Advanced driver-assistance system
Advanced driver-assistance systems, known as ADAS, are the sensors, cameras and software that help a driver drive or park. A rear camera is the simplest example, and it arrived because of an accident in a family driveway.
Cameron Gulbransen was two years old when his father backed the family SUV over him in their driveway without seeing him. Congress later put the boy's name on a 2007 law, the Cameron Gulbransen Kids Transportation Safety Act. The law led to a rule announced in March 2014 by the National Highway Traffic Safety Administration. It required rear view cameras on all new vehicles under 10,000 pounds, about 4,500 kg, by May 2018.
Rear cameras are one small piece of a much larger toolkit now built into cars. Several questions run through this documentary. How did cars learn to watch the road? How well do the systems work? Who names them, rates them and counts their crashes? And what should software do when a crash cannot be avoided?
After the Second World War, an engineer named Nathaniel Korman experimented with using radar to control the speed of a train. He had worked on radar systems during the war. His setup tied one train's speed to the speed of the train in front of it. Korman noted that the same idea could work for vehicles on the road.
Ralph Teetor applied for a US patent in 1948 for a "speed control device for resisting operation of the accelerator". Today that device is called cruise control, and the patent was granted in 1950. In 1959, General Motors displayed a concept car that used some variation of the system, with radar embedded in the front nacelles.
Production cars first carried ADAS in the 1970s, with the adoption of the anti-lock braking system. Electronic stability control, blind spot information systems, lane departure warning, adaptive cruise control and traction control count among the early systems. Mechanical changes can disturb them. Alignment adjustments or collision damage can affect these systems, so many manufacturers require an automatic reset after an alignment is performed. That fragility makes sense once you see how much the systems depend on what lies outside the car.
The line between ADAS and plain driver-assistance systems, called DAS, is drawn by data. ADAS depend on information describing the world outside the vehicle, while DAS rely on data from inside it.
Automotive imaging, LiDAR, radar, image processing, computer vision, telemetry and in-car networking all feed an ADAS. Other inputs can come from outside the vehicle's own platform. Vehicle-to-vehicle links, called V2V, carry telematics data from other cars. Vehicle-to-infrastructure links, called V2I, carry data from the infrastructure itself. Modern cars have ADAS built into their electronics. Manufacturers can add features during design or after production through over-the-air updates, known as OTA.
Preemptive priority scheduling is how an ADAS decides which task to do first. These systems count as real-time systems because they react quickly to many inputs at once and rank the incoming information to prevent crashes. Assigning those priorities wrongly can cause more harm than good.
Most ADAS features are installed by the vehicle manufacturer during production, which the industry calls OEM. A growing aftermarket sells systems that can be retrofitted, particularly for commercial fleets. The most common aftermarket type uses computer vision and a windshield-mounted camera, often combined with a dashcam. It continuously scans the road ahead and offers forward collision warning, lane departure warning and pedestrian collision warning. Because these devices record video of events, they often form part of video telematics. A telematic control unit sends the footage to a fleet management platform, where it supports driver coaching and incident analysis. Warnings like these are the simplest rung of the ADAS ladder, and the list of them is long.
The 2013 Cadillac ATS was the first model in a line of GM Cadillacs to offer vibrating seat warnings. If the car drifts out of its lane on a highway, the seat vibrates on the side of the drift. When a frontal threat is detected, the seat pulses on both sides at once.
Forward collision warning watches the speed of the car, the vehicle ahead and the open distance between them. It alerts the driver to a possible impending collision with a sound or a visual pop-up, and it does not take control. Lane departure warning also only alerts. Cameras read the lane markings, and the system speaks up when the driver partially merges into a lane without using a turn signal. A blind spot monitor adds cameras that watch the areas the driver cannot see from the seat. A rear cross-traffic alert warns of approaching traffic when the driver reverses out of a parking spot.
Driver drowsiness detection reads facial patterns, steering movement, driving habits, turn signal use and speed to judge whether the driver seems drowsy. If it suspects so, the car typically sounds a loud alert and may vibrate the seat. A driver monitoring system goes further and tracks attentiveness through eye-tracking, using infrared sensors and cameras.
