The main cause of traffic accidents is human error, but sorting out the tech challenges of driverless cars solves only half the problem.
Billed as 'luxury in motion,' the Mercedes F 015 has been wowing car show audiences around the world. It's not your ordinary luxury car - there are no windows, and the inside walls are covered with interactive screens. But most importantly, the driver can opt out and enjoy some quality time while the vehicle does the driving.
It's a concept car, and who knows if or when it'll get onto our streets. But Mercedes' move shows that more and more carmakers - including Tesla, Audi, Nissan and Ford - envisage a future for driverless cars and are investing heavily in the technology.
They are not alone. Silicon Valley tech giant Google has promised that by 2020 self-driving cars will be common, and is constantly trialling its own driverless concept cars. To date, Google's automated vehicles have clocked up more than 1 million miles on US streets and, provided new legislation is passed, we may see them in the UK very soon too.
There are a multitude of reasons automated technology could benefit us. An estimated 90 per cent of accidents are caused by driver error - a figure the industry hopes driverless cars could help reduce drastically thanks to more reliable and logical robotic thinking. More efficient traffic flows and economical driving would benefit the environment and bring down our stress levels as well.
But there's a long way to go before we'll see driverless cars take over. Aside from the engineering challenges and security concerns, according to John McCarthy, technical lead at UK-based design and engineering firm Atkins, there are a few fundamental questions we need to ask. 'Technology, the connection between systems and security are all important - but what is most important is our interaction with technology,' he says. 'We don't know how people will react as pedestrians and passengers. Will they be happy to give up control?'
The idea of a self-driving vehicle isn't new. In the 1960s researchers at the Transport and Road Research Laboratory in Berkshire developed a 'self-driving' Citroen DS car that was able to follow a series of magnetic cables embedded in the road. Since then, the industry has followed an evolutionary approach, gradually adding elements of automated technology to vehicles, like parking and lane sensors. Some of the technology, like telematics, has been around for over a decade. Telematics itself is a catch-all term for the converging of telecommunications technology with informatics. In recent times, this has led to a whole host of new developments, including parking sensors, GPS navigation, and automatic driver-assistance technology.
Driverless obstacle course
Creating a functioning and safe driverless car challenges the engineering fundamentals of today's vehicles. Traditional car design is all about putting the driver in control of his or her vehicle. The traffic code, road design and signage are also focused on the driver. Changing all this is a big task.
So if we want to go driverless, we first have to tackle miles of red tape and legal issues. In its recent 'Pathway to Driverless Cars' report, the UK Government set out its ambition to make Britain a key territory for the development of driverless cars. A £10m investment by Innovate UK (a science and technology funding body formerly known as the Technology Strategy Board) supports four driverless car projects across the country. The purpose of these projects is not to create new technologies, but to understand, measure and assess their potential impact on society and individuals.
One of the key challenges facing any team developing a driverless car is identifying a safe test environment, and the UK projects have opted for a virtual one. This is what racing car drivers do, after all, as they get used to the world's most challenging circuits. Williams, a Formula One team, uses a detailed simulator that enables drivers to get a feel for the track - and it's also a great deal cheaper than doing it for real. As part of a consortium dubbed Venturer, which brings together tech companies, universities, insurance firm AXA and local authorities in Bristol and South Gloucestershire, Williams F1 now has to create a virtual environment that mimics the real-life conditions driverless cars might experience. 'The computer has to believe it's in a real-life environment, which means we have to create a simulation with extreme accuracy,' explains McCarthy.
One of the biggest concerns is just how a machine will deal with the unexpected: those events or situations that are impossible to predict. 'The advanced simulator technologies we are working on can replicate rare or unnatural environments,' says McCarthy. 'We need to understand how the machine reacts to a situation, the connection between systems - including the communication amongst automated vehicles.'
The test environment is also about understanding the relationship between man and machine. The team will assess and understand how the passenger will deal with situations when the vehicle makes the decision. The research will also cover the reactions of other road users to driverless cars, how they might interact on the road and to what extent they could happily co-exist.
