Cars
Last reviewed
May 13, 2026
Sources
70 citations
Review status
Source-backed
Revision
v2 · 6,349 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Last reviewed
May 13, 2026
Sources
70 citations
Review status
Source-backed
Revision
v2 · 6,349 words
Add missing citations, update stale details, or suggest a clearer explanation.
See also: Cars ChatGPT Plugins
AI in cars covers the use of machine learning, computer vision, neural networks, and large language models in passenger vehicles, commercial trucks, and the systems that build and insure them. The catch-all label hides a very mixed reality. Some pieces, like lane keep, automatic emergency braking, and traffic sign recognition, are now standard on cheap cars. Others, like Level 4 robotaxis, work only inside small geofenced areas after a decade of work and tens of billions of dollars in capital. And the newest layer, conversational generative AI inside the cabin, is mostly a 2023 to 2026 phenomenon that nobody has yet figured out how to make money on.
This article covers driver assistance, autonomous driving, autonomous trucking, in-car voice assistants powered by LLMs, manufacturing AI, telematics, and the regulatory record around crashes and recalls, including the 2023 Cruise incident in San Francisco and the National Highway Traffic Safety Administration probes into Tesla Autopilot and Full Self-Driving.
AI shows up in cars at several different layers. The most visible layer is ADAS, short for advanced driver assistance systems. ADAS uses cameras, radar, and sometimes lidar feeding into perception and planning models to keep a car in its lane, follow the vehicle in front, and brake before a forward collision. ADAS is a feature on a human-driven car, not a replacement for the human. Almost every new car sold in the United States, Europe, Japan, and China in the mid 2020s ships some flavor of ADAS, and automatic emergency braking is mandatory in the United States by 2029.
A separate stack of technology is trying to replace the driver entirely. The work that started in academia and at DARPA in the 2000s turned into a billion-dollar industry, and as of 2026 Waymo operates a paid robotaxi service across multiple US cities and is providing several hundred thousand rides a week. Tesla launched a supervised robotaxi service in Austin in 2025. Pony.ai, Baidu Apollo Go, and WeRide run commercial robotaxi services in China. Heavy trucking has had a parallel arc, with Aurora Innovation becoming the first company to run fully driverless heavy trucks in commercial revenue service on US public roads, in Texas in April 2025.
Generative AI has taken longer to find a home in cars. Mercedes-Benz was first to run a public beta of ChatGPT inside its MBUX system in June 2023. Volkswagen and BMW announced their LLM integrations at CES 2024. By 2026 GM is rolling Google Gemini to roughly four million OnStar-connected vehicles. None of these systems drive the car. They take voice input, answer general questions, and increasingly control infotainment and climate functions.
The other thread that runs through the whole field is regulation. NHTSA has run several open investigations into Tesla Autopilot and Tesla FSD that resulted in a December 2023 recall affecting over two million vehicles. The California DMV pulled Cruise's driverless permit in October 2023 after the company misrepresented what had happened in a pedestrian crash. GM shut Cruise's robotaxi business down in December 2024. Apple cancelled its decade-long car project in February 2024. So the picture is not a smooth march of progress; some companies and bets won, others lost badly.
Research into self-driving cars goes back to the 1980s, with the Carnegie Mellon Navlab project under Red Whittaker and the Eureka PROMETHEUS Project in Europe. The modern era starts with the DARPA Grand Challenge in 2004, in which the United States Defense Advanced Research Projects Agency offered a one million dollar prize for an autonomous vehicle that could complete a 142 mile desert course from Barstow, California to Primm, Nevada. Fifteen teams competed and none finished. The lead vehicle managed just over seven miles.
The 2005 rematch went better. Stanley, a modified Volkswagen Touareg fielded by Stanford under Sebastian Thrun, finished the 132 mile course in six hours and 54 minutes. Carnegie Mellon's Sandstorm and H1ghlander came in second and third. The 2007 Urban Challenge introduced city traffic and an MIT team finished fourth behind Carnegie Mellon's Tartan Racing. The veterans of those teams ended up running the autonomous vehicle programs at Google, Uber, Aurora, Argo AI, and Zoox, so the DARPA challenges had outsized influence on the next twenty years of the field.