Parking sensors sound faster warnings as the car gets closer to an object. They may miss low objects such as parking stops, so they typically work alongside backup cameras. The backup camera switches on automatically when the driver puts the car in reverse.
In 2010, the VW Touareg introduced the first glare-free high beam headlamp system, which used a mechanical shutter to keep light from hitting specific traffic participants. Traffic sign recognition reads the shape and color of signs, such as a stop sign or a turn ahead sign. Poor lighting, extreme weather and partial obstruction can all make it less accurate. Night vision systems use sources such as infrared sensors, GPS, LiDAR and radar to spot pedestrians and other obstacles in the dark or in heavy weather.
A ruling from the US National Highway Traffic Safety Administration pushed for sound on quiet vehicles. It said 50 percent of quiet vehicles must carry a device that sounds when the car travels below 30 km/h, or 18.6 mph, by September 2019. The goal is to warn pedestrians and cyclists that a hybrid or plug-in electric vehicle is nearby.
Collision avoidance systems, also called pre-crash systems, use small radar detectors near the front of the car. When a crash looks possible, they can sound an alarm, tense the passengers' seat belts, close a sunroof or raise reclined seats. Pedestrian protection systems act at the moment of impact. The vehicle's bonnet lifts to cushion the gap between the pedestrian and the hard engine components, which lowers the chance of a severe head injury.
Emergency driver assistant watches for a driver who has stopped driving. If the driver has not touched the accelerator, the brake or the steering wheel after a set period, the car sends audio, visual and physical signals. If the driver still does not respond, it stops the vehicle, positions it away from oncoming traffic and turns on the hazard lights. That is a rare case of the car taking full charge, but many everyday systems share the work in smaller ways.
Cruise control holds a speed the driver sets until the driver presses the brake pedal or the clutch pedal, or switches the system off. Adaptive cruise control adds a second target, the distance to the vehicle ahead. It can brake or accelerate automatically, and versions with stop and go can halt completely and then return to the chosen speed. The driver must still stay alert, because it controls only speed and following distance.
Anti-lock brakes became standard as they grew in popularity during the 1990s. An anti-lock braking system restores traction by regulating brake pressure when the car begins to skid, for example on ice.
Electronic stability control targets two failures. Understeer happens when the front wheels lack the traction to turn the car. Oversteer happens when the car turns more than intended and spins out. The system slows the car and brakes individual wheels, working with anti-lock braking and traction control. The traction control system limits tire slip by cutting power delivery. It uses the same wheel-speed sensors as anti-lock brakes and brakes any single wheel that spins faster than the others.
Hill descent control switches on when a vehicle goes faster than 15 to 20 mph on a downgrade. It pulses the brakes and controls each wheel independently to keep traction. Hill-start assist, also called hill holder, keeps the brake applied while the driver moves from the brake pedal to the gas pedal. In manual cars it also covers the move through the clutch.
Crosswind stabilization reads yaw rate, steering angle, lateral acceleration and velocity sensors to keep a car from overturning in strong side winds. It distributes the wheel load according to the speed and direction of the wind. Rain sensors detect water and can raise open windows or close convertible tops. The frequency of raindrops also sets the speed of the windshield wipers.
Lane centering uses cameras to read lane markings, and it may take over the steering if the driver is at risk of leaving the lane. UNECE regulation 79 sorts lane change functions into categories of the automatically commanded steering function, or ACSF. In Category C, the driver commands a single lateral maneuver such as a lane change. In Category D, the system indicates that a maneuver is possible but acts only after the driver confirms. In Category E, the system continuously judges whether a maneuver is possible and completes it for extended periods without further driver command.
Automatic parking takes over steering, braking and acceleration to place the car in an available spot. The driver must still watch the surroundings and be willing to take control.
Ford's BlueCruise and General Motors' Super Cruise are sold in North America as "hands-off, eyes-on" systems. Drivers may take their hands off the wheel while the system is engaged, but they must keep their eyes on the road and be ready to act at all times. Figures for Level 2 systems list 100 million hands-free miles for BlueCruise out of about 150 million traveled. Super Cruise shows 77 million hands-free miles out of about 100 million.
