Thank your for your subscribe
Oops something went wrong. Please check your entry

Robotics Series - 1 - A coming loop revolution for driverless cars

The next automobile revolution will be the driverless car. Ever since the automobile industry was born, it has continuously represented a clear-cut vector for technological progress and societal transformation. Driverless cars, with general roadworthy models running by 2030, promise to bring sizeable changes to the way we live, to the environment, to our health, the economy and industry at large. Market potential is already visible but many questions still remain to be answered.

June 2014
lire en français
lire en français
Lire le résumé

The next automobile revolution will be the driverless car. Ever since the automobile industry was born, it has continuously represented a clear-cut vector for technological progress and societal transformation. Driverless cars, with general roadworthy models running by 2030, promise to bring sizeable changes to the way we live, to the environment, to our health, the economy and industry at large. Market potential is already visible but many questions still remain to be answered.

This article is the first of a series dedicated to robotics, which will be published within the next months.

Has the time come to herald the driverless car? Since Google launched its experimental Google car in 2010, there has been no lack of announcements. In its conclusions (in a report published 2014), the specialist agency Navigant Research estimates that sales of driverless cars would progress from less than 8,000 in 2020 to 95.4 million by 2035, i.e., at that horizon, a figure that represents 75% of all new, light cars on the road. From a technological point of view, the portended revolution has already begun. Over the coming decade, levels of production capacity and various technical improvements will lead to cost reduction such that it will be possible to fit all production line vehicles with most of the equipment needed.

A really driverless car?

The ultimate objective is to see a truly driverless car, with on-board systems that completely replace the human driver. Nevertheless, the changeover will be progressive. Part autonomy is already available in 2014 models, with a combination of several customized aids to driving. Certain models today propose speed regulators, automatic emergency braking, lane crossing warnings and self-parking. These aids only take over from the driver at the wheel in the event of specific scenarios occurring or within certain speed limits.

In 2014, the driverless car that has clocked up most miles – more than a million km to date - is the Google Car, for which Chris Urmson (Director, Self-Driving Car Project) proudly announced in a recent in-house post that it had already run more than 700 000 miles (>1.1 Mkm). Let us note that rather than a self-driving car, the focus is on the navigation system installed experimentally in 8 vehicles (Toyota Prius, Audi TT, Lexus RX-450h). The current cost of the equipment is 150 000 euros and schematically it operates as follows: a rotating multidirectional sensor is attached to the car’s rooftop, with a sensing range in excess of 60m, its remote detection relying on laser beams (a “lidar”) which provides a precise 3D map of the near environment of the vehicle. A movement sensor, positioned above the rear left wheel measures and records vehicle movements, even the smallest, for the purpose of accurate geo-localizing it. A camera is placed near the inside mirror to detect traffic lights and other signals and also serves to detect mobile obstacles, such as pedestrians and/or cyclists. Lastly, there are 4 sensors on the front and back fenders, 3 at the front and 1 at the back to measure distances from the vehicle to various obstacles, for example other vehicles moving fast on motorways.


As Chris Urmson noted, on one hand there has not been as single accident so far but the system is, nevertheless, experimental and calls for continuous upgrading and improvement, most of which involves computer science and data processing developments. [Quote] “As we’ve encountered thousands of different situations, we’ve built software models of what to expect, from the likely (a car stopping at a red light) to the unlikely (blowing through it)”,  adding  “We still have lots of problems to solve, including teaching the car to drive more streets in Mountain View [Google’s global headquarters town] before we tackle another town.” In fact the navigator does not operate in abstracto but in a special environment that the system has ‘learned and recorded’. We can also observe that the team in charge of the project were also the designers of StreetView®, a global geolocalization- visualization system that eventually will cover the entire planet (except, of course, in those countries who refuse).

Thus, the driverless car will be partly autonomous, capable of taking its own marks and partly connected to a data processing system which can be minimal (simple geolocalization, for example) or maximal (with computer modelling maps of an entire city). In like manner, the system can be installed on-board (for city, neighbourhood mapping, or for home-to-work commuting) or external (with mapping for larger regions).

As far as the “autonomy” criterion is concerned, system reliability is not yet satisfactory. With a 360° scanning sensor, the system is not affected by fatigue or inattention, which is a great advantage, but it is less skilful than humans when it faces unexpected situations, for example, when a policeman manually signals orders to drivers, or when snow obliterates the lane marks and disturbs the laser readings. Because of these limitations, driverless cars in year 2020 will only be self-drivable under certain conditions and the vehicle ‘passenger’ must be in a position to immediately resume his ‘driver’ capacity under given circumstances, for example when the vehicle enters a zone with road works.

