Self-driving car

Organizations such as the global standards body SAE International (SAE) have proposed terminology to describe technical capabilities. However, most terms have no standard definition and are employed variously by vendors and others. Proposals to adopt aviation automation terminology for cars has also not prevailed. The first consideration is the operational design domain (ODD). Vendors have taken a variety of approaches to the self-driving problem. Tesla's approach is to allow their "full self-driving" (FSD) system to be used in all ODDs as a Level 2 (hands/on, eyes/on) ADAS. Waymo picked specific ODDs (city streets in Phoenix and San Francisco) for their Level 5 robotaxi service. Mercedes Benz offers Level 3 service in Las Vegas in highway traffic jams at speeds up to . Mobileye's SuperVision system offers hands-off/eyes-on driving on all road types at speeds up to . GM's hands-free Super Cruise operates on specific roads in specific conditions, stopping or returning control to the driver when ODD changes. In 2024 the company announced plans to expand road coverage from 400,000 miles to . Ford's BlueCruise hands-off system operates on of US divided highways. Names such as AutonoDrive, PilotAssist, "Full-Self Driving" or DrivePilot are used even though the products offer an assortment of features that may not match the names. Despite offering a system dubbed Full Self-Driving, Tesla stated that its system did not autonomously handle all driving tasks. In the United Kingdom, a fully self-driving car is defined as a car so registered, rather than one that supports a specific feature set. The Association of British Insurers claimed that the usage of the word autonomous in marketing was dangerous because car ads make motorists think "autonomous" and "autopilot" imply that the driver can rely on the car to control itself, even though they do not. Concepts The following are useful in understanding the various definitions and criteria in use for self-driving cars. Driving systems Advanced driver-assistance systems (ADAS) automate specific driving features such as Forward Collision Warning (FCW), Automatic Emergency Braking (AEB), Lane Departure Warning (LDW), Lane Keeping Assistance (LKA) or Blind Spot Warning (BSW). An ADAS requires a human driver to handle tasks that the ADAS does not support. ADAS contrasts to an automated driving system (ADS), which would be classified by SAE J3016 as Level 3 or higher. Autonomy versus automation Autonomy implies that an automation system is under the control of the vehicle rather than a driver. Automation is function-specific, handling issues such as speed control, but leaves broader decision-making to the driver. The European car safety performance assessment programme Euro NCAP defines "autonomous" as "the system acts independently of the driver to avoid or mitigate the accident". In Europe, the words automated and autonomous can be used together. For instance, under Regulation (EU) 2019/2144: • "automated vehicle" means a vehicle that can move without continuous driver supervision, but that driver intervention is still expected or required in the operational design domains (ODD); According to SAE J3016, Self-driving The Union of Concerned Scientists defined self-driving as "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." The British Automated and Electric Vehicles Act 2018 defines a vehicle as "driving itself" if the vehicle is "not being controlled, and does not need to be monitored, by an individual". Another British government definition stated, "Self-driving vehicles are vehicles that can safely and lawfully drive themselves". British definitions In British English, the word "automated" has several meanings, such as in the sentence: "Thatcham also found that the automated lane keeping systems could only meet two out of the twelve principles required to guarantee safety, going on to say they cannot, therefore, be classed as 'automated driving', preferring 'assisted driving'". The first occurrence of the "automated" word refers to a UNECE automated system, while the second refers to the British legal definition of an automated vehicle. British law interprets the meaning of "automated vehicle" based on the interpretation section related to a vehicle "driving itself" and an insured vehicle. In November 2023 the British Government introduced the Automated Vehicles Bill. It proposed definitions for related terms: • Self-driving: "A vehicle 'satisfies the self-driving test' if it is designed or adapted with the intention that a feature of the vehicle will allow it to travel autonomously, and it is capable of doing so, by means of that feature, safely and legally." • Autonomy: A vehicle travels "autonomously" if it is controlled by the vehicle, and neither the vehicle nor its surroundings are monitored by a person who can intervene. • Control: control of vehicle motion. • Safe: a vehicle that conforms to an acceptably safe standard. • Legal: a vehicle that