Major brands with Level 2 features include Lexus, Audi, BMW, Mercedes-Benz, Tesla, Volvo, Tata, Citroën, Ford, Hyundai, Kia, Mazda, Nissan, Peugeot, Mahindra and Subaru. Full Level 2 systems include Tesla's Full Self-Driving, Volvo's Pilot Assist, Comma.ai's OpenPilot and Nissan's ProPILOT Assist. Level 3 is far thinner, with Honda Sensing Elite and Mercedes-Benz Drive Pilot as the named examples. How safe any of this is gets measured in crash reports.
A forward collision prevention system could cut crashes by 29 percent, according to the national crash database in the US. The same source puts the potential reduction at 19 percent for lane keeping assistance and 9 percent for blind zone detection. The reasoning behind all three is that most road crashes come from human error.
A 2021 research report from Canalys found that approximately 33 percent of new vehicles sold in the United States, Europe, Japan and China had ADAS. The firm predicted that 50 percent of all automobiles on the road would be ADAS-enabled by 2030. In Europe, 3 percent of passenger cars sold in the second quarter of 2018 had level 2 driving features. A year later, in the second quarter of 2019, that rose to 325,000 cars, or 8 percent of all new cars sold.
On the 29th of June 2021, the National Highway Traffic Safety Administration issued Standing General Order 2021-01. It required manufacturers of Level 1 and 2 ADAS and Levels 3-5 automated driving systems to report crashes that happened while the technology was in use. The order was amended on the 5th of August 2021 and ran for three years from its start. A crash is reportable if it happens on a publicly accessible US road and the system was engaged at any time in the 30 seconds before the crash began through its end.
Severe crashes carry a tight clock. A crash is severe if someone goes to a hospital or dies, if a vehicle is towed or an air bag deploys, or if a vulnerable road user is involved. Pedestrians, wheelchair users, motorcyclists and bicyclists count as vulnerable road users. Severe crashes must be reported within one calendar day, with an update within ten calendar days. Non-severe crashes involving automated driving systems are due on the fifteenth day of the month after the manufacturer learns of them.
The agency gathered data from the 1st of July 2021 through the 15th of May 2022 and released the first set in June 2022. Automated driving systems from 25 manufacturers were involved in 130 crashes. Waymo reported 62, Transdev Alternative Services 34, Cruise 23, General Motors 16 and Argo AI 10. The total of those numbers is higher than 130 because several manufacturers can report the same crash. In 108 of the 130 crashes, no injuries were reported, and only one serious injury was associated with the rest. The rear of the vehicle was the most commonly reported damage location.
Level 2 ADAS from 12 manufacturers were involved in 367 crashes over the same period. In all, 392 crashes were reported, but 25 either happened before July 2021 or had no date. Tesla led with 273, followed by Honda with 90 and Subaru with 10. Of the 392, 98 included injury reporting. Among those 98, 46 had no injuries reported, 5 resulted in serious injuries and 6 resulted in fatalities. For these cars the front of the vehicle was the most commonly reported damage location.
Manufacturers are not required to report how many equipped vehicles they have built, how many are operating, or how far they have traveled with the system active. Without those totals, a high count could simply reflect more cars on the road rather than more danger. The tally says little about whether drivers even understand the systems they are using.
Adaptive cruise control is called Adaptive Cruise Control by Fiat, Ford, GM, VW, Volvo and Peugeot. Nissan calls it Intelligent Cruise Control, Citroen and BMW call it Active Cruise Control, and Mercedes calls it DISTRONIC. One feature carries four different names across those makers.
AAA, Consumer Reports, J.D. Power, the National Safety Council, PAVE and SAE International all promote standard names. They favor terms such as "forward collision warning" and "automatic emergency braking" over "forward collision alert" or "smart city brake support". SAE International endorsed a series of recommendations for generic ADAS terminology, created with Consumer Reports, the American Automobile Association, J.D. Power and the National Safety Council.
