Self-Driving Cars Have Been Around for Longer Than You Think

The successful deployment of driverless taxis in US cities such as San Francisco and Phoenix, Arizona has shown that autonomous vehicle technology is now close to maturity, although perhaps not yet at the point of public acceptance given the continuing debate over the safety of autonomous vehicles in urban environments. As one of the most sophisticated technologies in common use today, you would be forgiven for thinking of self-driving cars as being a modern phenomenon but autonomous vehicle technology has a rich history involving many contributions stretching back more than 70 years. This article highlights three of these contributions, each of which was a major milestone in the history of autonomous vehicles.

The Automated Highway

In February 1958 General Motors issued a press release announcing the successful trialling of an automatically guided car at the US automotive behemoth’s Technical Center in Warren, Michigan. This concept demonstrator was the culmination of a 5-year collaboration with electronics company the Radio Corporation of America (RCA) to develop an automated highway system. The two firms had first worked together on the creation of the ‘Futurama’ exhibit for the 1939 New York World’s Fair, a huge animated scale model of a city of the future complete with radio-controlled electric cars which offered an intriguing foretaste of their achievements to come.

The GM/RCA automated highway system combined electromagnetic guidance with radio control. For steering, a pair of pick-up coils fitted to the car’s front bumper tracked the electromagnetic field generated by a steel cable embedded in the road which carried a low frequency alternating current. The signals from the coils were fed to an electronic analogue computer located in the car’s glovebox which compared amplitudes and sent the appropriate correction signal to a valve controlling the car’s hydraulic power steering if they were unequal.

For collision avoidance, a system devised by legendary RCA engineer Vladimir Zworykin utilised the inductance principle to detect each vehicle as it passed over a series of rectangular wire loops embedded in the road. This enabled the speed and relative proximity of vehicles to be determined and potential collisions identified. Roadside radio transmitters then communicated this information to the car’s on-board computer which adjusted its speed as necessary. Initially, speed control required the retrofitting of a velocity transducer to the car’s speedometer plus servomechanisms to its throttle and brakes but the project team expected this to be unnecessary on future versions by making use of factory-fitted cruise control systems such as the recently introduced Chrysler Auto-Pilot.

The GM/RCA system was the first in a number of similar automated highway projects conducted by the automotive industry during this period but all would remain technology demonstrators as the infrastructure costs would have been prohibitive. Successful implementation also relied on every vehicle that used the highway being fitted with the necessary control equipment, an unrealistic scenario given the 67 million cars already on US roads in 1958. A self-contained solution which could operate independently on any type of road would be far more practical. However, this would require the development of new technologies for collision avoidance plus a much more powerful computing capability that was small enough to fit within the vehicle. It would be almost two decades before these building blocks were in place.

The Intelligent Vehicle

One of the earliest efforts to develop a self-contained autonomous driving capability took place in the 1970s at the Mechanical Engineering Laboratory of the Japanese Ministry of International Trade and Industry (MITI). The ‘Intelligent Vehicle’ project was prompted by the Japanese government’s continued desire to ease traffic congestion in Japan’s densely populated cities following an earlier state-sponsored project to develop an integrated traffic control system using in-vehicle displays for route guidance and driving information.

Led by engineer Sadayuki Tsugawa, the Intelligent Vehicle project neatly combined two cutting edge technologies, machine vision for obstacle detection and a dead reckoning system for autonomous navigation. A saloon car fitted with hydraulic actuators for steering, brakes and throttle provided the test platform for these technologies. The machine vision system utilised a pair of analogue video cameras mounted one above the other on the car’s radiator grille. A stereo vision technique detected and located obstacles over a range of 5-20 metres in front of the vehicle. Image processing was performed using hard-wired logic circuits rather than a programmable computer. This sacrificed programmability for performance, allowing the system to operate in real-time without the need for a large remote computer. Dead reckoning was effected using rotary encoders fitted to the car’s rear wheels to measure their rotation. A 16-bit microcomputer then used this data to calculate the car’s position and heading relative to its starting point. The microcomputer also handled the overall control of the car, using the information supplied by the machine vision system plus additional data from sensors which measured the car’s speed and steering angle.

By 1977, Tsugawa’s team were able to demonstrate the car driving autonomously at speeds of up to 30 km/h while avoiding obstacles and guide rails. As the first successful application of machine vision to a road vehicle, this was a major breakthrough and all the more remarkable given the lack of computing power available at that time. However, the technology was incapable of recognising road markings or traffic signs. Until these essential features were added, autonomous vehicle technology would remain unsuitable for use on public roads.

