Vehicular automation

• 1. Introduction

  ›

  Vehicular automation replaces or assists the human operator of a car, truck, aircraft, rocket, military vehicle, or boat. In 2015, Google reported that its self-driving cars had failed at least 272 times in testing. Human drivers, the company said, had to intervene around 13 times to prevent fatalities. That single admission captures a tension running through the entire field: a machine can drive itself right up until the moment it can't. How does a vehicle without a driver actually build a picture of the road around it? Which machines, on rails, in the air, and out on open water, learned to operate themselves long before ordinary cars did? And when the technology fails, as it sometimes does, who or what ends up responsible?

• 2. Eyes, Maps and Brains

  ›

  The perception module ingests data from cameras, LIDAR, RADAR, and ultrasonic SONAR, building a picture of everything around the vehicle. It has to track other cars, pedestrians, and cyclists, along with road conditions and obstacles that could affect the trip. Various manufacturers combine cameras, radar, lidar, sonar, and even microphones, since layering sensors together helps cancel out each one's errors.

  A separate localization module cross-references 3D point cloud data, GPS, and an inertial measurement unit, or IMU. Together they fix the vehicle's exact position, orientation, velocity, and turning rate. A planning module then takes those readings, along with perception data, and calculates outputs such as velocity and steering angle. Deep neural networks typically support all of this work, letting the vehicle detect objects, read traffic patterns, and decide in real time. Increasingly, engineers also fuse multiple sensor streams together, to hold accuracy steady across changing weather and light.

  For navigation, the Global Positioning System, or GPS, guides air, water, and land vehicles alike, and does much of the work off-road. On paved roads, two rival approaches compete. Light maps lean on the vehicle's own perception to fill in gaps. Highly detailed maps shrink real-time guesswork instead, but need constant upkeep as streets change. Some systems crowdsource that upkeep, letting an entire fleet report back on new construction or shifting traffic. Roadside monitoring systems add further data, helping route vehicles across a limited road network. Real-time kinematic positioning and precise point positioning, two GNSS enhancements, sharpen a vehicle's location to sub-meter precision. That accuracy matters when a car must decide, in an instant, which lane it actually occupies.

  None of these systems, though, existed in working form when engineers first tried to make a vehicle steer itself. That attempt came decades before GPS or neural networks were even part of the picture.

• 3. The Cart That Crawled

  ›

  Hans Moravec built the Stanford Cart in the late 1970s while he was a graduate student, and it became the first experimental autonomous vehicle. The box-shaped machine rolled on four bicycle wheels and carried a camera, a battery, and a radio antenna linking it to a remote computer. It could also be steered by hand from a distance. Engineers built it partly as a rough stand-in for NASA's Moon and Mars lander projects. Radio signal delay, they reasoned, meant only autonomous control could work that far from Earth. Inside a room 100 feet long, the Cart could steer around large obstacles. Each crossing took roughly 5 hours. Its computer had to keep stopping to process new images before deciding where to move next.

  Roughly 20 years later, Carnegie Mellon University's Robotics Lab built ALVINN, short for Autonomous Land Vehicle In a Neural Network. Three onboard Sun Microsystems computers processed input from a camera and a laser range finder. That let the vehicle slowly track a road's white divider line and steer itself along it. European Union law eventually drew its own line between the two. It defines a hard split between driver-assistance systems and true autonomous vehicles, based on where liability falls when something goes wrong.

  The AAA Foundation for Traffic Safety tested automatic emergency braking systems. The models included the 2016 Volvo XC90, Subaru Legacy, Lincoln MKX, Honda Civic, and Volkswagen Passat. Researchers found that systems built to prevent a crash outright cut vehicle speed by twice as much as systems designed merely to soften the impact. When the two test vehicles closed to within 30 mph of each other, even the gentler systems avoided a collision 60 percent of the time.

  The Sartre project's name stood for Safe Road Trains for the Environment. Its goal was to let a human-driven lead vehicle guide a train of other cars and trucks behind it. Planners expected these trains to deliver comfort and safe passage to any vehicle that joined. A human driver who encountered a train on the road could simply merge in and hand off control.

  None of that testing, though, involved the small, low-speed shuttles already carrying paying passengers on fixed routes around the world.

• 4. Six Seats to Twenty

  ›

  The ParkShuttle has run since 1999 on a dedicated route in Capelle aan den IJssel in the Netherlands. It uses the free-ranging on grid, or FROG, system and a magnetic track. That same route connects the Rivium business park to Rotterdam, terminating at the Kralingse Zoom metro station. In 2005, the system experienced a crash there that investigators traced to human error.