PACTS argues that the lack of standardization makes ADAS hard for drivers to understand. Button names and locations and dashboard symbols change from car to car. Behavior changes too. In most cars the adaptive cruise speed can be temporarily overridden, while some cars switch to standby after one minute.
One pre-collision system might use 12 pages to explain 23 exceptions where it may operate when not needed. The same pages list 30 exceptions where it may not operate even though a collision is likely. These limits show why the lack of shared names and behavior matters.
In October 2023, Consumer Reports rated 17 active driving assistance systems. Its criteria were capabilities and performance, clarity about when the system is safe to use, ease of use, keeping the driver engaged and handling an unresponsive driver. Ford and Lincoln's BlueCruise topped the list with 84. General Motors' Super Cruise followed with 75 and Mercedes-Benz's Driver Assistance scored 72. Tesla's Autopilot scored 61, and the lowest score, 47, went to Hyundai, Kia and Genesis Highway Driving Assist.
The Insurance Institute for Highway Safety issued its first partial automation safeguard ratings in March 2024. Its criteria included monitoring both the driver's gaze and hand position, and using multiple types of rapidly escalating alerts. Automated lane changes must be started or confirmed by the driver, and features must not work with the seat belt unfastened. No system earned a rating of good. Lexus Teammate with Advanced Drive was rated acceptable. Super Cruise and Nissan's ProPILOT Assist with Navi-link were marginal. BlueCruise, Tesla's Autopilot and Full Self-Driving (Beta) were among the many rated poor. The same BlueCruise that topped one ranking landed in the bottom group of the other. Neither scorecard says what a car should do when a crash cannot be avoided.
In 1967, the English philosopher Philippa Foot introduced the trolley problem. A trolley's brake has failed, and five people stand ahead. The driver can go straight and kill the five, or turn onto a side track and kill one pedestrian. Before autonomous vehicles, this was a dilemma between utilitarianism and deontological ethics. Now it is a question for the programming of self-driving cars. Because crashes are unavoidable, it raises the question "whose lives should be prioritized in the event of an inevitable crash?"
One approach lets artificial intelligence learn human ethics from data about human actions. This helps when rules cannot be written out in full. It has limits. Many human actions come from self-preservation instincts, which are realistic but not ethical. The data must also be carefully selected to avoid undesired outcomes.
Noah J. Goodall proposed a three-phase approach. First, car manufacturers, transportation engineers, lawyers and ethicists agree on a system, and it is set transparently. Second, artificial intelligence learns human ethics while bound by that system. Third, the system gives constant feedback that humans can understand.
Market Research Future expects the autonomous vehicle market to hit over $65 billion by 2027. Manufacturers are advised to carry four kinds of insurance. Technology errors and omissions covers the cost of a crash when the technology fails. Auto liability and physical damage covers third-party injuries and technology damage. Cyber liability covers lawsuits and regulator penalties tied to cybersecurity. Directors and officers coverage protects a company's balance sheet and assets from bad management. Cars that record every minute on the road may also reduce fraud, including staged crashes. In September 2016, the National Highway Traffic Safety Administration published the Federal Automated Vehicles Policy for highly automated vehicles.
Intelligent transportation systems resemble ADAS but cover any enterprise that safely moves people, including public transport. By tying transportation technology to a city's infrastructure, they point toward a smart city. They could add 22.5 percent capacity to roads on average.
A draft regulation named driver control assistance systems, or DCAS, would allow systems such as Tesla's Full Self-Driving in Europe. The UNECE plan says DCAS must ensure the driver performs the driving task, with hands on the wheel. The system would also monitor the driver's visual engagement.
Updated
Common questions
What are advanced driver-assistance systems (ADAS)?
Advanced driver-assistance systems, known as ADAS, are the sensors, cameras and software that help a driver drive or park. They detect nearby obstacles or driver errors and respond with alerts or actions.
How do advanced driver-assistance systems work?
ADAS draw on outside data such as automotive imaging, LiDAR, radar, computer vision, telemetry and in-car networking. They are real-time systems that rank incoming information with preemptive priority scheduling to prevent crashes.
When was cruise control first patented in the history of advanced driver-assistance systems?