The industrial and economic challenges

Development of driverless vehicles will produce significant economic effects, for the sector’s industrialists, of course, but also beyond them.

Industry itself is already in battle order, notably the European constructors (and especially the Germans), who are moving forward. At the 2013 Frankfurt Car Show, Mercedes-Benz’ CEO, Dieter Zestche, created a sensation by arriving on the company stand sitting at the back of a driverless Class S. Audi and Nissan are also running in the front line. But it will not necessarily be the traditional manufactures who make the best of things. This is a brand-new market slot and there will no doubt be a hard fight with ‘breakthrough’ competitors such as major technology-intensive companies with a mastery of two coupled and necessary areas: managing ecosystems and handling big data efficiently. Google is in the fore here and some pundits foresee that the Corporation will carve itself a leadership niche such that it will not be possible to challenge them at a later stage. The Mountain View Company is already furbishing its arms in the field of legal requirements and has already ‘favoured’ votes on laws by the authorities of Nevada that will become the first US State to regulate driverless cars on its public highways.

Dieter Zetsche

The technologies needed are being improved and finalized very quickly and the automobile parts manufacturers are already seeking to make cost reductions. The first objective is to attain a panoramic HD 360° vision system using 4 miniature digital cameras that will offer the drivers a panoramic view of their vehicle, alerting them in the case, for example, of a child who suddenly moves onto the highway ahead or behind (if reversing). The advantage here is to have an accurate, constantly sharp image even in poor light conditions or with a high contrast environment such as that met in underground parking lots.    

The French company Valeo acquired Tuam, an Irish start-up specialized in driverless technologies to fit mass production car series: the Tuam cameras, combined with ultrasonic sensors and radar devices, assist the driving functions. Valeo is developing driverless software packages to analyse both static and mobile obstacles, i.e., other vehicles and/or pedestrians. If it is a pedestrian, the car can automatically carry out an emergency braking stop. Better still, it can check if the driver is looking ahead - theoretically therefore seeing the obstacle - and only engages the emergency stop if this is not the case.


The consultancy company KPMG published a fully documented analysis in 2012, noting that automated driving would lead to 3 concrete consequences in terms of consumption. Firstly, having less traffic jams implies substantial fuel savings. Secondly, excessive fuel consumption due to poor driving habits will be eliminated. Thirdly and lastly, the almost 100% reduction of road accidents will allow future cars to be made substantially lighter, by removal of the steel reinforcement struts currently fitted to car frames to counter crash damage. Lighter cars will need less fuel. The Earth Institute at the University of Columbia, NY-NY, published detailed report in 2013 entitled “Transforming Personal Mobility”, which led to some impressive results in 3 test scenarios.

In Ann Arbor (Michigan), a shared fleet of driverless vehicles would offer the same degree of mobility as driven, personal cars, plus a 20% cost saving, on the basis of 16 000 km/year/passenger. In Babcock Ranch, a “green city” currently under construction in Florida, some 4 000 shared vehicles could serve the mobility need of an estimated 50 000 inhabitants, with a waiting list of less than 1 minute (the customer would order the vehicle via smartphone and the nearest, free, driverless car would automatically come to the point where the call was made (home, etc.). Once the trip has been completed, the customer client would only need to press a button and the car would head back to its parking space. Last example, in Manhattan, NY, 9 000 driverless cars could ‘efficiently’ replace the 13 000 famous Yellow Cabs, for only 10-15% of the cost of the cab fleet today. In each case envisioned, the community makes huge savings in terms of parking costs and driving time spent “behind the wheel”. Americans are used to driving 250 hours/year, compared with the 1 500 to 2 000 annual hours devoted to actually working, in developed countries. If the hours spent previously at the wheel were converted into work hours, global productivity would clearly benefit.

Many actors in the automobile sector ecosystem are going to see their business model upset, notably if a rapid rise in the number of driverless vehicles coincides with a development of the sharing concept – which, when we think about it, is close to the idea of lending a car which has become “depersonified”. Automobile manufacturers are going to experience a drop in their orders, unless transport needs rise to a huge demand. In our cities, the drop in needs for parking places will offer new building construction opportunities to the city authorities, but will lead to a drop in revenues from the previous parking lots. Public transport, in particular, buses, will begin a fierce commercial fight with driverless cars and the situation may prove even more merciless with the taxi companies who will find it hard to survive. In general terms, all driving professions, in taxis, lorries, or buses …  stand to lose a lot. And, last but not least, given that there will be less accidents, the insurance companies will be forced to reduce their premiums and also, consequently, register lower revenues.

