History of Self-Driving Cars
The promise of self-driving cars of making our transportation safer, more convenient and more efficient is clearly immensely attractive. No longer would it be necessary to concentrate on the driving task, and we would have more time for other activities. But numerous other reasons justify the research and development of driverless technology. Some examples include making transportation more productive and affordable, ensuring access to transportation for people with disabilities, decreasing pollution and tailpipe emissions by up to 66 %, and many more⁶.
The dream of autonomous driving is likely as old as the automobile itself, yet complete autonomy still remains one of the grand challenges of computer science and engineering. In this article, we will look into the historical background of the development of driverless cars and the many setbacks that highlighted the challenges of autonomous driving.
The Beginnings of the Automotive Industry
While we can go as far as the 18th century when talking about the first automobiles, the first sensible milestone for the scope of this article is the year 1886. That year, Carl Benz developed the first gasoline-powered automobile with an internal combustion engine — the ”Benz Patent-Motorwagen Nummer 1”⁸.
In 1908, the Ford Motor Company introduced the Ford Model T — the first automobile to be mass-produced on a moving assembly line⁴. Naturally, all of these cars and models introduced in the following decades were driven exclusively by a human driver.
First Autonomous Steps
The first attempt towards autonomous driving was made in 1925 by Houdina Radio Control with the ”American Wonder” — a radio-controlled car driving through heavy traffic on New York City streets. This car was technically driverless, but it was still fully controlled by an operator in another car following the vehicle. Therefore, even though it marked an important step towards the invention of self-driving cars, it is still far from what we think of as fully autonomous driving.
For the next few decades, others followed a similar idea of merely navigating the car via an external controller. A popular approach was embedding some kind of navigational circuit into the roadway. The first example of such guidance system was exhibited at the 1939 World’s Fair sponsored by General Motors — their prototype was guided via magnetic cables embedded directly in the roadway. This idea was continued by many researchers, e.g. RCA Labs, throughout the 1950s, going from controlling a miniature car by wires laid on a laboratory floor in 1953 to a full-size demonstration in 1957 and 1960.
During the 1950s and 1960s, GM showcased a series of experimental cars called ”Firebirds”, described as having an ”electronic guidance system [that] can rush it over an automatic highway while the driver relaxes”.
Similar experiments were carried out during the 1960s in the UK’s Transport and Road Research Laboratory with the Citroen DS19, presented in 1970. It was able to go through the test track at 80 miles per hour and was expected to prevent around 40 % of the accidents, should the development have continued. This was one of the last milestones in technology based on an infrastructure of cables built into the road, as the funding for these experiments was withdrawn by the mid-1970s, followed by preliminary research into the intelligent systems conducted at the Coordinated Science Laboratory of the University of Illinois.
Towards Real Autonomy
During the 1980s, the first vision-based navigation systems were implemented in three major projects. The first vehicle to use this technology was Carnegie Mellon University’s Navlab project¹³, developed between 1986 and 1995 — the last iteration reported an autonomy of 98 % 2850 miles across the USA.
The second example was Mercedes-Benz’s robotic van, which achieved a speed of nearly 60 miles per hour on streets with no traffic¹⁰. Subsequently, the EUREKA organisation conducted the largest RD project ever in the field of driverless cars by providing funding of 749,000,000 to the Prometheus Project from 1987 to 1995³. The project demonstrated several components of autonomous driving, such as collision avoidance, lane-keeping support, and travel and traffic information systems.
Another example in the same decade was ALV, or ALVINN (Autonomous Land Vehicle in a Neural Network)⁵, funded by DARPA, develped in 1988. It used a 2-layer fully connected neural network to convert input from LIDAR and road images to vehicle commands, driving at the speed of up to 19 miles per four.
Technological Revolution
While the development of autonomous vehicles gradually progressed as the systems became better and better, there are three waves of a “technological revolution” that were of particular importance to the field.
First Wave — Navigation Systems and the DARPA Challenge
The first wave focused on exploitation of the emerging navigation technology, namely Global Positioning System (GPS) and Inertial Measurement Units (IMU), developed around the turn of the century. With GPS, autonomous driving systems could determine the car’s position within one meter — while this was an incredible achievement, it was not sufficient for precise driving and navigation in road traffic. This was improved with the introduction of IMUs, which measure the acceleration forces within the car and which increased the accuracy of the GPS navigation to up to 5 cm.
Open challenges and large competitions are essential factors for further innovations in many research areas — for example, the ILSVRC challenge⁹ intended to spur the development and research of image recognition techniques. The equivalent in the autonomous driving domain was the Defense Advanced Research Projects Agency (DARPA) Grand Challenge in 2004. The first competition took place in the Mojave Desert region of the USA, and the task was to navigate through a 150-mile route. A prize of one million dollars was offered for first place; however, none of the participating teams finished the route. The farthest distance was travelled by CMU’s Humvee, completing only 7.32 miles before getting hung up on a rock after making a switchback turn. Therefore, a second DARPA Grand Challenge was scheduled for the following year — this time, five teams managed to finish the route autonomously to the end.
Second Wave — Tech Giants Join the Battle
Over the years, the quality of sensors used in the vehicles improved drastically, and vision-based systems became more and more popular. Around 2006, better LIDAR systems were developed, camera systems improved significantly, and first methods for precise 3D reconstructions were introduced, which helped with recognising and locating obstacles. These technologies significantly advanced the research of autonomous driving, and soon, large tech companies entered the field. For example, in 2009, Google began secretly developing their prototype called ”Waymo”. By 2015, they reported an investment of $1 billion into the research and development. Recently, they also published a large-scale reconstruction of San Francisco from 2.8 million photos using a technique based on Neural Radiance Fields (NeRF)¹¹. Compared to the previous methods in large-scale reconstruction, Waymo achieved superb results.
