Experiments have been conducted on automating driving since at least the 1920s; trials began in the 1950s. The rst truly automated car was developed in 1977, by Japan's Tsukuba Mechanical Engineering Laboratory. The vehicle tracked white street markers, which were interpreted by two cameras on the vehicle, using an analog computer for signal processing. The vehicle reached speeds up to 30 kilometres per hour (19 mph), with the support of an elevated rail.
Autonomous prototype cars appeared in the 1980s, with Carnegie Mellon University's Navlab and ALV
projects funded by DARPA starting in 1984 and Mercedes-Benz and Bundeswehr University Munich's EUREKA Prometheus Project in 1987. By 1985, the ALV had demonstrated self-driving speeds on two-lane
roads of 31 kilometers per hour (19 mph) with obstacle avoidance added in 1986 and o -road driving in
day and nighttime conditions by 1987. From the 1960s through the second DARPA Grand Challenge in
2005, automated vehicle research in the U.S. was primarily funded by DARPA, the US Army and the U.S.
Navy yielding incremental advances in speeds, driving competence in more complex conditions, controls and sensor systems. Companies and research organizations have developed prototypes.
The U.S. allocated $650 million in 1991 for research on the National Automated Highway System, which
demonstrated automated driving through a combination of automation, embedded in the highway with
automated technology in vehicles and cooperative networking between the vehicles and with the highway infrastructure. The program concluded with a successful demonstration in 1997 but without clear direction or funding to implement the system on a larger scale. Partly funded by the National Automated Highway System and DARPA, the Carnegie Mellon University Navlab drove 4,584 kilometers (2,848 mi) across America in 1995, 4,501 kilometers (2,797 mi) or 98% of it autonomously. Navlab's record achievement stood unmatched for two decades until 2015 when Delphi improved it by piloting an Audi, augmented with Delphi technology, over 5,472 kilometers (3,400 mi) through 15 states while remaining in self-driving mode 99% of the time. In 2015, the US states of Nevada, Florida, California, Virginia, and Michigan, together with Washington, D.C., allowed the testing of automated cars on public roads.
In 2017, Audi stated that its latest A8 would be automated at speeds of up to 60 kilometres per hour (37
mph) using its \Audi AI." The driver would not have to do safety checks such as frequently gripping the
steering wheel. The Audi A8 was claimed to be the rst production car to reach level 3 automated driving,
and Audi would be the rst manufacturer to use laser scanners in addition to cameras and ultrasonic sensors for their system.
In November 2017, Waymo announced that it had begun testing driverless cars without a safety driver in the driver position; however, there is still an employee in the car. In July 2018, Waymo announced that its test vehicles had traveled in automated mode for over 8,000,000 miles (13,000,000 km), increasing by 1,000,000 miles (1,600,000 kilometers) per month (Fig. 7).

Terminology

There is some inconsistency in terminology used in the self-driving car industry. Various organizations have proposed to de ne an accurate and consistent vocabulary. Such confusion has been documented in SAE J3016 which states that \Some vernacular usages associate autonomous speci cally with full driving automation (level 5), while other usages apply it to all levels of driving automation, and some state legislation has de ned it to correspond approximately to any ADS at or above level 3 (or to any vehicle equipped with such an ADS)."

Words de nition and safety considerations

Modern vehicles provide partly automated features such as keeping the car within its lane, speed controls
or emergency braking. Nonetheless, di erences remain between a fully autonomous self-driving car on one hand and driver assistance technologies on the other hand. According to the BBC, confusion between those concepts leads to deaths.
Association of British Insurers considers the usage of the word autonomous in marketing for modern cars to be dangerous, because car ads make motorists think `autonomous' and `autopilot' means a vehicle can drive itself, when they still rely on the driver to ensure safety. Technology alone still is not able to drive the car.
When some car makers suggest or claim vehicles are self-driving, when they are only partly automated,
drivers risk becoming excessively con dent, leading to crashes, while fully self-driving cars are still a long
way o in the UK.

Autonomous vs. automated

Autonomous means self-governing. Many historical projects related to vehicle automation have been automated (made automatic) subject to a heavy reliance on arti cial aids in their environment, such as magnetic strips. Autonomous control implies satisfactory performance under signi cant uncertainties in the environment and the ability to compensate for system failures without external intervention.
One approach is to implement communication networks both in the immediate vicinity (for collision avoidance) and farther away (for congestion management). Such outside in uences in the decision process reduce an individual vehicle's autonomy, while still not requiring human intervention.
Wood et al. (2012) wrote, \This Article generally uses the term `autonomous,' instead of the term `automated.' " The term \autonomous" was chosen \because it is the term that is currently in more widespread use (and thus is more familiar to the general public). However, the latter term is arguably more accurate. `Automated' connotes control or operation by a machine, while `autonomous' connotes acting alone or independently. Most of the vehicle concepts (that we are currently aware of) have a person in the driver's seat, utilize a communication connection to the Cloud or other vehicles, and do not independently select either destinations or routes for reaching them. Thus, the term `automated' would more accurately describe these vehicle concepts." As of 2017, most commercial projects focused on automated vehicles that did not communicate with other vehicles or with an enveloping management regime.
Put in the words of one Nissan engineer, \A truly autonomous car would be one where you request it to
take you to work and it decides to go to the beach instead."
EuroNCAP de nes autonomous in \Autonomous Emergency Braking" as: \the system acts independently
of the driver to avoid or mitigate the accident." which implies the autonomous system is not the driver.