  From 2012 to 2016, the European Union funded CityMobil2, a project that trialed shared autonomous vehicles and the passenger experience across seven cities. That work fed directly into the EasyMile EZ10, a shuttle later tested on a medical campus in Toulouse and in cities from Dubai to Helsinki. By November 2021, EasyMile became the first driverless provider authorized to run at Level 4 in mixed traffic on a public European road.

  Jacksonville's Transportation Authority announced in December 2016 that it would replace the Jacksonville Skyway monorail with driverless vehicles able to continue onto ordinary streets. The renamed Ultimate Urban Circulator, or U2C, project has since tested shuttles from six different manufacturers, at an estimated cost of $379 million.

  Baidu said in July 2018 that it had built 100 of its 8-seat Apolong shuttle. By July 2021, though, the model had still not entered volume production. Toyota unveiled its 20-passenger e-Palette in December 2020, intending it for the 2021 Tokyo Olympic Games and for wider commercial use before 2025. During those Olympics, a fleet of twenty e-Palette vehicles ferried athletes around the Athletes' Village. Each vehicle could carry 20 people or 4 wheelchairs, at a top speed of 20 mph. On the 27th of August 2021, one of the vehicles struck and injured a visually impaired pedestrian, pausing the service. It resumed on the 31st of August with improved safety measures.

  Navya's investor report in January 2021 forecast global autonomous shuttle sales reaching 12,600 units by 2025, worth an estimated 1.7 billion euros. By June 2021, Chinese manufacturer Yutong said it had delivered 100 of its 10-seat Xiaoyu 2.0 buses for use in Zhengzhou. There, the shuttles already ferry workers between factory buildings.

  A 2021 survey covering more than 100 shuttle trials across Europe found that low operating speed was the leading barrier to wider use. Researchers also pointed to cost, pricing the vehicles at around 280,000 euros each. They also cited the ongoing need for a human attendant riding along.

  Britain's first autonomous bus trial began in mid-2019. An Alexander Dennis Enviro200 bus, fitted with software from Fusion Processing, drove itself around Stagecoach Manchester's Sharston depot. Passenger service followed in January 2023, first across the Forth Road Bridge on Stagecoach Fife's route. A separate trial used a Fiat Ducato minibus serving Oxfordshire's Milton Park.

  Every one of those shuttles carries a handful of passengers over a short, fixed loop. That scale of ambition sits worlds apart from freight trucks already driving themselves across entire countries.

• 5. Freight Without a Driver

  ›

  Suncor Energy, a Canadian energy company, and the Rio Tinto Group were among the first to swap human-driven mining trucks for computer-run ones. In April 2016, trucks built by Volvo and the Daimler Company completed a week of autonomous driving across Europe. The convoy was organized by the Dutch. IHS Incorporated projected, in a June 2016 report, that US sales of self-driving trucks would reach 60,000 units by 2035.

  Popular Science reported in June 1995 that engineers were developing self-driving trucks for military convoys, where only the lead truck carried a human driver. The trailing trucks were meant to rely on satellite links, an inertial guidance system, and ground-speed sensors to follow along. Caterpillar Incorporated partnered with Carnegie Mellon's Robotics Institute in 2013 to raise efficiency and cut costs at mining and construction sites.

  Google's parent, Alphabet Incorporated, spun off Waymo. In March 2018, the company said it would apply its technology to semi trucks, hauling freight to data centers near Atlanta, Georgia. Uber completed the first driverless delivery on public roads in October 2016. It moved a trailer of Budweiser beer overnight from Fort Collins, Colorado to Colorado Springs. The Colorado State Police closed the highway ahead of the truck. That truck ran on technology from Otto, a company Uber had acquired in August 2016. A human sat in the cab, but not behind the wheel.

  Embark Trucks drove an automated semi 2,400 miles from Los Angeles to Jacksonville, Florida, along Interstate 10. Announced in February 2018, it was the first cross-country automated trip. Tesla, Incorporated, led by Elon Musk, revealed its Tesla Semi prototype in November 2017. The vehicle, the company said, could drive itself and travel in self-following platoons. Starsky Robotics chose a narrower path in 2017, building toward Level 3 autonomy. There, a human driver still has to be ready to intervene on request. In December 2018, Anthony Levandowski unveiled Pronto, a San Francisco company. It built Level 2 driver-assistance technology specifically for commercial trucking.

  A PwC Strategy and Report analysis estimated self-driving trucks would affect around 3 million truck drivers in the United States. It also flagged around 4 million further jobs in gas stations, restaurants, bars, and hotels along trucking routes. That same appetite for automation, though, has already spread past trucks and highways. It now reaches machines that never touch a road at all.