Safety factors and responsibilities

Safety requirements and associate regulations are going to favour the emergence of smart cars, which represent the first step on the way to building truly driverless cars. The American road transport authority, NHTSA, adopted a text in April 2014 that makes it an obligation for all vehicles on US roads to be retrofitted with a camera to see pedestrians behind them. Every year, there are 15 000 injured (200 fatal) persons, mainly children, who are victims of drivers who are in too much of a hurry to start the engine and move.

Human errors are the cause of 93% road accidents, as reported the KPMG consultants in 2012. Handing over the driving function to a software package could eliminate almost all such accidents, saving some 30 000 lives a year and economizing 300 billion $US in the USA alone, says the American Automobile Association (AAA). Morgan Stanley Bank –aggregating all the direct and indirect consequences of such a change, goes as far as suggesting up to 1 300 billion $US savings.

However, even if driverless cars, globally speaking, will prove safer than those with humans at the wheel, there will still be accidents due to unforeseen/undetected obstacles such as pedestrians crossing the road, fallen trees, whatever …(as it happens with human drivers). The question then arises as to how the driverless vehicle will react. If the impact is inevitable, can the vehicle be equipped to “optimize” the effects of a crash? Certain scenarios are confusing and as for all robotic systems, the options and system decisions will lead on to some very difficult ethical issues: would it be more acceptable to mow down a child or an old-age pensioner? Is it preferable to crash an older model rather a brand-new car? If the aim is to minimize costs, then there the answer is definitely to have the driverless car crash into the older model, if there is a choice of course. If the impending collision is serious, with a high risk of pasengers being injured, the choice will be to crash into the safest model, viz., the one which protects its passengers best. Or, maybe, the choice might be to crash into a motor-cycle rider wearing a helmet, rather than one without one.

Inside a vehicle, the weak link is the human being. Even in a driverless car, the passenger will have to stay alert, even when the vehicle is not moving. In practice terms, we have to be able to instantly change a passive passenger into an instantly alert driver. Connected systems – data, leisure and entertainment – play a key role in automated driving. The car manufacturers will force new owners to use only the integrated systems, so as to be able to draw the ‘driver’s’ attention in the event of an emergency, if he/she has to take over the vehicle controls at very short notice, i.e., immediately.  Integrated systems can incorporate a cut-off function for films, e-mails and reading screens. A passenger typically needs 5 to 10 seconds to reconcentrate on driving the vehicle safely.

The main obstacles on the road to driverless cars are not technical, in fact, but more in relation to questions of system reliability and legal responsibilities (in the event of accidents). In many countries all vehicles must have a driver on-board and in charge at all times. The Highway Code, subordinated to the Vienna Convention 1968, stipulates that the person sitting behind the steering wheel must be in constant control of the vehicle. Certain US States and certain European countries have begun issuing licenses to companies to carry out driverless tests on public highways but the issue of responsibilities has not yet been settled. The question is simple: who is responsible in the event of an accident? Is it the car owner, the manufacturer, the software package designer, or the GPS supplier/programmer? The concept of “driver” will be replaced by the notion of the “operator” who programmes the GPS unit. In theory a driving license will no longer be required. Does this suggest that children could travel unaccompanied by adults in a driverless car? The answer here is not self-evident: the question of legal responsibility will complicate even further the definition of the “driver”.

As far as transport infrastructures are concerned, they will have to be adapted to meet the systemic consequences as and when driverless cars are allowed on the roads. The Sartre project, financed by the European Commission, will investigate the concept of convoy driving, viz., several vehicles electronically “tethered” to a pilot vehicle with a professional operator at the controls. We shall have envisage limited access motorways for convoy traffic. If the vehicles learn how to load and unload passengers and freight, they will be able to use parking spaces distant from the passengers’ home (or target) address. Moreover, since the driverless vehicles will follow lanes more closely than human drivers, their width can be narrower. Lastly, we shall have to solve the issue of transition: if driverless cars are on the same roads as human-driver cars, maybe the former will have to learn to adapt to human driver road behaviour, especially where speed limits are concerned, so as to assure continuous safety separation distances to other road users? Or again, perhaps we shall have to invent flexible safety rules that adapt in real time to local conditions?