Google started its in-house self driving car project in January 2009 under Thrun, with engineering led by Chris Urmson and Anthony Levandowski. The project ran inside Google for seven years, racked up over two million miles of testing on public roads, and was renamed Waymo when it was spun out as an independent Alphabet company in December 2016. Tesla launched Autopilot on the Model S in October 2015, using hardware and software from Mobileye; the partnership ended after the Joshua Brown fatality in May 2016.
GM acquired Cruise Automation in March 2016 for around $581 million. Aurora Innovation was founded by Urmson, ex-Tesla Autopilot lead Sterling Anderson, and ex-Uber autonomy chief Drew Bagnell in 2017. Argo AI, backed by Ford and Volkswagen, also launched in 2017 and shut down in October 2022. Uber sold its Advanced Technologies Group to Aurora in late 2020 after a 2018 fatal crash in Tempe, Arizona that killed Elaine Herzberg, the first known pedestrian death involving a self-driving test vehicle.
Waymo opened Waymo One to the public in Phoenix in December 2018 with safety drivers. Fully driverless rides for the public came in late 2020. Cruise launched paid driverless robotaxi service in San Francisco in June 2022. By 2024 Waymo was running in Phoenix, San Francisco, Los Angeles, and Austin. In April 2025 Aurora started driverless commercial trucking on the Dallas to Houston corridor. Tesla launched a supervised robotaxi pilot in Austin in June 2025. Zoox opened a Las Vegas service in September 2025. By the end of 2025 Waymo alone was providing roughly 450,000 paid rides a week.
ADAS combines cameras, radar, ultrasonic sensors, and on some vehicles lidar, feeding into perception and planning algorithms that warn or intervene to avoid a crash. Most modern systems are SAE Level 2 at best, meaning the human driver remains responsible for the driving task and must keep watching the road. Several flagship suites such as Drive Pilot on the Mercedes EQS and S-Class qualify as Level 3 within a tight operational design domain.
| System | Vendor or OEM | First shipped | Notes |
|---|---|---|---|
| EyeSight | Subaru | 2008 in Japan, 2013 in the US | Stereo camera, no radar; standard on most Subaru models |
| Mobileye EyeQ | Mobileye / Intel | EyeQ1 in 2004 | Vision system on chip; over 200 million chips shipped through 2024 |
| Bosch ADAS Integration Platform | Bosch | Multi-generation | Market leader for radar; Level 0 to 3 sensor fusion |
| Continental ARS radar | Continental | 1990s onward | Front radar widely used for adaptive cruise |
| Autopilot | Tesla | October 2015 | Vision based after 2021 radar removal |
| Super Cruise | GM | 2017 on Cadillac CT6 | Hands-free on mapped highways |
| BlueCruise | Ford | 2021 | Hands-free on Mustang Mach-E and F-150 |
| ProPilot Assist | Nissan | 2016 | Lane centering and traffic-aware cruise |
| Honda Sensing | Honda | 2014 | Standard across the US lineup by 2022 |
| Toyota Safety Sense | Toyota | 2015 | Pedestrian detection, automatic emergency braking |
| Drive Pilot | Mercedes-Benz | 2022 in Germany, 2023 in US | SAE Level 3, freeway only, 40 mph cap initially |
| Highway Pilot | Honda | 2021 in Japan on Legend | Level 3 system limited to 50 km/h |
| NIO Pilot, ZEEKR, Xpeng XNGP | Chinese OEMs | 2020 onward | City NGP features rolling out in Chinese cities |
| Pilot Assist | Volvo | Mid 2010s | Camera and radar based |
In April 2024 NHTSA finalized a rule requiring automatic emergency braking and pedestrian AEB on all new passenger cars and light trucks sold in the United States by September 2029. Similar mandates have been in place in the European Union since 2022 under General Safety Regulation 2.