Autonomous versus cooperative

To make a car travel without any driver embedded within the vehicle some system makers used a remote
driver.
But according to SAE J3016, some driving automation systems may indeed be autonomous if they perform all of their functions independently and self-suciently, but if they depend on communication and/or cooperation with outside entities, they should be considered cooperative rather than autonomous.

Self-driving car

Techemergence says.
"Self-driving" is a rather vague term with a vague meaning
-?Techemergence
PC mag de nition is: A computer-controlled car that drives itself. Also called an \autonomous vehicle"
and \driverless car," self-driving cars date back to the 1939 New York World's Fair when General Motors
predicted the development of self-driving, radio-controlled electric cars.
UCSUSA de nition is: Self-driving vehicles are cars or trucks in which human drivers are never required
to take control to safely operate the vehicle. Also known as autonomous or \driverless" cars, they combine sensors and software to control, navigate, and drive the vehicle. Currently, there are no legally operating, fully-autonomous vehicles in the United States.
NHTSA de nition is: These self-driving vehicles ultimately will integrate onto U.S. roadways by progressing
through six levels of driver assistance technology advancements in the coming years. This includes everything from no automation (where a fully engaged driver is required at all times), to full autonomy (where an automated vehicle operates independently, without a human driver).

Classi cation

A classi cation system based on six di erent levels (ranging from fully manual to fully automated systems)
was published in 2014 by SAE International, an automotive standardization body, as J3016, Taxonomy and De nitions for Terms Related to On-Road Motor Vehicle Automated Driving Systems. This classi cation
system is based on the amount of driver intervention and attentiveness required, rather than the vehicle
capabilities, although these are very loosely related. In the United States in 2013, the National Highway
Trac Safety Administration (NHTSA) released a formal classi cation system, but abandoned this system
in favor of the SAE standard in 2016. Also in 2016, SAE updated its classi cation, called J3016 201609.

Levels of driving automation

In SAE's automation level de nitions, \driving mode" means \a type of driving scenario with characteristic
dynamic driving task requirements (e.g., expressway merging, high speed cruising, low speed trac jam,
closed-campus operations, etc.)"
Level 0: Automated system issues warnings and may momentarily intervene but has no sustained vehicle
control.
Level 1 (\hands on"): The driver and the automated system share control of the vehicle. Examples are
Adaptive Cruise Control (ACC), where the driver controls steering and the automated system controls
speed; and Parking Assistance, where steering is automated while speed is under manual control. The driver must be ready to retake full control at any time. Lane Keeping Assistance (LKA) Type II is a further
example of level 1 self-driving.
Level 2 (\hands o "): The automated system takes full control of the vehicle (accelerating, braking, and
steering). The driver must monitor the driving and be prepared to intervene immediately at any time if the automated system fails to respond properly. The shorthand \hands o " is not meant to be taken literally. In fact, contact between hand and wheel is often mandatory during SAE 2 driving, to con rm that the driver is ready to intervene.
Level 3 (\eyes o "): The driver can safely turn their attention away from the driving tasks, e.g. the driver
can text or watch a movie. The vehicle will handle situations that call for an immediate response, like
emergency braking. The driver must still be prepared to intervene within some limited time, speci ed by the manufacturer, when called upon by the vehicle to do so. As an example, the 2018 Audi A8 Luxury Sedan was the rst commercial car to claim to be capable of level 3 self-driving. This particular car has a so-called Trac Jam Pilot. When activated by the human driver, the car takes full control of all aspects of driving in slow-moving trac at up to 60 kilometers per hour (37 mph). The function works only on highways with a physical barrier separating one stream of trac from oncoming trac.
Level 4 (\mind o "): As level 3, but no driver attention is ever required for safety, i.e. the driver may safely
go to sleep or leave the driver's seat. Self-driving is supported only in limited spatial areas (geofenced) or
under special circumstances, like trac jams. Outside of these areas or circumstances, the vehicle must be able to safely abort the trip, i.e. park the car, if the driver does not retake control.
Level 5 (\steering wheel optional"): No human intervention is required at all. An example would be a robotic taxi.
In the formal SAE de nition below, note in particular what happens in the shift from SAE 2 to SAE 3: the
human driver no longer has to monitor the environment. This is the nal aspect of the \dynamic driving
task" that is now passed over from the human to the automated system. At SAE 3, the human driver still
has the responsibility to intervene when asked to do so by the automated system. At SAE 4 the human
driver is relieved of that responsibility and at SAE 5 the automated system will never need to ask for an
intervention.