Ralph Teetor applied for a US patent in 1948 on a speed control device for resisting operation of the accelerator. The patent was granted in 1950, and the device is now known as cruise control.
How much can advanced driver-assistance systems reduce crashes?
According to the US national crash database, forward collision prevention could reduce crashes by 29 percent. Lane keeping assistance could reduce them by 19 percent and blind zone detection by 9 percent.
What is Standing General Order 2021-01 for advanced driver-assistance systems?
The National Highway Traffic Safety Administration issued Standing General Order 2021-01 on the 29th of June 2021. It requires manufacturers to report crashes that occur while Level 2 ADAS or Levels 3-5 automated driving systems are in use.
Why are advanced driver-assistance systems hard for drivers to understand?
ADAS feature names, button locations and dashboard symbols vary from car to car. Adaptive cruise control, for example, is called DISTRONIC by Mercedes and Intelligent Cruise Control by Nissan.
How did the Insurance Institute for Highway Safety rate advanced driver-assistance systems in 2024?
In March 2024 the Insurance Institute for Highway Safety issued its first partial automation safeguard ratings, and no system earned a rating of good. Lexus Teammate with Advanced Drive was rated acceptable, and most other systems were rated poor.
All sources
91 references cited across the entry
- 1NewsTesla Autopilot Crash Avoidance Model S Autopilot saves man's lifeZach Epstein — 2016-07-21
- 2JournalBehavioural impacts of Advanced Driver Assistance Systems–An OverviewKarel A. Brookhuis et al. — 2001-06-01
- 3Book2017 IEEE Conference on Systems, Process and Control (ICSPC)Umar Zakir Abdul Hamid et al. — 2017-12-01
- 4JournalImpact analysis of Advanced Driver Assistance Systems (ADAS) regarding road safety – computing reduction potentialsMichael Aleksa et al. — 2024-06-27
- 5NewsAutomotive electronics revolutionRaj Kumar Nagpal et al. — AspenCore — 2022-05-18
- 10JournalHistory and future of driver assistanceMarco Galvani — 2019-02-04
- 11Understanding ADAS Calibration: What You Need To KnowSteve — 2025-04-16
- 12JournalAn Overview of Vehicular CommunicationsFabio Arena et al. — 2019-01-24
- 13Book2011 seventh International Computer Engineering Conference (ICENCO'2011)Adnan Shaout et al. — 28 December 2011
- 14Aftermarket ADAS can help fleets avoid collisionsJuly 23, 2019
- 17Which Cars Have Self-Driving Features for 2020?Kelsey Mays — 2020-04-04
- 18Guide to Cars With Advanced Safety Systems2020-05-08
- 20JournalDetecting Driver Drowsiness Based on Sensors: A ReviewArun Sahayadhas et al. — 7 December 2012
- 21JournalReal-time Driver Drowsiness Detection for Android Application Using Deep Neural Networks TechniquesRateb Jabbar et al. — 2018
- 26Intelligent Speed Adaptation (ISA)Anonymous — 2016-10-17
- 30JournalTyre pressure monitoring using a dynamical model-based estimatorGiulio Reina et al. — 2015-04-03
- 32JournalWelcome to ROSA PSarah Simpson et al. — November 2015
- 33JournalEvaluation of a New Pedestrian Head Injury Protection System with a Sensor in the Bumper and Lifting of the Bonnet's Rear PartRikard Fredriksson et al. — 2001-06-04
- 34JournalThe influence of Cruise Control and Adaptive Cruise Control on driving behaviour – A driving simulator studyMarkVollrath et al. — 19 December 2010
- 37Handbook of Intelligent VehiclesAzim Eskandarian — Springer — 2012
- 38How Pre-Collision Systems Work2009-04-22
- 39JournalAutomated simulation of scenarios to guide the development of a crosswind stabilization functionKlaus-Dieter Hilf et al. — 2010-07-01
- 40How Cruise Control Systems Work2001-01-15
- 41How Electronic Stability Control Works2009-10-05
- 43How Do Hill Descent Control Systems Work?Jeremy Laukkonen — 2021-10-25