| Company | Founded | Status as of 2026 | Notable milestones |
|---|---|---|---|
| Waymo | 2009 (Google project), 2016 spin-off | Active; ~450K rides per week end of 2025 | First public commercial robotaxi service, December 2018; fully driverless to public, late 2020; Waymo Driver 6 unveiled August 2024 |
| Cruise | 2013 | Robotaxi unit folded into GM December 2024 | GM acquired 2016; SF paid service June 2022; California DMV suspended permit October 24, 2023; $1.5M NHTSA fine for false report |
| Tesla | 2003 (Autopilot from 2014) | Active; FSD v14 rolling out; supervised robotaxi pilot in Austin from June 2025 | Autopilot launched October 2015; FSD Beta opened October 2020; FSD v12 end-to-end neural network 2024; Cybercab revealed October 10, 2024 |
| Aurora Innovation | 2017 | Active; public via SPAC November 2021 | Launched first commercial driverless trucking, Dallas to Houston, April 2025 |
| Mobileye | 1999 | Active; IPO October 2022 | Intel acquired 2017 for $15.3B; Mobileye Drive Level 4 platform; SuperVision on Zeekr 001 |
| Wayve | 2017 | Active | UK end-to-end driving startup; $1.05B Series C led by SoftBank, NVIDIA, Microsoft, May 2024; further $1.2B round 2026 |
| Zoox | 2014 | Active; Amazon subsidiary | Acquired by Amazon for ~$1.3B in 2020; public Las Vegas launch September 2025 |
| Pony.ai | 2016 | Active; Nasdaq IPO November 2024 | Toyota invested $400M in 2020; 250+ robotaxis in 2024; expanding to Europe |
| WeRide | 2017 | Active; Nasdaq IPO October 2024 | Robobus, robotaxi, and robosweeper services in China |
| Baidu Apollo Go | 2017 (Apollo platform) | Active | Over 17M cumulative rides through November 2025; 1,000-vehicle fleet milestone May 2025 |
| Argo AI | 2017 | Shut down October 2022 | Ford and VW backed; ran tests in Miami and Pittsburgh |
| Uber ATG | 2015 | Sold to Aurora December 2020 | Elaine Herzberg fatality March 18, 2018 |
| TuSimple | September 2015 | Exited US, rebranded as CreateAI | Nasdaq IPO April 2021; US wind-down December 2023; delisted January 2024 |
| Embark Trucks | 2016 | Wound down March 2023 | SPAC merger November 2021 at $5.2B valuation |
| Kodiak Robotics | 2018 | Active | First customer-owned driverless commercial truck delivery for Atlas Energy Solutions, December 2024 |
| Plus | 2016 | Active | Level 4 SuperDrive on Scania and International trucks |
| Apple Project Titan | 2014 | Cancelled February 27, 2024 | Estimated several billion dollars spent |
| Aptiv / Motional | 2020 JV | Active | Hyundai and Aptiv joint venture; Las Vegas Lyft pilot |
| Nuro | 2016 | Pivoted from delivery pods to driver software | Texas pilots; raised $106M Series E February 2025 |
| DiDi Autonomous Driving | 2019 spin-off | Active | China operations |
| Bot Auto | 2024 | Early stage | Founded by ex-TuSimple CEO Xiaodi Hou |
SAE International publishes the J3016 taxonomy, the de facto industry definition for driving automation. The standard distinguishes six levels.
| Level | Name | Who drives | Typical example |
|---|---|---|---|
| 0 | No driving automation | Human, always | Conventional car with maybe blind spot warning |
| 1 | Driver assistance | Human, with system help in one axis | Adaptive cruise control only, or lane keep only |
| 2 | Partial driving automation | Human, system helps with steering and speed | Tesla Autopilot, GM Super Cruise, Ford BlueCruise, Honda Sensing |
| 3 | Conditional driving automation | System within ODD; human takes back when asked | Mercedes Drive Pilot, Honda Sensing Elite on Legend |
| 4 | High driving automation | System inside ODD; no human fallback | Waymo One in Phoenix and SF; Cruise in SF before suspension; Aurora driverless trucks |
| 5 | Full driving automation | System everywhere, in all conditions | No production example exists in 2026 |
ODD is short for operational design domain, the specific conditions, geography, weather, and speeds under which a system is rated to operate. A Level 4 system that works flawlessly inside the Phoenix metro at 45 mph is still Level 4. The level alone says very little about the car's actual capability; it only says how much responsibility the system takes inside its operating domain.
The key confusion in marketing is that Tesla's Full Self-Driving (Supervised) and most other consumer ADAS suites are Level 2, even when the system handles most of the driving most of the time. Whenever the human is expected to take over without prior warning, the system is Level 2 by definition.
Tesla offered Autopilot pre-orders in October 2014, then shipped the first working version, called Autopilot 1, in October 2015 with software update 7.0. The hardware came from Mobileye and the system handled adaptive cruise, lane keep, and automatic lane changes on highways. Joshua Brown, a Tesla Model S owner, died in a May 7, 2016 crash in Williston, Florida when his car drove under a tractor trailer making a left turn across the highway. The Mobileye partnership ended in July 2016. Tesla then moved to a second hardware platform built around an Nvidia GPU, and in March 2019 launched HW3 with a custom Tesla designed inference chip.