The Seeing Passenger Car

The next significant advance in autonomous vehicle technology came with a project that began in the early 1980s at the Universit?t der Bundeswehr München in Munich, Germany. The aim of the project was to develop machine vision systems for a range of vehicle guidance applications and was directed by veteran aerospace engineer Ernst D Dickmanns. In 1986 the team began to focus exclusively on road vehicles following an approach by German automotive manufacturer Daimler-Benz suggesting a collaborative project to mark the 100th anniversary of the company’s first production car, the Benz Patent-Motorwagen. The result was VaMoRs (Versuchsfahrzeug für autonome Mobilit?t und Rechnersehen), in English ‘Test vehicle for autonomous mobility and computer vision’, a Mercedes-Benz 508 D panel van fitted with a self-contained autonomous driving system based on machine vision.

The choice of a van instead of a car was made for VaMoRs as it provided more space to accommodate the system’s bulky rack-mounted electronics and electric power generation equipment. The system featured a pair of CCD video cameras mounted centrally behind the van’s windscreen for capturing stereo images. Electromechanical actuators were also fitted to the pedals and steering mechanism to operate the controls. With microprocessor-based computers not quite powerful enough to handle the computational demands of real-time machine vision, the team needed to find a workable shortcut. They developed a novel technique based on a spatiotemporal (4-D) dynamic model, where the motion of objects in 3-D space from one image to the next is predicted. This removed the requirement to store previous images, thereby reducing the computational load on the system. Further gains were made by identifying and analysing only the most relevant regions of an image, such as areas of high contrast. Processing was performed by an on-board computer system comprising ten Intel 8086 16-bit microprocessors operating in parallel. The combination of spatiotemporal machine vision and parallel processing allowed the van to drive autonomously at speeds of up to 96 km/h.

In 1987, VaMoRs was demonstrated driving autonomously on a public road over a distance of 20 kilometres, albeit on a closed-off section of Autobahn with no other traffic to deal with and a human driver in the cab ready to assume control if any mishap occurred. Nevertheless, the basic capabilities had been proven. That same year the Eureka PROMETHEUS (PROgraMme for a European Traffic of Highest Efficiency and Unprecedented Safety) initiative was launched. Boasting an eye-popping €749 million budget and the involvement of automotive manufacturers from six European countries, PROMETHEUS was aimed at supporting collaborative projects between industry and academia which would lead to improvements in vehicle safety and efficiency. For the VaMoRs team, the timing was perfect.

Dickmanns submitted a research proposal which argued against the use of electromagnetic guidance via embedded cables, which some PROMETHEUS participants were advocating, in favour of a self-contained autonomous navigation solution based on the VaMoRs machine vision technology. The proposal was endorsed by Daimler-Benz, with the manufacturer also promising to provide two Mercedes-Benz 500 SEL saloon cars for use as test vehicles. The team’s track record and the backing of Daimler-Benz helped to ensure that the proposal was successful, and the additional funding allowed the team to extend the capability of VaMoRs considerably.

Designated VaMoRs-P, the extended system added an obstacle recognition and avoidance capability which employed both forward facing and rearward facing video cameras to detect objects located at any position around the periphery of the vehicle. Each camera pair could simultaneously track up to 6 objects over a range of 5-80 metres. Intelligent control software then used this information to continuously assess the traffic situation around the vehicle and react accordingly. An upgraded computer system consisting of 60 Inmos transputers, high-performance microprocessors specifically designed for parallel processing, provided the necessary processing power. Though much more compact than the original VaMoRs equipment, the new system still took up all the available space in the car, leaving only the front seats free for a safety driver and passenger.

VaMoRs-P was unveiled at the PROMETHEUS wrap-up meeting held in Paris in October 1994. Both cars were demonstrated driving autonomously on a public road in heavy traffic at speeds of up to 130 km/h. This remarkable demonstration also featured the cars driving in convoy and performing autonomous lane changes and overtaking manoeuvres subject to the approval of the safety driver. The following year, one of the cars completed a journey of 1,758 kilometres from Munich to Copenhagen, Denmark, and back, covering up to 158 kilometres at a time without human intervention. Dickmanns’ team had highlighted the potential of autonomous driving technology in spectacular fashion but the equipment was extremely costly, bulky and complicated. Consequently, it would be many years before such technology was deemed sufficiently mature for use in a production vehicle.

The Rest of the Story

Of course, these projects are only part of the story. Complex technologies are seldom developed in isolation. Most evolve from earlier inventions or are an amalgamation of existing technologies. Today’s self-driving cars would not exist without the pioneering efforts of scientists and engineers engaged in the development of autonomous mobile robots during the 1970s and 80s, or the encouragement of the US military’s Defense Advanced Research Projects Agency in the early 2000s. For a fuller account of the development of autonomous vehicle technology, see Chapter 7 of my book, ‘The Story of the Robot: A Short History of Automation and Robotics’.