A realistic test environment is just one part of the project, with the Venturer team ultimately aiming to have an autonomous vehicle driving the streets of Bristol and South Gloucestershire by 2016. But McCarthy is keen to establish that this is about sharing knowledge and expertise, creating a test environment that is attractive to driverless car manufacturers from around the world.
In another Innovate-funded UK project, the streets of Greenwich will soon be buzzing with the sound of zero-emission automated vehicles. Starting later this year, the Greenwich Automated Transport Environment (GATEway) trial will see self-driving shuttles taking passengers around the historic peninsula, and offer autonomous valet parking. Led by the Transport Research Laboratory (TRL), the team will test the small, pod-like vehicles, trying to understand - among other things - how people will react to them.
Proving the technology is safe and can interact with humans is the most important challenge for Professor Nick Reed, academy director at the Transport Research Laboratory and technical lead of the Gateway project. 'All of our vehicles will be designed with security in mind and safeguards will be put in place to prevent unlawful access, whether that is physical or virtual,' he stresses.
Mapping the road
But such trials are happening within strictly controlled parameters where researchers can manage the variables; once these vehicles take to the open road, the big challenge is to create a system of navigation that can adapt to a dynamic and unpredictable environment.
Self-driving cars need something a little more precise than the 2D GPS maps we're used to. One of leaders in 3D map development is Here, which was recently purchased by a consortium of German carmakers for EUR2.8bn, recognising that this technology is essential to the performance and safety of driverless vehicles. Here uses lidar, a remote-sensing system that measures distance with spinning lasers, bouncing light off the surrounding area. This approach is much more expensive than radar boxes and cameras, but it's also way more precise - essential when even a few centimetres can mean the difference between life and death.
'HD maps model road surfaces down to the number of lanes and their width, the curvature and slope of the road and surrounding signage,' says Martin Birkner, head of automotive product marketing at Here. The cars 'automatically detect more than 10,000 street sign types in various countries, as well as road markings, all of which are computed into the map so that vehicles can interpret and understand them.'
The unit is capable of creating an incredibly detailed 3D HD map with a margin of error of up to 10cm. This is ten times as accurate as 2D maps that operate with a margin of error of up to a metre.
To travel safely on the streets, the system creates a seamless exchange between the vehicle and the cloud, updating the mapped environment to the real world. 'All of this data, as well as traffic information, needs to be updated in real time to allow the vehicle to react to changes on the road in a timely manner. If there are roadworks, a change in road layout or a change in traffic density, the map must be able to inform the car,' Birkner points out.
The technology isn't just about creating an incredibly accurate mapped environment, though; it's also about providing the intelligence for the car to make the correct decision. For instance, just because a road has a speed limit of 60mph, it doesn't mean the corner should - or can - be taken at 60mph.
Then there is the data issue. The sensor and mapping technologies within driverless cars create huge amounts of data, and to make sense of it, cars will become the supercomputers of the future.
'On the current range of self-driving cars you might find four or five cameras, all over two megapixels. Each relays data to the processor over 60 times per second, all of which needs to be analysed, processed and acted upon,' says Davide Santo, product line manager for ADAS (advanced driver assistance systems) at Freescale Semiconductors, a US-based company that produces microprocessors. There's also the continuous ongoing data exchange with the cloud. In terms of numbers, it's approaching the terabyte per second level.
To manage such a load, Santo's team is working on creating a highly powered computer that can operate within a constrained environment. 'We are looking at lower processor geometry, reducing power leakage and creating a system with dynamic consumption. It's an incredible technological challenge that needs to deliver huge power in a small, low-power environment,' he says.
Creating accurate maps, and the processing power to manage them, are fundamental to the task of creating an automated car that's safe - but would it be secure?
Traditionally, the automotive industry has focused on the ISO 26262 standard established by the International Organization for Standardization, which outlines the accepted norms for protecting electrical systems in cars. The attention has been on dealing with so-called random failures, but according to Santo, the industry has been slow to tackle the systematic failures that could affect automated cars. To function effectively, a driverless car needs to be constantly networked, says Santo, which raises the possibility of hacking. To what extent can this be protected against, and what would happen if this caused an accident?