Full Self-Driving Beta opened to a small group of owners in October 2020. The early versions were essentially Autopilot extended into city streets, with explicit code for stop signs, traffic lights, and unprotected turns. FSD Beta v9 in July 2021 removed radar and went vision only. FSD Beta v11 in 2023 unified highway and city stacks. FSD v12, which started rolling out in early 2024, replaced large portions of the hand coded driving stack with an end-to-end neural network trained from video. FSD v13 followed for HW4 vehicles in late 2024, and v14 began rolling out in October 2025.
Hardware 4, branded as AI4, launched on production vehicles in early 2023. It uses an updated custom Tesla chip and a new camera array. HW3 vehicles continue to receive FSD updates but lag behind HW4 by one major version. Some HW3 owners, mainly buyers of the original $10,000 to $15,000 FSD package, have argued Tesla owes them the same capability promised at purchase. The dispute is ongoing.
Tesla revealed Dojo at AI Day on August 19, 2021. The custom D1 chip, fabricated on TSMC's 7nm process, packed 50 billion transistors and 362 TFLOPS into a 645 mm2 die. Twenty five D1 chips made a training tile, and 120 tiles formed a full Dojo cabinet. The first ExaPOD reached around an exaflop of compute. Bloomberg reported on August 7, 2025 that Tesla had disbanded the Dojo team. Pete Bannon, the lead on the project, left the company. Musk publicly called Dojo 2 an evolutionary dead end and Tesla shifted to a heavier reliance on Nvidia GPUs at its Cortex training facility in Texas plus AI6 chips from Samsung under a $16.5 billion deal.
Tesla unveiled the Cybercab and Robovan at the "We, Robot" event at Warner Bros. Studios in Burbank, California on October 10, 2024. The Cybercab is a two seat purpose-built robotaxi without a steering wheel or pedals, with a stated target price below $30,000 and inductive charging. Twenty prototypes drove guests around the studio backlot. Tesla launched a small supervised robotaxi pilot in Austin in June 2025 using Model Y vehicles with a Tesla employee in the front passenger seat. The Cybercab itself is not yet in production.
Waymo grew out of the Google Self-Driving Car Project that started in 2009 inside Google X. Sebastian Thrun, who had led the Stanford team that won the 2005 DARPA Grand Challenge, started the program with Chris Urmson and Anthony Levandowski. Levandowski left in 2016 to found Otto, was sued by Waymo over alleged trade secret theft, and was pardoned by Donald Trump in January 2021 after pleading guilty. Urmson left in 2016 to start Aurora. John Krafcik became CEO of Waymo in 2015 and oversaw the December 2016 spin-out from Alphabet.
Waymo One launched in Phoenix in December 2018, the first public commercial autonomous ride hailing service in the world. Initial service used safety drivers and operated in a small geofenced area covering Chandler, Tempe, Mesa, and Gilbert. Fully driverless rides for the public came in late 2020. Tekedra Mawakana and Dmitri Dolgov became co-CEOs in 2021 after Krafcik departed. Service expanded to San Francisco in 2024 and to Los Angeles and Austin during 2024. Atlanta service launched via Uber in 2025.
In August 2024 Waymo announced the sixth generation of the Waymo Driver, riding on a purpose built Zeekr van called the Ojai. The new generation uses 13 cameras, 4 lidars, and 6 radars, down from 29 cameras in the prior generation, and Waymo says compute and capability went up while sensor costs dropped substantially. A second integration on the Hyundai Ioniq 5 is also planned. Service area in the Bay Area grew almost fivefold during 2025, from around 55 to over 260 square miles. November 2025 added paid freeway rides in San Francisco, Los Angeles, and Phoenix.
By the end of 2025 Waymo was providing more than 14 million paid trips for the year, averaging around 450,000 a week, up from about 175,000 a week at the start of the year. The company announced plans to launch in roughly a dozen new US cities through 2026 and to operate in London as its first overseas market.
Cruise was founded in 2013 by Kyle Vogt and Daniel Kan. General Motors bought it in March 2016 for roughly $581 million plus stock. The company began testing in San Francisco and obtained a permit from the California Public Utilities Commission in June 2022 to operate a paid driverless robotaxi service at night in part of the city. Service expanded in 2023 to broader hours and additional cities including Phoenix and Austin.