Many of the major automotive manufacturers, such as GM, Ford, Audi, BMW, …, have been working on driverless car systems since 2005. In 2010, a self-driving car drove nearly 10,000 miles from Italy to Shanghai as a part of the VisLab Intercontinental Autonomous Challenge (VIAC), autonomously following a guide car driving ahead. More experiments in small, geofenced areas with real traffic were carried out, such as the Stadtpilot project developed by Technical University of Braunschweig in 2010⁷. During the same year, first concerns related to laws and ethics were raised — specifically, an attorney for the California Department of Motor Vehicles said that ”the technology is ahead of the law in many areas”, citing the state law requiring the vehicles to have a human operator at all times¹².
Third Wave — Technological Revolution
The next major milestone of the development that could be considered the third wave of the technological revolution occurred in 2012. With the advancement of Machine Learning, Representation Learning and Deep Learning methods, self-driving systems became more and more accurate. New competitions and benchmarks made it possible to compare different technologies and methods in a uniform manner, which allowed for an incentive for many research teams to keep developing new techniques.
In 2014, Mercedes-Benz introduced its S Class model equipped with features such as automatic lane keeping, autonomous steering, acceleration and braking, as well as collision avoidance at high speeds. In the same year, the Society of Automotive Engineers (SAE) published a taxonomy comparing different levels of self-driving technologies, creating five categories of autonomy¹¹. According to this system, the S Class achieved a Level 2 Autonomy.
Self-Driving Industry of Today
The increasing involvement of large companies in developing certain components or complete systems for self-driving cars soon formed an entire industry. Many large companies, such as Uber and Tesla, started whole research departments in 2015 and 2016, hiring numerous scientists and engineers. Unfortunately, the development was again marked by setbacks, especially in 2016 and 2018, due to two fatal Tesla vehicle accidents, raising legal questions and lowering trust in autonomous driving technology.
In 2017, Audi announced that their new A8 model would be fully self-driving. Contrary to the other cars, there would be no need to do safety checks such as touching the steering wheel every 15s. To reference the Levels of Autonomy chart, this would mean Level 3 autonomy. However, the system was never fully implemented, and in 2020, Audi announced that the system was not going to be activated².
In the 2020s, Tesla released a beta version of their FSD system to a small group of testers in the USA. Mercedes-Benz has received German approval for a Level 3 autonomous system focusing on lane keeping. Later, they launched the sales of their Drive Pilot system in Germany for the S Class and EQS models, operating at Level 3.
Summary
This article summarized the major milestones of the development of autonomous vehicles. As we have seen, self-driving has a long history. People started thinking about ways to make vehicles autonomous earlier than one might think. Over the past decades, this technology has continued to improve and become more and more reliable until, in 2019, the first vehicle that realised Level 3 autonomy for highways was presented. In the race for more and more autonomous cars, there are now many players, including well-known ones, with Tesla and Waymo seemingly leading the field. But Level 4 and 5 autonomy remain unattained and extremely difficult to implement. In addition, several setbacks have lowered confidence in self-driving and brought ethical and legal issues to the fore. Self- driving is a complex but promising technology, and it is expected to become increasingly important in the coming decades.
References
[1] SAE International. AdaptIVe system classification and glossary on Automated driving.
Archived from the original on October 7, 2017.
[2] Stephen Edelstein. Audi gives up on Level 3 autonomous driver-assist
system in A8. 2020.
[3] EUREKA. Eureka Project E!45 PROMETHEUS.
Archived from the original on August 14, 2018.
[4] History.com. Model T: The First Car That Most People Could Afford to
Buy. 2020.
[5] Takeo Kanade, Chuck Thorpe, and William Whittaker. Autonomous Land Vehicle Project at CMU. In Proceedings of the 1986 ACM Fourteenth
Annual Conference on Computer Science.
[6] Nikolaus Lang, Michael R ̈ußmann, Jeffrey Chua, and Xanthi Doubara.
Making autonomous vehicles a reality: Lessons from Boston and beyond. 2017.
[7] Tobias Nothdurft, Peter Hecker, Sebastian Ohl, Falko Saust, Markus Maurer, Andreas Reschka, and J ̈urgen R ̈udiger B ̈ohmer. Stadtpilot: First fully autonomous test drives in urban traffic. 2012.
[8] German Patent and Trade Mark Office. Der Streit um den ”Geburstag”
des modernen Automobils.
Archived from the original on January 2, 2017.
[9] Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh,
Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bern-
stein, Alexander C. Berg, and Li Fei-Fei. ImageNet Large Scale Visual
Recognition Challenge. International Journal of Computer Vision (IJCV). 2015.
[10] Jurgen Schmidhuber. Prof. Schmidhuber’s highlights of robot car history. 2009.
[11] Matthew Tancik, Vincent Casser, Xinchen Yan, Sabeek Pradhan, Ben
Mildenhall, Pratul Srinivasan, Jonathan T. Barron, and Henrik Kretzschmar. Block-NeRF: Scalable large scene neural view synthesis. 2022.
[12] The New York Times. Google Cars Drive Themselves, in Traffic. 2010.
[13] Carnegie Mellon University. The Carnegie Mellon University Autonomous Land Vehicle Project (NAVLAB). 2014.