offers an acceptably low risk of committing a traffic infraction. SAE classification is classified as an SAE Level 2 system. In 2014, the association for transport and other industries SAE International published a six-level classification system – ranging from fully manual to fully automated – as J3016, Taxonomy and Definitions for Terms Related to On-Road Motor Vehicle Automated Driving Systems; the details are revised occasionally. The system was adopted by the United Nations Economic Commission for Europe (UNECE) in the form of UN Regulation No. 157, which came into force in January 2021, and used in over 50 countries. After SAE updated its classification in 2016, (J3016_201609), the US National Highway Traffic Safety Administration (NHTSA) adopted the SAE standard. The classification is a topic of debate, having been criticized for its technological focus, with various revisions proposed. It has been argued that the structure of the levels suggests that automation increases linearly and that more automation is better, which may not be the case. SAE Levels also do not account for changes that may be required to infrastructure and road user behavior. Automation levels The SAE classification of Levels is based on the role of the driver, rather than the vehicle's capabilities, although these are related in the form of a "driving mode" (aka driving scenario). The mode is determined by both an operational design domain (ODD) and a "dynamic driving requirement." The ODD is the circumstance in which the car is driving, and the driving requirement is what the system must do while remaining safe within the boundaries of that ODD. These two things define the SAE Level. Cars may therefore switch levels according to the driving mode. A human analogy might be a mode in which you can stand unassisted on one leg (the requirement) if you are on the ground (the ODD), but would need support when standing on a tightrope. Above Level 1, level differences are related to how responsibility for safe movement is divided/shared between the advanced driver-assistance system (ADAS) and driver, rather than specific driving features (see Concepts). A car might therefore have Level 3 capability for lane management (the driving requirement) up to 100 km/h (the ODD), but Level 2 at speeds faster than that. Or it might have Level 4 for driverless navigation on designated highways, but Level 2 on urban side roads. ADAS that are considered Level 1 are adaptive cruise control, emergency brake assist, automatic emergency brake assist, lane-keeping, and lane centering. ADAS that are considered Level 2 are: highway assist, autonomous obstacle avoidance, and autonomous parking. At Level 3 and above, if a car is driving and the ODD changes - for instance, a clear highway becomes a construction zone with hand-signals from a worker - the system must recognize that it is exiting its ODD and fall back accordingly, including asking for driver intervention if necessary. A fully self-driving car would therefore stay at Level 5 for all dynamic driving requirements in all circumstances (speeds, roads, terrains, visibilities, obstacle types, jurisdictions, etc.), while a car with no ADAS controlling the driving requirements would rely on the human driver for that, and stay at Level 0 regardless of the domain. Mobileye terminology Mobileye CEO Amnon Shashua and CTO Shai Shalev-Shwartz proposed an alternative taxonomy for autonomous driving systems, claiming that a more consumer-friendly approach was needed. Its categories reflect the amount of driver engagement that is required. Some vehicle makers have informally adopted some of the terminology involved, while not formally committing to it. Eyes-on/hands-on The first level, hands-on/eyes-on, implies that the driver is fully engaged in operating the vehicle, but is supervised by the system, which intervenes according to the features it supports (e.g., adaptive cruise control, automatic emergency braking). The driver is entirely responsible, with hands on the wheel and eyes on the road. Eyes-on/hands-off Eyes-on/hands-off allows the driver to let go of the wheel. The system drives, the driver monitors, and remains prepared to resume control as needed. Eyes-off/hands-off Eyes-off/hands-off means that the driver can stop monitoring the system, leaving the system in full control. Eyes-off requires that no errors be reproducible (not triggered by exotic transitory conditions) or frequent, that speeds are contextually appropriate (e.g., on limited-access roads), and that the system handles typical maneuvers (e.g., getting cut off by another vehicle). The automation level could vary according to the road (e.g., eyes-off on freeways, eyes-on on side streets). No driver The highest level does not require a human driver in the car: monitoring is done either remotely (telepresence) or not at all. Safety MRM A critical requirement for the higher two levels is that the vehicle be able to conduct a Minimum Risk Maneuver and stop safely out of traffic without driver intervention. == Technology ==

