Driverless cars are not a new idea. Like other forms of transportation, efforts to automate cars began not long after the car was invented. And though we are finally on the cusp of that elusive goal, automating the task of driving has confounded scientists, engineers and inventors for the past century.
For the sake of thoroughness though, we need to go back even further to fully cover the history of this technology. Most likely, you don’t need to be convinced that Leonardo da Vinci was a man ahead of his time. He was dreaming of helicopters, planes, robots, parachutes, and scuba gear centuries before these technologies saw their first real use. But, if you are still a sceptic or you simply like to be amazed, consider this: da Vinci invented what many consider the first driverless vehicle nearly four hundred years before Karl Benz produced the first automobile. His “cart” was self-propelled and would travel a straight or curved route based on the way it was “programmed.”
The first public demonstration of autonomous vehicles didn’t come until 1925. A company called Houdina Radio Control operated a car (later call the Phantom Auto) with no-one behind the wheel along Broadway and Fifth Avenue in New York City during the middle of a traffic jam. Much like today, that demonstration, and subsequent demonstrations, were driven by safety concerns. Accidents involving cars in that era were horrendously frequent. Alas, the demonstration was more sparkle than substance, more a highlight of radio technology than autonomous driving. Rather than being self-driving, the car was operated via remote control by a driver in a nearby car. The Phantom Car didn’t represent a meaningful leap towards true autonomous driving.
Fifteen years after the Phantom Auto cruised through New York City, Norman Bel Geddes and GM captured the world’s imagination at the 1939 World’s Fair with Futurama. The exhibit represented his vision of cities in the the not-too-distant future. A key part of that vision was autonomous cars cruising safely along highways to and through cities, technological leaps he thought would become reality by 1960. The vehicles would be controlled by radio and propelled electromagnetically with cables embedded in the pavement. Importantly, the path he envisioned towards autonomy required smart infrastructure more than it required a smart car. And, much like we hear today, autonomous vehicles promised to reduce accidents, congestion and parking demand.
By 1953, RCA built a miniature working protoype of a driverless car based on Bel Geddes’ vision. Rather than using the cable to propel the car, however, RCA used the embedded cable to guide it. Sensors in the car were used to track the cable and steer the car along the path. Sufficiently impressed with the demonstration, the Nebraska Department of Motors decided to partner with RCA to build a full scale model in Lincoln, NE. The test was a success! A second demonstration was successfully completed in Princeton, NJ a few years later. Similar tests were conducted successfully in the UK around the same time. A study from the UK tests predicted the technology would reduce congestion by 50% and accidents by 40%, and that the cost of adding the technology would be paid off by the end of the century. Despite those findings, no broader implementation of the technology was pursued in the US or the UK. However, GM did develop multiple prototypes of a car called the Firebird that was designed to work on these smart highways.
The dawn of the computer age brought a fundamental shift in efforts to automate driving. Previously, research focused on designing smart highways. Computers opened the door to smart cars.
Efforts to imbue vehicles with senses and intelligence began small. Like da Vinci before them, a team at Stanford began by experimenting with a cart. The goal was to test the feasibility of operating a lunar rover from space. The cart had four bicycle wheels, a camera and a remote control cable. In reality, that first test was more akin to Houdina’s Phantom Car than an autonomous vehicle, but it proved that controlling a lunar rover from Earth wasn’t feasible. Over the years, the cart was gradually upgraded until, in 1979, it successfully navigated a chair filled room without human intervention in around 5 hours. The Stanford Cart thus had the basic building blocks of a smart car. It could sense it surroundings and make navigation decisions towards a destination without intervention from a human.
These abilities were first added to a car by a team from Japan in 1977. The vehicle was equipped with two cameras and an analog computer. It was capable of traveling up to 19 mph, but required a rail for guidance.
Autonomous cars took a giant leap forward in the 1980s with research conducted by German engineer Ernst Dickmanns. He started by outfitting a van with two cameras, a 16-bit Intel Microprocessor and a suite of software. The first iteration of the van was capable of traveling 56 mph and it marked the first time a vehicle could operate autonomously without infrastructure improvements.
