Introduction
Whether you welcome self-driving cars as the new overlords of personal transport, or foresee a Terminator-style apocalypse on public roads, one thing is for sure: they're are coming soon. Volvo recently announced plans to give driverless cars to 100 punters in the Swedish city of Gothenburg and see how everyone gets on. Oh, yeah, and that's going to happen in 2017.
With that in mind, let's take a look at the state of the art in self-driving technology. Volvo has been more explicit than most manufacturers in breaking autonomous car tech down to its key components.
What are the key technologies for driverless cars? What works already, and what's the stuff that might make for an Arnie-style slaughter on our streets if they don't get it sorted?
Artificial intelligence
AI is at the heart of fears regarding automated technology of all kinds. But it's debatable whether driverless cars actually require anything that truly qualifies as artificial intelligence.
Instead, what they need are a series of robust systems developed to do specific jobs really well, such as detecting the edge of the road or managing traffic.
If there are grey areas, two obvious and related candidates present themselves: object recognition and what you might call accident prioritisation. The first involves distinguishing solid objects from (for example) a plastic bag being blown across the road. One might require evasive action, the other most certainly not.
Once you know what is around you, the next challenge is deciding what to do in the event of an unavoidable accident. What will your car decide to do, for example, if it has to choose between an accident that will kill its single adult occupant and an evasive move that saves the on-board passenger but involves ploughing into line of school children on the side of the road?
In reality, such a contrived scenario is unlikely to occur – but it still has to be accounted for. Ultimately, it's an ethical decision we can make as a society, and quite why people think such decisions are better made in a split second behind the wheel of a car is a mystery.
As for the recognition part of the problem, that isn't really an AI problem as much as an algorithm and machine learning problem. Volvo has already shown systems that are quite adept at detecting and distinguishing everything from children to moose. It's just a matter of time before the problem is thoroughly cracked.
That said, the question of who is in control remains a problem in terms of public perception. Consequently, some companies remain very cautious how they present autononous driving. Audi UK, for instance, told us: "Here at Audi, we make a point of using the term piloted driving, which means that the driver remains in control of the vehicle. The person at the steering wheel ultimately decides what the vehicle does, and can override the piloted driving assistance systems at any time."
Sensors
Software aside, sensors are probably the key enabling technology for driverless cars. More than anything, driverless cars need situational awareness. It's one thing knowing where you are in the world, quite another to have a detailed idea of what's around you and how it's all behaving.
This might lead you to think that driverless cars need sci-fi style uber-sensors. Certainly, Google's prototype driverless car is famous for its exotic LIDAR (Light Detection And Ranging) laser system – a 64-beam system to scan all around the vehicle.
However, according to most car manufacturers, existing sensor technology is actually good enough. For instance, Volvo has developed a sensor solution for its Drive Me technology that starts with a combined radar and camera, a device Volvo is already fitting to its new XC90 uber SUV.
It can read traffic signs, detect curvature in the road and pick up certain objects, such as other cars.
Next up in the Volvo system is a full 360-degree surround radar and surround vision system, including radar at each corner of the car and four cameras – two under the rear-view mirror, one in the rear bumper and one in the front grille.
Volvo does use a laser scanner with a 150-metre range, but it's nothing like as complex as the $70,000 laser Google has been using. Then there's another trifocal camera for detail object recognition, ultrasonic sensors for low-range and low-speed driving (used for years as parking sensors in many cars) and ultra long-range radar.
Each individual sensor, therefore, probably isn't anything special. But combined it's comprehensive stuff and allows the car to build a detailed picture of everything around it.
Other car companies are developing similar technologies and, like Volvo, already putting it into existing production cars. According to Audi UK, "the majority of the sensors on the Audi A7 piloted driving concept are sensors that are already in use on our series production cars."
So, the critical difference with autonomous cars isn't the sensor tech or gathering of the data. It's what the car does with that information.
High-performance vehicle positioning
High-def sensors for detailed road positioning and detecting hazards are all very well, but a driverless car also needs to know where it is in the world.
Luckily, the technology to achieve this already exists. For Volvo, high performance GPS is one part of positioning control that is enhanced by a combination of an advanced GPS, a 3-degrees of freedom accelerometer and a 3-degrees of freedom gyro. By matching the 360-degree image created by the multitude of sensors with the map image, the car builds a full picture of its position in relation to the surroundings.