Architecture The perception system processes visual and audio data from outside and inside the car to create a local model of the vehicle, the road, traffic, traffic controls and other observable objects, and their relative motion. The control system then takes actions to move the vehicle, considering the local model, road map, and driving regulations. Educational robotics modules, such as the NSDL-hosted "Thinking Robotics" activity, show how foundational decision-making and sensor-based navigation concepts are used in autonomous vehicles. Several classifications have been proposed to describe ADAS technology. One proposal is to adopt these categories: navigation, path planning, perception, and car control. Vendors such as Tesla and Motional have opted for monolithic "end to end" (E2E) neural networks, moving away from modular systems that separate perception from control. Navigation Navigation involves the use of maps to define a path between origin and destination. Hybrid navigation is the use of multiple navigation systems. Some systems use basic maps, relying on perception to deal with anomalies. Such a map understands which roads lead to which others, whether a road is a freeway, a highway, are one-way, etc. Other systems require highly detailed maps, including lane maps, obstacles, traffic controls, etc. Perception ACs need to be able to perceive the world around them. Supporting technologies include combinations of cameras, LiDAR, radar, audio, and ultrasound, GPS, and inertial measurement. Deep neural networks are used to analyse inputs from these sensors to detect and identify objects and their trajectories. Some systems use Bayesian simultaneous localization and mapping (SLAM) algorithms. Another technique is detection and tracking of other moving objects (DATMO), used to handle potential obstacles. Other systems use roadside real-time locating system (RTLS) technologies to aid localization. Tesla's "vision only" system uses eight cameras, without LIDAR or radar, to create its bird's-eye view of the environment. Path planning Path planning finds a sequence of segments that a vehicle can use to move from origin to destination. Techniques used for path planning include graph-based search and variational-based optimization techniques. Graph-based techniques can make harder decisions such as how to pass another vehicle/obstacle. Variational-based optimization techniques require more stringent restrictions on the vehicle's path to prevent collisions. The large scale path of the vehicle can be determined by using a voronoi diagram, an occupancy grid mapping, or a driving corridor algorithm. The latter allows the vehicle to locate and drive within open space that is bounded by lanes or barriers. Maps Maps are necessary for navigation. Map sophistication varies from simple graphs that show which roads connect to each other, with details such as one-way vs two-way, to those that are highly detailed, with information about lanes, traffic controls, roadworks, and more. One issue with highly detailed maps is updating them as the world changes. Vehicles that can operate with less-detailed maps do not require frequent updates or geo-fencing. Sensors Sensors are necessary for the vehicle to properly respond to the driving environment. Sensor types include cameras, LiDAR, ultrasound, and radar. Control systems typically combine data from multiple sensors. Multiple sensors can provide a more complete view of the surroundings and can be used to cross-check each other to correct errors. For example, radar can image a scene in, e.g., a nighttime snowstorm, that defeats cameras and LiDAR, albeit at reduced precision. After experimenting with radar and ultrasound, Tesla adopted a vision-only approach, asserting that humans drive using only vision, and that cars should be able to do the same, while citing the lower cost of cameras versus other sensor types. By contrast, Waymo makes use of the higher resolution of LiDAR sensors and cites the declining cost of that technology. Drive by wire Drive by wire is the use of electrical or electro-mechanical systems for performing vehicle functions such as steering or speed control that are traditionally achieved by mechanical linkages. Driver monitoring Driver monitoring is used to assess the driver's attention and alertness. Techniques in use include eye monitoring, and requiring the driver to maintain torque on the steering wheel. It attempts to understand driver status and identify dangerous driving behaviors. Vehicle communication Vehicles can potentially benefit from communicating with others to share information about traffic, road obstacles, to receive map and software updates, etc. while ISO/TC 204 specifies information, communication and control systems in surface transport. International standards have been developed for ADAS functions, connectivity, human interaction, in-vehicle