On October 2, 2023, around 9:30 pm, a pedestrian crossing Market Street near Fifth Street was struck by a human driven Nissan and thrown into the path of a Cruise Bolt operating without a driver. The Cruise vehicle ran her over, then attempted a pullover maneuver and dragged her approximately 20 feet before coming to a stop. Cruise initially showed regulators a video that cut off at the first stop. The California DMV suspended Cruise's driverless deployment and testing permits on October 24, 2023, citing public safety concerns and a misrepresentation of the incident.
Cruise paused all driverless operations across the country immediately afterward. CEO Kyle Vogt resigned on November 19, 2023. The company laid off about a quarter of its workforce in December 2023. Cruise paid a $1.5 million NHTSA penalty in 2024 and another $500,000 fine to the US Attorney's Office for the Northern District of California in November 2024 after admitting it had filed a false report to influence a federal investigation. A separate $8 million settlement with the injured pedestrian was reported in May 2024.
On December 10, 2024, GM announced it would no longer fund Cruise's robotaxi development and would fold the unit into its core engineering organization. CEO Mary Barra cited the cost and competitive dynamics of the robotaxi market. GM had invested an estimated $10 billion in Cruise over the course of its life. Remaining engineering work was redirected toward personal autonomous vehicles.
Long haul trucking is in some ways an easier autonomy problem than urban driving. Trips run on highways with predictable geometry, limited pedestrian and cyclist exposure, and consistent customer demand. Driver shortages and freight cost pressure made it commercially attractive. The economics also favor autonomy because driver labor is roughly a third of operating cost on a long haul move.
Aurora Innovation became the first company to operate fully driverless commercial trucking on public US highways. Aurora launched paid driverless freight runs on the Dallas to Houston corridor with two trucks in April 2025, scaling to a small fleet through 2025 and expanding to El Paso and Phoenix routes. Aurora went public via a SPAC in November 2021 valued at $13 billion. Its Aurora Driver platform integrates with Paccar's Peterbilt 579 and Volvo VNL truck platforms.
Kodiak Robotics was founded in 2018 by Don Burnette and Paz Eshel. Kodiak delivered the first customer owned and operated autonomous trucks to Atlas Energy Solutions in December 2024, running short haul frac sand routes in West Texas oil fields. By mid 2025 Atlas owned four Kodiak Driver equipped trucks. Kodiak also has US Department of Defense contracts for autonomous off-road military trucks.
TuSimple was founded in September 2015 by Mo Chen and Xiaodi Hou and went public on Nasdaq in April 2021 at a valuation around $8.5 billion. The company ran the first driverless truck test on US public highways in December 2021 in Arizona. Internal disputes, allegations of technology transfer to a Chinese partner, and a CFIUS national security review forced an unwind of US operations beginning in December 2023. TuSimple delisted from Nasdaq in January 2024 and rebranded as CreateAI in late 2024, pivoting to AI animation and gaming.
Embark Trucks, founded in 2016 by Alex Rodrigues and Brandon Moak, merged with a SPAC at a $5.2 billion valuation in November 2021. The company laid off 70 percent of staff in late 2022 and shut down in March 2023, selling its remaining IP to Applied Intuition. Plus is still operating, partnering with TRATON Group brands and Hyundai for factory-built Level 4 trucks. Xiaodi Hou, the ousted CTO of TuSimple, started a new Houston based autonomous trucking company called Bot Auto in 2024.
Deepway, Inceptio, and other Chinese autonomous trucking startups operate in China but have a much smaller US footprint. Volvo Autonomous Solutions, the autonomous truck unit of Volvo Group, runs driverless yard trucks at customer facilities and has partnerships with Aurora and DHL.
LLM powered voice assistants started entering production cars in 2023. The basic pattern is that the existing in-car voice assistant routes complex requests to a cloud LLM and reads back the answer, while local intent recognition handles vehicle control like climate, navigation, and seat heaters. The marketing is more aggressive than the underlying capability so far; most systems still fall back to traditional command grammars for anything safety relevant.