The test was impressive enough to trigger Prometheus, an autonomous car research project funded with a staggering investment (roughly $1.3B in today’s dollars) from EUREKA. The project culminated in two autonomous cars (VaMP and VITA-2) and a demonstration in real traffic at speeds of up-to 80 mph. The vehicles included four cameras, radar and robust array of computers.
Around the same time as Prometheus, DARPA launched a project called ALV (Autonomous Land Vehicle) in the US. The vehicle was designed for an off-road, rather than an urban, application, but it represented the first meaningful use of Lidar to enhance machine vision in autonomous vehicles.
The United States government broadened its commitment to autonomous vehicles in 1991 with the ISTEA Transportation Authorization Bill. The legislation directed the USDOT to demonstrate automated vehicles and highways before the end of the decade. In achievement of its congressional mandate, USDOT and a host of public, private and academic partners presented a demo of the technology in San Diego. The demo included tightly packed vehicle platoons and mixed-traffic operations. Government funding for the research largely ceased not long after the demonstration. As with previous demonstrations, the technology was impressive but fell short of the capabilities necessary to handle the complexity and sheer range of potential driving environments.
DARPA’s Grand Challenge: The Tide Turns
DARPA jumped back into the driverless vehicle fray in a big way with its Grand Challenges. The first Grand Challenge involved a $1M prize to the contestant that could develop a car capable of autonomously navigating an off-road course. Fifteen vehicles were entered into the competition, but not one of them was capable of completing the course.
One year later, the challenge was repeated. This time there were 23 entrants into the competition. The progress that was made in a single year is stunning. Five vehicles finished the course and all but one vehicle performed better than the entire field from the preceding year. The vehicles used an incredible array of cameras, radar, lidar, etc. to observe their environments and sophisticated software to interpret the data and make navigation decisions. The challenge highlighted the huge advances that had been made in both the hardware and software of autonomous driving.
The third DARPA Grand Challenge was nicknamed the Urban Challenge because it focused on driving in an urban condition with simulated traffic. Amazingly, 53 team’s proposed for the competitions. Only 11 were allowed to compete because of safety considerations. A team from Carnegie Mellon won the competition, but 6 teams completed the course, which is remarkable given the complexity of the challenge.
The significance of the Grand Challenge’s is that it demonstrated the enormous advances that had been made in machine vision and computer technology that had converged to make smart autonomous vehicles feasible. Prior efforts and demonstrations, while remarkable in their own right, highlighted the limited ability of autonomous vehicles to handle the messiness of our traffic environments. Prior iterations of autonomous vehicles would have required dedicated lanes or new infrastructure to safely operate. The Grand Challenge hinted that that might no longer be the case.
The Grand Challenge did not immediately spur the avalanche of funding and research into autonomous vehicles that we see today. But it did catch the attention of one company: Google. Not long after the final Grand Challenge, Google hired a team of engineers (many of whom competed in the challenge) to begin a secret project on autonomous vehicles. When Google’s efforts finally became public, many, including most car manufacturers, didn’t take them seriously. But Google persisted and their technology improved. They quickly mastered highway driving and moved their attention to city driving. By 2014, they were confident enough in their technology that they developed a prototype for a vehicle that had no steering wheel and no pedals, just an on and off button.
By this time, Google had firmly caught the attention of the industry and an army of entrepreneurs. It was not be long before every major automaker, a host of well-resourced tech companies and an army of start-ups, were investing heavily in research and development around autonomous vehicles. Companies are now racing to be the first to offer fully autonomous technology to the public. Others are focused on refining the hardware for a self-driving application. Some companies have started rolling out partially autonomous features and others have already started piloting robo-taxi services. The launch of the first commercial, fully autonomous vehicle is, after nearly a century of dreaming and tinkering, finally nigh.
Wikipedia: History of autonomous cars
Wired Magazine: Autonomous Cars through the Ages
Computer History Museum: Where to? A History of Autonomous Vehicles
ThoughtCo.: History of Self-Driving Cars
Engineering.com: The Road to Driverless Cars: 1925-2025