systems, management/engineering, dynamic map and positioning, privacy and security. Rather than communicating among vehicles, they can communicate with road-based systems to receive similar information. Software update Software controls the vehicle, and can provide entertainment and other services. Over-the-air updates can deliver bug fixes and additional features over the internet. Software updates are one way to accomplish recalls that in the past required a visit to a service center. In March 2021, the UNECE regulation on software update and software update management systems was published. Safety model A safety model is software that attempts to formalize rules that ensure that ACs operate safely. IEEE is attempting to forge a standard for safety models as "IEEE P2846: A Formal Model for Safety Considerations in Automated Vehicle Decision Making". In 2022, a research group at National Institute of Informatics (NII, Japan) enhanced Mobileye's Reliable Safety System as "Goal-Aware RSS" to enable RSS rules to deal with complex scenarios via program logic. Notification In accordance with the proposed SAE J3134 "ADS Marker Lamps" specification, the US has standardized the use of turquoise lights to inform other drivers that a vehicle is driving autonomously. It will be used in the 2026 Mercedes-Benz EQS and S-Class sedans with Drive Pilot, an SAE Level 3 driving system. As of 2023, the turquoise light had not been standardized in China or by the UNECE. Artificial Intelligence Artificial intelligence (AI) plays a pivotal role in the development and operation of autonomous vehicles (AVs), enabling them to perceive their surroundings, make decisions, and navigate safely without human intervention. AI algorithms empower AVs to interpret sensory data from various onboard sensors, such as cameras, LiDAR, radar, and GPS, to understand their environment and improve its technological ability and overall safety over time. == Challenges ==

Obstacles The primary obstacle to ACs is the advanced software and mapping required to make them work safely across the wide variety of conditions that drivers experience. In addition to handling day/night driving in good and bad weather Other obstacles include cost, liability, consumer reluctance, ethical dilemmas, security, privacy, and legal/regulatory framework. Further, AVs could automate the work of professional drivers, eliminating many jobs, which could slow acceptance. Concerns Deceptive marketing Tesla calls its Level 2 ADAS "Full Self-Driving (FSD) Beta". At that time, Mercedes-Benz rejected the claims and stopped its "self-driving car" ad campaign that had been running. In August 2022, the California Department of Motor Vehicles (DMV) accused Tesla of deceptive marketing practices. With the Automated Vehicles Bill (AVB) self-driving car-makers could face prison for misleading adverts in the United-Kingdom. Security In the 2020s, concerns over ACs' vulnerability to cyberattacks and data theft emerged. Espionage In 2018 and 2019, former Apple engineers were charged with stealing information related to Apple's self-driving car project. In 2021 the United States Department of Justice (DOJ) accused Chinese security officials of coordinating a hacking campaign to steal information from government entities, including research related to autonomous vehicles. China has prepared "the Provisions on Management of Automotive Data Security (Trial) to protect its own data". Cellular Vehicle-to-Everything technologies are based on 5G wireless networks. , the US Congress was considering the possibility that imported Chinese AC technology could facilitate espionage. Testing of Chinese automated cars in the US has raised concern over which US data are collected by Chinese vehicles to be stored in China and any link with the Chinese communist party. Driver communications ACs complicate the need for drivers to communicate with each other, e.g., to decide which car enters an intersection first. In an AC without a driver, traditional means such as hand signals do not work (no driver, no hands). Behavior prediction ACs must be able to predict the behavior of possibly moving vehicles, pedestrians, etc, in real time in order to proceed safely. Handover The ADAS has to be able to safely accept control from and return control to the driver. Robotaxis operating in San Francisco received pushback over perceived safety risks. Automatic elevators were invented in 1900, but did not become common until operator strikes and trust was built with advertising and features such as an emergency stop button. However, with repeated use of autonomous driving functions, drivers' behavior and trust in autonomous vehicles gradually improved and both entered a more stable state. At the same time, this also improved the performance and reliability of the vehicle in complex conditions, thereby increasing public trust. Economics Autonomy also presents various political and economic implications. The transportation sector holds significant