| OEM or product | Model partner | Launched | Notes |
|---|---|---|---|
| Mercedes-Benz MBUX with ChatGPT | OpenAI via Azure | June 16, 2023 (US beta) | First major automaker to ship ChatGPT in a car; opt-in via Mercedes me app; about 900,000 vehicles eligible |
| Volkswagen IDA with Cerence Chat Pro | OpenAI via Cerence | CES 2024, deployed Q2 2024 | ID.7, ID.4, ID.5, ID.3, Tiguan, Passat, Golf; rolled out in five languages across VW Group brands including Cupra, Seat, Skoda |
| BMW Intelligent Personal Assistant | Amazon Alexa LLM | CES 2024 demo, in production with OS 9 | Conversational car expert and drive mode control |
| GM with Google Gemini | Announced October 2025, OTA April 2026 | ~4 million model year 2022+ Cadillac, Chevrolet, Buick, GMC; OnStar integration | |
| Hyundai, Kia, Genesis Pleos Connect with Gleo AI | In-house plus NVIDIA Nemotron | 2025 rollout on Korean Grandeur first | Target of 20 million Hyundai Group vehicles by 2030 |
| Honda 0 Series with Renesas-based assistant | TBD | Announced CES 2025 | Honda's in-house ASIMO OS |
| Stellantis with Mistral and SoundHound | Mistral, SoundHound Chat AI | 2024 fleet pilots | Italian and French markets first |
| Ford with Anthropic Claude | Anthropic | Pilot announced 2024 | Engineering and dealer assistant first, in-car later |
| Polestar with Google Gemini | Announced 2024 | Polestar 4 launch | |
| NIO Nomi GPT | In-house plus partners | 2024 | Replaced first-generation Nomi |
The deeper question is what the LLM is actually doing in the car. For most current implementations the answer is: routing voice queries through a remote endpoint and reading the answer aloud. The model does not steer. It does not interpret sensor data. It may, with newer implementations from GM and Hyundai, control infotainment, climate, and route planning by function calling. Tesla's voice control remains relatively spartan compared with the LLM-enabled systems, though Grok integration was added to vehicles starting in mid 2025.
Automakers have used machine learning in factories since well before the current generative AI wave. The applications include defect detection from cameras on the line, predictive maintenance for press lines and welders, demand forecasting, and increasingly robotic vision for assembly. The economics are straightforward; a few percent improvement in first time quality or yield on an EV battery pack line is worth tens of millions of dollars per plant per year.
Tesla applies computer vision throughout its body in white welding and final assembly, with neural networks reused from the Autopilot stack. The company also operates one of the largest non-cloud GPU clusters in the world for training, known as Cortex, located at the Texas Gigafactory and totaling over 230,000 H100 equivalent GPUs by mid 2025. The same training infrastructure supports both FSD and the Optimus humanoid robot, which Tesla is targeting for use inside its own factories before external sale.
Toyota's production engineering work uses ML for press line defect detection and for paint shop anomaly detection. Audi launched an AI-based dent detection system in its Ingolstadt and Neckarsulm press shops in the late 2010s that became a corporate standard. BMW has worked with NVIDIA on AI digital twins of its plants under the Omniverse program. Siemens supplies many automakers with predictive maintenance AI and is the dominant industrial automation vendor in European auto plants. Foxconn, Hon Hai's contract manufacturing arm, is increasingly providing AI-driven contract assembly for EVs including the Lordstown Endurance and several Stellantis models.
Chinese EV makers like BYD, NIO, Xpeng, and Li Auto have been particularly aggressive about lights-out manufacturing using ML-driven QA, in part because they were building greenfield plants from 2017 onward and could design for full automation from the start. The Tesla Shanghai Gigafactory uses similar designs and is widely reported to be Tesla's most efficient assembly plant.
AI in telematics is older than AI in self driving cars, even though it gets less attention. The simplest form is a smartphone app or aftermarket OBD II device that records acceleration, braking, cornering, and phone-handling behavior, scores the driver, and feeds the score into an insurance premium model. Progressive's Snapshot, State Farm's Drive Safe and Save, and Allstate Drivewise are the best known consumer products in the United States.
Cambridge Mobile Telematics, founded in 2010 out of MIT by Hari Balakrishnan and Sam Madden, builds the DriveWell platform that powers many of those carrier programs, including Progressive, USAA, Intact, and Discovery Insure. CMT's app reads phone sensors to detect distracted driving, hard braking, and crashes, and runs claims reconstruction with ML. The company has raised over $850 million and counts SoftBank as its largest investor.