Legal de nition

In the district of Columbia (DC) code, \Autonomous vehicle" means a vehicle capable of navigating
District roadways and interpreting trac-control devices without a driver actively operating any of the
vehicle's control systems. The term \autonomous vehicle" excludes a motor vehicle enabled with active
safety systems or driver- assistance systems, including systems to provide electronic blind-spot assistance, crash avoidance, emergency braking, parking assistance, adaptive cruise control, lane-keep assistance, lanedeparture warning, or trac-jam and queuing assistance, unless the system alone or in combination with other systems enables the vehicle on which the technology is installed to drive without active control or monitoring by a human operator.
In the same district code, it is considered that:
An autonomous vehicle may operate on a public roadway; provided, that the vehicle:
  1. Has a manual override feature that allows a driver to assume control of the autonomous vehicle at any time; 
  2.  Has a driver seated in the control seat of the vehicle while in operation who is prepared to take control of the autonomous vehicle at any moment; and
  3. Is capable of operating in compliance with the District's applicable trac laws and motor vehicle laws and trac control devices.

Technical challenges

The challenge for driverless car designers is to produce control systems capable of analyzing sensory data in order to provide accurate detection of other vehicles and the road ahead. Modern self-driving cars generally use Bayesian simultaneous localization and mapping (SLAM) algorithms, which fuse data from multiple sensors and an o -line map into current location estimates and map updates. Waymo has developed a variant of SLAM with detection and tracking of other moving objects (DATMO), which also handles obstacles such as cars and pedestrians. Simpler systems may use roadside real-time locating system (RTLS) technologies to aid localization. Typical sensors include Lidar, stereo vision, GPS and IMU. Udacity is developing an open-source software stack. Control systems on automated cars may use Sensor Fusion, which is an approach that integrates information from a variety of sensors on the car to produce a more consistent, accurate, and useful view of the environment.
Driverless vehicles require some form of machine vision for the purpose of visual object recognition. Automated cars are being developed with deep neural networks, a type of deep learning architecture with many computational stages, or levels, in which neurons are simulated from the environment that activate the network. The neural network depends on an extensive amount of data extracted from real-life driving scenarios, enabling the neural network to \learn" how to execute the best course of action.
In May 2018, researchers from MIT announced that they had built an automated car that can navigate
unmapped roads. Researchers at their Computer Science and Arti cial Intelligence Laboratory (CSAIL)
have developed a new system, called MapLite, which allows self-driving cars to drive on roads that they
have never been on before, without using 3D maps. The system combines the GPS position of the vehicle, a "sparse topological map" such as OpenStreetMap, (i.e. having 2D features of the roads only), and a series of sensors that observe the road conditions.

Human factor challenges

Alongside the many technical challenges that autonomous cars face, there exist many human and social
factors that may impede upon the wider uptake of the technology. As things become more automated, the human users need to have trust in the automation, which can be a challenge in itself.

Testing

Testing vehicles with varying degrees of automation can be done physically, in closed environments, on public roads (where permitted, typically with a license or permit or adhering to a speci c set of operating principles) or virtually, i.e. in computer simulations. When driven on public roads, automated vehicles require a person to monitor their proper operation and \take over" when needed. Apple is currently testing self-driven cars, and has increased the number of test vehicles from 3 to 27 in January 2018, and to 45 in March 2018.
One way to assess the progress of automated vehicles is to compute the average distance driven between
\disengagements", when the automated system is turned o , typically by a human driver. In 2017, Waymo reported 63 disengagements over 352,545 miles (567,366 km) of testing, or 5,596 miles (9,006 km) on average, the highest among companies reporting such gures. Waymo also traveled more distance in total than any other. Their 2017 rate of 0.18 disengagements per 1,000 miles (1,600 km) was an improvement from 0.2 disengagements per 1,000 miles (1,600 km) in 2016 and 0.8 in 2015. In March, 2017, Uber reported an average of 0.67 miles (1.08 km) per disengagement. In the nal three months of 2017, Cruise Automation (now owned by GM) averaged 5,224 miles (8,407 km) per disruption over 62,689 miles (100,888 km). In July 2018, the rst electric driverless racing car \Robocar" completed 1.8 kilometers track, using its navigation
system and arti cial intelligence (Table 1).