sway in many political and economic landscapes. For instance, many US states generate much annual revenue from transportation fees and taxes. The advent of self-driving cars could profoundly affect the economy by potentially altering state tax revenue streams. Furthermore, the transition to autonomous vehicles might disrupt employment patterns and labor markets, particularly in industries heavily reliant on driving professions. Additionally, taxi and delivery drivers represented approximately 370,400 positions, and bus drivers constituted a workforce of over 680,000. Collectively, this amounts to a conceivable displacement of nearly 2.9 million jobs, surpassing the job losses experienced in the 2008 Great Recession. Ethical issues Equity and inclusion The prominence of certain demographic groups within the tech industry inevitably shapes the trajectory of autonomous vehicle (AV) development, potentially perpetuating existing inequalities. Pedestrian detection In 2019, research from Georgia Tech revealed that autonomous vehicle detection systems were generally five percent less effective at recognizing darker-skinned individuals. This accuracy gap persisted despite adjustments for environmental variables like lighting and visual obstructions. Rationale for liability Standards for liability have yet to be adopted to address crashes and other incidents. Liability could rest with the vehicle occupant, its owner, the vehicle manufacturer, or even the ADAS technology supplier, possibly depending on the circumstances of the crash. Additionally, the infusion of ArtificiaI Intelligence technology in autonomous vehicles adds layers of complexity to ownership and ethical dynamics. Given that AI systems are inherently self-learning, a question arises of whether accountability should rest with the vehicle owner, the manufacturer, or the AI developer. Trolley problem The trolley problem is a thought experiment in ethics. Adapted for ACs, it considers an AC carrying one passenger confronting a pedestrian who steps in its way. The ADAS notionally has to choose between killing the pedestrian or swerving into a wall, killing the passenger. Possible frameworks include deontology (formal rules) and utilitarianism (harm reduction). One public opinion survey reported that harm reduction was preferred, except that passengers wanted the vehicle to prefer them, while pedestrians took the opposite view. Utilitarian regulations were unpopular. Additionally, cultural viewpoints exert substantial influence on shaping responses to these ethical quandaries. Another study found that cultural biases impact preferences in prioritizing the rescue of certain individuals over others in car accident scenarios. Road infrastructure ACs make use of road infrastructure (e.g., traffic signs, turn lanes) and may require modifications to that infrastructure to fully achieve their safety and other goals. In March 2023, the Japanese government unveiled a plan to set up a dedicated highway lane for ACs. In April 2023, JR East announced their challenge to raise their self-driving level of Kesennuma Line bus rapid transit (BRT) in rural area from the current Level 2 to Level 4 at . == Testing ==

Approaches ACs can be tested via digital simulations, in a controlled test environment, and/or on public roads. Road testing typically requires some form of permit or a commitment to adhere to acceptable operating principles. For example, New York requires a test driver to be in the vehicle, prepared to override the ADAS as necessary. 2010s and disengagements 's self-driving car, navigating public streets in Mountain View, California in 2017 In California, self-driving car manufacturers are required to submit annual reports describing how often their vehicles autonomously disengaged from autonomous mode. This is one measure of system robustness (ideally, the system should never disengage). 2020s Disengagement definitions Reporting companies use varying definitions of what qualifies as a disengagement, and such definitions can change over time. Standards In April 2021, WP.29 GRVA proposed a "Test Method for Automated Driving (NATM)". In October 2021, Europe's pilot test, L3Pilot, demonstrated ADAS for cars in Hamburg, Germany, in conjunction with ITS World Congress 2021. SAE Level 3 and 4 functions were tested on ordinary roads. In November 2022, an International Standard ISO 34502 on "Scenario based safety evaluation framework" was published. Collision avoidance In April 2022, collision avoidance testing was demonstrated by Nissan. Waymo published a document about collision avoidance testing in December 2022. Simulation and validation In September 2022, Biprogy released Driving Intelligence Validation Platform (DIVP) as part of Japanese national project "SIP-adus", which is interoperable with Open Simulation Interface (OSI) of ASAM. Toyota In November 2022, Toyota demonstrated one of its GR Yaris test cars, which had been trained using professional rally drivers. Toyota used its collaboration with Microsoft in FIA World Rally Championship since the 2017 season. Pedestrian reactions In 2023 David R. Large, senior research fellow with the Human Factors Research Group at the University of Nottingham, disguised himself as a car seat in a study to test people's reactions to driverless cars. He said, "We wanted to explore how pedestrians would interact with a driverless car and developed this unique methodology to explore their reactions." The study found that, in the absence of someone in the driving seat, pedestrians trust certain visual prompts more than others when deciding whether to cross the road. == Safety ==

