning techniques process vast amounts of sensor data from the vehicles and combine it to create an understanding of the environment. These allow the vehicles to detect traffic lights and road markings, with LiDAR providing precise distances. Other AI systems categorise objects such as other vehicles, pedestrians and cyclists, allowing the car to react to the presence of these, such as avoiding traffic hazards and obstacles.
The Society of Automotive Engineers defines six levels of driving automation, from 0 (no automation) to 5 (full automation). The Waymo taxis are defined as level 4, meaning that they can handle all tasks in most conditions without driver intervention, but may have limitations in certain environments. Their combination of radar, cameras, and LiDAR allows them to drive autonomously within specific geographic locations, such as the cities mentioned. At present, no company has developed a car capable of level 5 autonomy, though active research continues.
There are clear advantages to autonomous vehicles. Such cars do not get tired or drink alcohol, nor lose attention. They can offer improved fuel efficiency, reduce congestion by maintaining consistent speeds, and use real-time data to find efficient routes. Although some people enjoy driving, driverless cars will allow others to relax, and potentially be more productive by being able to read or even work during journeys. Because they can be heavily utilised, they have the potential to improve parking congestion. Cars are parked for 95% of the time, but driverless cars can serve a range of customers in a day, stopping only to pick and drop people off, and to recharge. A form of AI called Cached Decentralised Federated Learning allows cars to potentially share information with each other, even without direct contact between vehicles. This is analogous to current mapping technologies like Waze, which pass real-time traffic information from other drivers to their users.
Autonomous vehicles still have many hurdles to overcome, despite their considerable progress. Handling unpredictable human behaviour, bad weather, road accidents, and unexpected situations is a challenge for their AI systems. Edge cases that a human can handle, such as a deer jumping out into the road, are tricky for AI systems to deal with. There are even moral questions, such as how they deal with critical situations. The hypothetical “trolley problem” (would you kill one person or five in an unavoidable accident) moral dilemma may be of limited use in the case of autonomous vehicles, but the vehicle designers need to consider many safety scenarios when they are building their decision-making AIs for vehicles.
There are other issues. Charging infrastructure needs to be developed further. Building public acceptance will take time, and if the technology does well, then governments will also have to consider the effects on employment. There may be 18 million taxi drivers in the world, with 6 million active Uber drivers alone. Uber itself is trialling autonomous vehicles, and the job displacement effect may be considerable. There have already been public protests against autonomous vehicles, and protestors have taken to using traffic cones to disable autonomous taxis. Although the Waymo data suggest that autonomous vehicles can be safer than humans, there have been over 3,000 incidents involving such cars, with 83 fatalities. The economics of autonomous cars are somewhat unclear, as so many of the current ones are pilot projects or experiments, funded by companies or governments. However, McKinsey reckon that they may amount to an industry worth hundreds of billions within a decade. There are effects to consider on regular cars, too, since the technology used in autonomous cars is creeping into mainstream production. Many cars now have adaptive cruise control and lane-keeping assistance, as well as automatic emergency braking. Such technology may start to reduce accident rates in regular vehicles, which itself may impact the car insurance industry. Governments will need to adapt legislation – who is liable for an accident in an autonomous car? How about one with partial autonomy but where a driver is still at the wheel? Such legislation is already being passed in some countries, as in the 2024 Automated Vehicles Act 2024 in the UK.
Despite such challenges, autonomous vehicles are slowly becoming established, at least within a limited number of urban settings, and seem certain to expand further. Their safety record is showing mixed signals so far, but will presumably improve over time. The extensive use of artificial intelligence in the interpretation of sensor data and car control systems may also spur its usage in other settings. The self-driving car may take many more years to become fully established on our roads, but the science fiction world where they become the norm does not seem quite so far off as it once did.