In July 2015, professional hackers Charlie Miller and Chris Valasek, director of security intelligence at IO Securities, illustrated just how dangerous a driverless car hack could be. To expose the vulnerabilities of the system, they cracked the internal code of a Jeep by wirelessly accessing its onboard computer. They then took control of it, disabling the brakes and leaving it to sink into a ditch - all of this from the comfort of Miller's house 10 miles away. The security flaw could have affected tens of thousands of vehicles across the US.
Called a zero-day exploit, the hack led to Jeep's parent company Fiat Chrysler Automobiles recalling 1.4 million vehicles in July this year.
Speaking in 2014, Hugh Boyes, the Institution of Engineering and Technology's cyber-security lead, highlighted how 98 per cent of all software applications have serious defects. To address the issue, the Department for Transport's 'Pathway to Driverless Cars' report insisted on the need for fail-safe systems that allow passengers to take manual control in the event of a driverless systems failure or cyber attack.
Taking this hybrid approach, Audi has moved away from referring to its own RS7 concept car as an automated one, preferring the term 'piloted driving'. The car recently completed a 550-mile journey from Silicon Valley to Las Vegas. The driver switches the autopilot on and then holds on to a joystick, pressing a button that enables the system to work. If removed, the car brings itself to a halt. It's networked - to a point - but the driver always has the option to take control.
It's the culmination of an estimated 15 years of development in the laboratory and the racetrack and is, for Audi at least, a natural progression of the automated parking and sensor technologies that already exist within its current production vehicles. For car manufacturers, the integration of new technologies within traditional standard designs is a comfortable medium, and reduces the risk of exposure to negative outside influence.
Security experts talk about creating a secure gateway where vehicles themselves are protected through highly secure information channels to the cloud, but where technology goes, hackers are sure to follow. Santo considers it arguably the biggest challenge for the industry: 'Investment in security is high, and the need to encrypt data is essential. The company that can safely manage and protect big data has a clear competitive advantage.'
The blame game
There are many practical challenges on the route to safe and secure driverless cars, but experts are clearly confident that these can be overcome. Google has captivated the public's attention with its driverless car trials near its headquarters in Mountain View, California. The company claims that in the six years it has been running trials, its fleet of 48 automated vehicles had only twelve minor accidents in over 1.8 million miles of automated driving. The giant is now expanding its choice of trial locations, planning to test cars in Austin, Texas, and in the UK, where its self-driving Land Rover began navigating the roads of the University of the West of England's Frenchay campus in April this year.
What is more complicated is the ethical dimension: when the car must choose between crashing into a child on the road running after a ball, or swerving to avoid her and running into a couple kissing on the pavement, or yet again, swerving in some other direction and hitting a building, injuring the passenger. In rare situations, we humans may sacrifice ourselves to save others, but can a computer do this, and if so, who would be to blame for the crash?
'Challenging liability questions need to be considered before driverless cars can be a realistic part of everyday life,' says Mark Fenning, an expert in road-traffic legal issues at Stephensons, a UK-based law firm. The programmer has to create a logical system that may, in certain cases, need to make seemingly illogical choices. It will also need to calculate the impact of those choices. As human drivers, we do this all the time subconsciously, but it remains to be seen whether this can be done by a computer.
This raises a number of conundrums the industry needs to tackle, says Fenning. 'Would the manufacturer seek to point the finger at the software provider? Would each of these parties have their own legal team, or would the insurance companies look to cover liability for both car owners and car manufacturers under one policy?'
At the moment, there are many more questions than there are answers. It's one of the reasons why both Venture and Gateway are working closely with the insurance industry to develop entirely new models of risk management. It's a case of society innovating at the same pace as the technology.
Having said all that, it is not a question of if, but when, we will see automated vehicles on our streets. The industry seems to have made up its mind, with every major car manufacturer, and a whole host of new entrants, investing billions in the technology. It's exciting, but there is a lot to consider before we take our hands off the wheel.
McCarthy points out the benefits for society in job creation and the development of increasingly important and valuable intellectual property. He's confident that adoption will be speedy, and so is Santo. 'The next five years will be much more dramatic than the last five,' Santo says. 'Humans accept changes when there are benefits to the user. We will accept it.'