A meta-analysis published in Nature Communications in 2024 compared various sources of safety data for autonomous (AV) and human-driven vehicles (HDV). This collected 2,100 AV and 35,133 HDV incident records which accurately reflected the incident details. Some AVs in the comparisons (such as robotaxis) were effectively autonomous, while others were equipped with Advanced Driving Systems (ADS) or Advanced Driver Assistance Systems (ADAS). The study concluded that AVs are safer in most circumstances, and AVs had far fewer crashes involving pedestrians (3% against 15%) per mile travelled. The comparison also highlighted some other disaparities: while AVs were found to be significantly less likely to crash in heavy rain or fog than an unassisted human, they were more than five times more vulnerable to collisions at dawn and dusk. On 20 January 2016, the first of five known fatal crashes of a Tesla with Autopilot occurred, in China's Hubei province. Initially, Tesla stated that the vehicle was so badly damaged from the impact that their recorder was not able to determine whether the car had been on Autopilot at the time. However, the car failed to take evasive action. Another fatal Autopilot crash occurred in May in Florida in a Tesla Model S that crashed into a tractor-trailer. In a civil suit between the father of the driver killed and Tesla, Tesla documented that the car had been on Autopilot. According to Tesla, "neither Autopilot nor the driver noticed the white side of the tractor-trailer against a brightly lit sky, so the brake was not applied." Tesla claimed that this was Tesla's first known Autopilot death in over with Autopilot engaged. Tesla claimed that on average one fatality occurs every across all vehicle types in the US. However, this number also includes motorcycle/pedestrian fatalities. The ultimate National Transportation Safety Board (NTSB) report concluded Tesla was not at fault; the investigation revealed that for Tesla cars, the crash rate dropped by 40 percent after Autopilot was installed. In February 2025, a Tesla Cybertruck crashed while in Full-Self Driving mode, raising concerns about autonomous driving and prompting an investigation from Tesla who said the crash would be probed ""in line with standard protocol when any of our electric vehicles are involved in an accident while in FSD mode." Google Waymo In June 2015, Google confirmed that 12 vehicles had suffered collisions as of that date. Eight involved rear-end collisions at a stop sign or traffic light, in two of which the vehicle was side-swiped by another driver, one in which another driver rolled a stop sign, and one where a driver was controlling the car manually. In July 2015, three employees suffered minor injuries when their vehicle was rear-ended by a car whose driver failed to brake. This was the first collision that resulted in injuries. According to Google Waymo's accident reports as of early 2016, their test cars had been involved in 14 collisions, of which other drivers were at fault 13 times, although in 2016 the car's software caused a crash. On 14 February 2016 a Google vehicle attempted to avoid sandbags blocking its path. During the maneuver it struck a bus. Google stated, "In this case, we clearly bear some responsibility, because if our car hadn't moved, there wouldn't have been a collision." Google characterized the crash as a misunderstanding and a learning experience. No injuries were reported. Uber's Advanced Technologies Group (ATG) In March 2018, Elaine Herzberg died after she was hit by an AC tested by Uber's Advanced Technologies Group (ATG) in Arizona. A safety driver was in the car. Herzberg was crossing the road about 400 feet from an intersection. Some experts said a human driver could have avoided the crash. Arizona governor Doug Ducey suspended the company's ability to test its ACs citing an "unquestionable failure" of Uber to protect public safety. Uber also stopped testing in California until receiving a new permit in 2020. The NTSB's final report determined that the immediate cause of the accident was that safety driver Rafaela Vasquez failed to monitor the road, because she was distracted by her phone, but that Uber's "inadequate safety culture" contributed. The report noted that the victim had "a very high level" of methamphetamine in her