• 6. Wheels, Rails, Wings and Water

  ›

  Honda, inspired by its Uni-Cub personal mobility device, built a motorcycle that lowers its own center of balance by extending its wheelbase. It then steers itself upright at a stop. BMW Motorrad's ConnectRide concept goes further still, handling emergency braking, tight turns, intersections, and front-impact avoidance. It can even drive itself at normal speed before returning to a set location. Yamaha's Motoroid can balance on its own, recognize its rider, and travel to a chosen spot on a hand gesture. It also communicates back through a gentle squeeze against the rider's lower back at dangerous speeds. Harley-Davidson filed a patent for a rear-mounted gyroscope that holds a motorcycle upright below 3 miles per hour, disengaging automatically above that speed.

  The Port Island Line in Kobe, Japan opened in 1981 as the world's first driverless urban transit system. London's Thameslink route later launched the first self-driving train in the UK. The city's Docklands Light Railway runs as another automated network. Potsdam trialed its first autonomous trams in 2018.

  An automated guided vehicle follows floor markers, wires, magnets, or lasers to move materials around a factory or warehouse. That use expanded through the late 20th century. Aircraft rely on autopilot for autonomous navigation, and delivery drones now come from UPS Flight Forward, Alphabet's Wing, and Amazon Prime Air. Zipline, an American medical drone delivery company, runs the largest active drone delivery operation in the world and flies at Level 4 autonomy. Iceland, alongside Costa Rica, Italy, the UAE, Sweden, and Norway, has taken an unusually permissive approach to licensing commercial drone flights.

  DARPA launched the Sea Hunter, an autonomous unmanned surface vehicle, in 2016 as part of its Anti-Submarine Warfare Continuous Trail Unmanned Vessel, or ACTUV, program. Underwater vehicles, meanwhile, focus on tasks like pipeline inspection and mapping the sea floor. Spot, a four-legged robot, navigates indoor and outdoor terrain using 360-degree vision cameras and gyroscopes, staying balanced even when pushed. It can carry heavy loads for construction crews or military personnel across rough ground.

  None of these machines, on rails, underwater, or in the air, faces quite the same public skepticism. That skepticism has followed cars sharing ordinary roads with everyone else.

• 7. Trust, Doubt and the Rulebook

  ›

  Around 2015, companies including Nissan and Toyota promised self-driving cars would be ready by 2020, a target that proved far too optimistic. Companies remain focused on Level 4 automation, which works only under specific conditions, since fully unrestricted Level 5 autonomy is still unsolved. Researchers still argue over whether to include lidar at all. Some camera-only algorithms rival lidar's performance, though cameras alone sometimes draw inaccurate bounding boxes and produce poor predictions.

  Kelley Blue Book's nationwide 2016 survey found that most people preferred keeping some control over their own vehicle. They trusted that more than complete, Level 5, autonomy. Half of those respondents said perceived safety actually declined as autonomy increased. A separate 2019 survey by the AAA Foundation for Traffic and Safety found the same distrust persisting. The deepest skepticism was aimed at Level 5 vehicles. That survey also found trust rose alongside people's understanding of the technology.

  At least 113 autonomous vehicle related accidents occurred by 2018, despite the technology's aim of reducing crashes and their severity. The National Transportation Safety Board found that a self-driving Uber car was involved in a 2018 accident. It failed to identify a victim in time to slow down and avoid the crash. Tesla, Incorporated recorded one fatal accident involving its automated driving system, in a Model S. An inattentive driver was at fault, and the autopilot failed to recognize the obstruction ahead. The book The Driver in the Driverless Car: How Your Technology Choices Create the Future describes a separate incident. A driver used an app to summon his Tesla after an overnight software update. The car then crashed into his garage door.

  Ford's Blue Cruise technology lets geofenced cars drive themselves only within mapped and approved areas.

  Researchers Steven Umbrello and Roman V. Yampolskiy have proposed a value sensitive design approach. Their aim is to build ethical human values directly into how autonomous vehicles are designed. A fully autonomous, Level 5 vehicle facing an unavoidable accident can only act on what it was programmed to do. A human driver, by contrast, relies on split-second judgment.

  A proposed amendment to the Vienna Convention on Road Traffic defines an automated driving system. It is one that exercises sustained dynamic control using both hardware and software. That amendment was due to take effect on the 14th of July 2022, unless governments rejected it before the 13th of January 2022. The British Highway Code ties itself to the Automated and Electric Vehicles Act 2018. It still requires a driver using assisted features to stay ready to resume control at any moment.

  There is already a push to introduce dead man's switches into automotive use, aimed primarily at heavy vehicles. Penalty switches for cruise controls have been raised as a further possibility. That kind of switch, more than any sensor or algorithm, still puts the final call in a human hand.