- 44How Hill-Start Control Works2009-10-05
- 46Tech'splaining: What is Lane Centering Assist, How Does It Work?Christian Wardlaw — 2019-07-08
- 47Book2011 IEEE Intelligent Vehicles Symposium (IV)Stefan Habenicht et al. — 5 July 2011
- 48Lane Change Assistance – Level Five Supplies11 January 2019
- 49UN Regulation No. 79, Revision 4. Uniform provisions concerning the approval of vehicles with regard to steering equipmentUnited Nations Economic Commission for Europe — 2018-11-07
- 51Traction Control Explained2005-09-07
- 52JournalMicromirror based virtual image automotive head-up displayChao Fan et al. — 2017-06-01
- 53JournalAutomotive night vision systemNancy S. Martinelli et al. — International Society for Optics and Photonics — 1999-03-19
- 54How Rearview Cameras Work2008-06-23
- 56JournalGlare-free high beam optimization based on country road traffic simulationJ Kobbert et al. — 2018-11-14
- 59Hands-free driving is becoming more popular — but is it safe?Andrew J. Hawkins — 31 July 2023
- 60Hit >1 billion miles driven on FSDTesla AI — 4 April 2024
- 61On June 9, 2023
- 63Press releaseHonda Unveils Next-generation Technologies of Honda SENSING 360 and Honda SENSING EliteHonda — 2022-12-01
- 64Mercedes Launches SAE Level 3 Drive Pilot SystemJay Ramey — 2022-05-09
- 65Standing General Order on Crash Reporting For incidents involving ADS and Level 2 ADASUnited States Department of Transportation, National Highway Traffic Safety Administration
- 66ReportIn re: Standing General Order 2021-01 Incident Reporting for Automated Driving Systems (ADS) and Level 2 Advanced Driver Assistance SystemsUnited States Department of Transportation, National Highway Traffic Safety Administration — June 29, 2021
- 67ReportIn re: First Amended Standing General Order 2021-01 Incident Reporting for Automated Driving Systems (ADS) and Level 2 Advanced Driver Assistance SystemsUnited States Department of Transportation, National Highway Traffic Safety Administration — August 5, 2021
- 68Press releaseNHTSA Releases Initial Data on Safety Performance of Advanced Vehicle TechnologiesUnited States Department of Transportation, National Highway Traffic Safety Administration — June 15, 2022
- 69ReportSummary Report: Standing General Order on Crash Reporting for Automated Driving SystemsUnited States Department of Transportation, National Highway Traffic Safety Administration — June 2022
- 70ReportSummary Report: Standing General Order on Crash Reporting for Level 2 Advanced Driver Assistance SystemsUnited States Department of Transportation, National Highway Traffic Safety Administration — June 2022
- 71What does my Car do?Shaun Helmen et al. — Parliamentary Advisory Council for Transport Safety — 2019-09-19
- 72NewsAdvanced driver assistance systems: how the latest car safety tech is testedHugo Griffiths — 2021-05-07
- 73SAE International endorses generic ADAS terminology recommendationsJohn Huetter — 2020-05-15
- 74New Safety Features in Cars (or Just New to You)Tom Voelk — 2020-10-08
- 79JournalThe Trolley ProblemJudith Jarvis Thomson — 1985
- 80JournalThe Ethics of Accident-Algorithms for Self-Driving Cars: an Applied Trolley Problem?Sven Nyholm et al. — 2016-11-01
- 81BookArtificial intelligence : a modern approachStuart Jonathan Russell — Pearson — July 2019
- 82JournalEthical Decision Making during Automated Vehicle CrashesNoah J. Goodall — 1 January 2014
- 83JournalTRENDS AND ISSUES IN SAFE DRIVER ASSISTANCE SYSTEMS: Driver Acceptance and Assistance for Elderly DriversSadayuki Tsugawa — 2006-01-01
- 85A top auto safety group tested 14 partial automated systems — only one passedAndrew J. Hawkins — 2024-03-12
- 86Partial automation safeguard ratingsMarch 2024
- 87BookSmart and Green Solutions for Transport SystemsWojciech Lewicki et al. — Springer International Publishing — 2019-11-14