body. The board called on federal regulators to carry out a review before allowing automated test vehicles to operate on public roads. In September 2020, Vasquez pled guilty to endangerment and was sentenced to three years' probation. NIO Navigate on Pilot On 12 August 2021, a 31-year-old Chinese man was killed after his NIO ES8 crashed in a tunnel. NIO's self-driving feature was in beta and could not deal with static obstacles. The vehicle's manual clearly stated that the driver must take over near construction sites. Lawyers of the deceased's family questioned NIO's private access to the vehicle, which they argued did not guarantee the integrity of the data. Pony.ai In November 2021, the California Department of Motor Vehicles (DMV) notified Pony.ai that it was suspending its testing permit following a reported collision in Fremont on 28 October. In May 2022, DMV revoked Pony.ai's permit for failing to monitor the driving records of its safety drivers. Cruise In April 2022, Cruise's testing vehicle was reported to have blocked a fire engine on emergency call, and sparked questions about its ability to handle unexpected circumstances. Ford In February 2024, a driver using the Ford BlueCruise hands-free driving feature struck and killed the driver of a stationary car with no lights on in the middle lane of a freeway in Texas. In March 2024, a drunk driver who was speeding, holding her cell phone, and using BlueCruise on a Pennsylvania freeway struck and killed two people who had been driving two cars. The first car had become disabled and was on the left shoulder with part of the car in the left driving lane. Total incidents The US National Highway Traffic Safety Administration began mandating incident reports from autonomous vehicle companies in June 2021. Some reports cite incidents from as early as August 2019, with current data available through June 17, 2024. There have been a total of 3,979 autonomous vehicle incidents (both ADS and ADAS) reported during this timeframe. 2,146 of those incidents (53.9%) involved Tesla vehicles. == Public opinion surveys ==

2010s In a 2011 online survey of 2,006 US and UK consumers, 49% said they would be comfortable using a "driverless car". A 2012 survey of 17,400 vehicle owners found 37% who initially said they would be interested in purchasing a "fully autonomous car". However, that figure dropped to 20% if told the technology would cost US$3,000 more. In a 2012 survey of about 1,000 German drivers, 22% had a positive attitude, 10% were undecided, 44% were skeptical and 24% were hostile. A 2013 survey of 1,500 consumers across 10 countries found 57% "stated they would be likely to ride in a car controlled entirely by technology that does not require a human driver", with Brazil, India and China the most willing to trust automated technology. In a 2014 US telephone survey, over three-quarters of licensed drivers said they would consider buying a self-driving car, rising to 86% if car insurance were cheaper. 31.7% said they would not continue to drive once an automated car was available. In 2015, a survey of 5,000 people from 109 countries reported that average respondents found manual driving the most enjoyable. 22% did not want to pay more money for autonomy. Respondents were found to be most concerned about hacking/misuse, and were also concerned about legal issues and safety. Finally, respondents from more developed countries were less comfortable with their vehicle sharing data. The survey reported consumer interest in purchasing an AC, stating that 37% of surveyed current owners were either "definitely" or "probably" interested. In a 2016 US survey of 1,584 people, "66 percent of respondents said they think autonomous cars are probably smarter than the average human driver". People were worried about safety and hacking risk. Nevertheless, only 13% of the interviewees saw no advantages in this new kind of cars. In a 2017 survey of 4,135 US adults found that many Americans anticipated significant impacts from various automation technologies including the widespread adoption of automated vehicles. In 2019, a test of a method of surveying opinions about self-driving cars, termed the Autonomous Vehicle Acceptance Model (AVAM), found that users were less accepting of high autonomy levels and displayed significantly lower intention to use highly autonomous vehicles. Additionally, partial autonomy (regardless of level) was perceived as requiring uniformly higher driver engagement (usage of hands, feet and eyes) than full autonomy. In the 2020s In 2022, a survey reported that only a quarter (27%) of the world's population would feel safe in self-driving cars. In July 2024, a biannual YouGov survey asking British adults how safe or unsafe people would feel in a self-driving car found that 4% would feel "Very safe" while 37% would feel "Very unsafe". == Regulation ==