For commercial fleets the market leaders are Lytx with DriveCam and Nauto with its in-cab AI dashcam. Both products run computer vision on a forward facing camera and a driver facing camera to detect distraction, drowsiness, mobile phone use, and tailgating, and to flag the events to a fleet safety manager. The Virginia Tech Transportation Institute ran a benchmark in 2024 that found Nauto delivered alerts faster than competitors on most distraction events. Samsara and Motive are larger fleet management platforms that bundle telematics, AI dashcams, and ELD compliance.
Geotab is the largest commercial telematics vendor by connected vehicles in North America, with over 4 million subscriptions, and has been adding ML-based predictive maintenance and accident reconstruction features through 2024 and 2025. Verizon Connect, formerly Fleetmatics, is a smaller competitor but a frequent OEM partner.
The National Highway Traffic Safety Administration is the lead US regulator for vehicle safety, including automated driving systems on public roads. NHTSA opens investigations through its Office of Defects Investigation. The most consequential cases involving AI driven vehicles are summarized below.
| Date | Action | Vehicle or company | Notes |
|---|---|---|---|
| May 7, 2016 | Fatal Autopilot crash | Tesla | Joshua Brown, Williston, Florida; first known Autopilot fatality |
| March 18, 2018 | Fatal pedestrian crash | Uber ATG | Elaine Herzberg, Tempe, Arizona; first pedestrian death involving a self-driving test vehicle |
| August 13, 2021 | NHTSA opens PE21-020 | Tesla | Preliminary evaluation into Autopilot collisions with stationary emergency vehicles; 765,000 vehicles |
| June 8, 2022 | Upgraded to EA22-002 | Tesla | Engineering analysis; broader scope |
| October 2, 2023 | Pedestrian dragging incident | Cruise | Driverless Bolt in San Francisco dragged pedestrian after a hit-and-run by a human driver |
| October 24, 2023 | DMV permit suspension | Cruise | California suspended driverless permits |
| December 12, 2023 | Recall 23V-838 | Tesla | OTA recall of 2,031,220 vehicles to add Autopilot driver attention controls; closed EA22-002 |
| February 27, 2024 | Project cancellation | Apple | Project Titan shut down after roughly 10 years |
| April 26, 2024 | Final rule, automatic emergency braking | NHTSA | Mandatory AEB on new light vehicles by September 2029 |
| April 25, 2024 | NHTSA opens new probe | Tesla | EA24-005 into adequacy of the December 2023 Autopilot recall |
| October 18, 2024 | NHTSA opens investigation | Tesla FSD | EA24-031, prompted by four FSD-engaged crashes including a fatal pedestrian impact |
| November 14, 2024 | $500,000 federal fine | Cruise | Admitted false report to federal investigators |
| December 10, 2024 | Robotaxi shutdown | Cruise | GM ended Cruise driverless funding |
| April 28, 2025 | Driverless commercial launch | Aurora | First driverless commercial heavy truck on US public road |
| August 7, 2025 | Dojo team disbanded | Tesla | Pete Bannon departure; Musk redirects to Nvidia and Samsung |
European Union regulation runs along a parallel track via UN ECE Regulation 157 for automated lane keeping systems, which permitted Level 3 systems to operate up to 60 km/h initially and was raised to 130 km/h in 2022. Mercedes Drive Pilot was the first system certified under R157 in Germany in December 2021. The EU AI Act, finalized in March 2024 and entering main provisions in 2026, classifies safety-critical components of automated driving systems as high-risk AI and imposes additional documentation and oversight obligations.
Chinese regulation is set at the national level by the Ministry of Industry and Information Technology with provincial overlays. Wuhan, Beijing, Guangzhou, and Shenzhen have all granted full driverless taxi permits. After several high profile crashes in 2025 involving lane assistance features sold as autonomous, the MIIT clamped down on marketing language; "intelligent driving" features can no longer be sold as full self driving in mainland China without specific clearance.
Machine learning has begun to show up inside battery management systems for electric vehicles, where models estimate state of charge, state of health, and remaining useful life from voltage, current, and temperature traces. SES AI, a battery cell developer that went public via SPAC in 2022, sells a service called Avatar that uses LLMs and analytics to accelerate cell formation cycling and defect detection at battery plants. Solidion Technology, Sila Nanotechnologies, and battery startups working on silicon and lithium metal anodes use ML for cycle life forecasting from short experiments. Most automakers have in-house battery analytics teams as well; Tesla's battery state estimation models are reported to run continuously on the inverter compute module of every car in its fleet and feed back into next generation cell designs.