While various models of car may be described by their manufacturers as being at a certain Level, the SAE technical specification means that a car can move between levels depending on the driving task and the circumstances it is operating in at a given time. Vehicles operating below Level 5 still offer many advantages. Level 2 – Partial automation SAE Level 2 features are available as part of the ADAS systems in many vehicles. In the US, 50% of new cars provide driver assistance for both steering and speed. Ford started offering BlueCruise service on certain vehicles in 2022; the system is named ActiveGlide in Lincoln vehicles. The system provided features such as lane centering, street sign recognition, and hands-free highway driving on more than 130,000 miles of divided highways. The 2022 1.2 version added features including hands-free lane changing, in-lane repositioning, and predictive speed assist. In April 2023 BlueCruise was approved in the UK for use on certain motorways, starting with 2023 models of Ford's electric Mustang Mach-E SUV. Tesla's Autopilot and its Full Self-Driving (FSD) ADAS suites are available on all Tesla cars since 2016. FSD offers highway and street driving (without geofencing), navigation/turn management, steering, and dynamic cruise control, collision avoidance, lane-keeping/switching, emergency braking, obstacle avoidance, but still requires the driver to remain ready to control the vehicle at any moment. Its driver management system combines eye tracking with monitoring pressure on the steering wheel to ensure that drives are both eyes on and hands on. Tesla's FSD rewrite V12 (released in March 2024) uses a single deep learning transformer model for all aspects of perception, monitoring, and control. It relies on its eight cameras for its vision-only perception system, without use of LiDAR, radar, or ultrasound. Level 3 – Conditional automation , two car manufacturers have sold or leased Level 3 cars: Honda in Japan, and Mercedes in Germany, Nevada and California. Mercedes Drive Pilot has been available on the EQS and S-class sedan in Germany since 2022, and in California and Nevada since 2023. Development Honda continued to enhance its Level 3 technology. As of 2023, 80 vehicles with Level 3 support had been sold. Mercedes-Benz received authorization in early 2023 to pilot its Level 3 software in Las Vegas. California also authorized Drive Pilot in 2023. BMW commercialized its AC in 2021. In 2023 BMW stated that its Level-3 technology was nearing release. It would be the second manufacturer to deliver Level-3 technology, but the only one with a Level 3 technology which works in the dark. In 2023, in China, IM Motors, Mercedes, and BMW obtained authorization to test vehicles with Level 3 systems on motorways. In September 2021, Stellantis presented its findings from its Level 3 pilot testing on Italian highways. Stellantis's Highway Chauffeur claimed Level 3 capabilities, as tested on the Maserati Ghibli and Fiat 500X prototypes. Polestar, a Volvo Cars' brand, announced in January 2022 its plan to offer Level 3 autonomous driving system in the Polestar 3 SUV, a Volvo XC90 successor, with technologies from Luminar Technologies, Nvidia, and Zenseact. In January 2022, Bosch and the Volkswagen Group subsidiary CARIAD released a collaboration for autonomous driving up to Level 3. This joint development targets Level 4 capabilities. Hyundai Motor Company is enhancing cybersecurity of connected cars to offer a Level 3 self-driving Genesis G90. Kia and Hyundai Korean car makers delayed their Level 3 plans, and will not deliver Level 3 vehicles in 2023. Level 4 – High automation In 2024, companies such as Waymo began offering robotaxi services in parts of the US with fully autonomous vehicles without safety drivers. These services all operate at a loss as of 2025, with operating costs of about , compared to for personal cars. The consultancy McKinsey estimated that bringing costs down to less than would take until 2035. In April 2023 in Japan, a Level 4 protocol became part of the amended Road Traffic Act. ZEN drive Pilot Level 4 made by AIST operates there. Development In July 2020, Toyota started public demonstration rides on Lexus LS (fifth generation) based TRI-P4 with Level 4 capability. In August 2021, Toyota operated a potentially Level 4 service using e-Palette around the Tokyo 2020 Olympic Village. In September 2020, Mercedes-Benz introduced world's first commercial Level 4 Automated Valet Parking (AVP) system named Intelligent Park Pilot for its new S-Class. In November 2022, Germany's Federal Motor Transport Authority (KBA) approved the system for use at Stuttgart Airport. In September 2021, Cruise, General Motors, and Honda started a joint testing programme, using Cruise AV. In 2023, the Origin was put on indefinite hold following Cruise's loss of its operating permit. In January 2023, Holon announced an autonomous shuttle during the 2023 Consumer Electronics Show (CES). The company claimed that the vehicle is the world's first Level 4 shuttle built to automotive standard. ==See also==