By combining information from the sensors and the map, Volvo says the Drive Me car is able to choose the best course in real time, factoring in variables such as the curvature of the road, speed limit, temporary signs and other traffic.
What's more, if you combine that with cloud-connected and car-to-car data, a driverless car that knows where it is can react to hazards long before a human driver could see them, much less react to them. Overall, positioning is a problem that driverless cars are already well equipped to tackle and gives them a huge advantage over the vagaries of human driven cars.
Cloud connections
Autonomy may be a key concept for driverless cars. But connectivity to a central cloud-based system is also critical to making the most of the technology.
One of the most obvious benefits will be traffic management. Imagine a future where all cars can be directed by a central traffic management system and jams would pretty much be a thing of the past. Before traffic had a chance to really snarl up, it could be redirected and load balanced across the available road network.
Central management would also make it much easier to make temporary changes to road systems. Want to close a road to traffic for a day? Very easy if there's an element of central control to all cars.
As for how plausible all this is, well, there are a number of factors at play. Keeping all cars permanently connected is very doable with existing wireless technologies and the creation of a central control system is hardly beyond the wit of man.
What's more, you don't need all cars to be autonomous or centrally managed to enjoy many of the benefits. Studies show that by centrally managing even a relatively small proportion of traffic, the majority of traffic jams can be significantly eased.
According to Audi, cloud connectivity will also make self-driving cars cleverer over time. That's because they'll act like data collecting drones, sending what they've learned "to an IT back-end in the cloud, over a super-fast online connection from Audi connect. There, it will be processed using machine-learning algorithms and sent back to the car."
So that's not just self-driving, but self-learning, too. Nice.
On the other hand, none of this will come cheap and it will likely require governments to enact legislation. And that usually means time consuming bun fights at best.
Then there's the spectre of hacking. Hacking an individual driverless car throws up some scary scenarios. But if the bad guys got control of the entire system, that really is Doomsday stuff.
Back-up systems
Failure is not an option, so it's good to hear car makers are planning to fit back-up systems. According to Dr Erik Coelingh, Technical Specialist at Volvo Cars, Volvo is taking a similar approach to the aircraft industry.
"Our fail-operational architecture includes backup systems that will ensure that Autopilot will continue to function safely also if an element of the system were to become disabled," Coelingh says.
Volvos says, for example, that the probability of brake system failure is very small. But a self-driving vehicle needs a second independent system to brake the vehicle to a stop, as it is unlikely that the driver will be prepared to press the brake pedal.
That final point is key. Very likely, as soon as people become accustomed to driverless cars, they'll also become entirely dependent upon them. In other words, there's slim to no chance of a human occupant being ready and able to intervene should the autonomous systems suddenly go offline.
Of course, all of this only applies to a back-up system that's completely reliable. If something has brought the primary pilot down, you have to worry if the same thing would apply to any secondary system.
What's more, the need for multiple back-up system is only going to add to cost and complexity. Given widespread doubts in the population at large over the suitability of driverless cars for public roads, the things will likely need to be as affordable as possible to encourage widespread adoption. Aircraft levels of system redundancy and low cost do not make for happy partners.
Autonomous road trains
Whatever you think about the likely future of driverless cars, it won't quite happen overnight. Exactly how we get from today's messy highway of humans to the fully robocar future isn't totally clear. But one of the stepping stones is very likely autonomous road trains.
What we're talking about is a kind of hybrid between human-driven and fully automonous cars. One version of autonomous road trains is a technology called the SARTRE Project and supported by Volvo.
SARTRE involves a sort of follow-the-leader approach. A professionally human-driven vehicle leads the way while other cars fall in behind under human control before switching to autonomous control and riding along like virtual cars of a rail train. It's an approach that works best on multi-lane motorways.
The benefits are several. Most obviously, humans can avoid the tedium of motorway driving and get something productive done or simply rest.
Similarly, road trains also allow cars to travel much more closely together. That's good for both slipstreaming to save fuel and in terms of more efficient use of road space. Fully autonomous trains with no human-driven lead car and the ability for autonomous vehicles to join and leave the train dynamically are the next step on.
Autonomous road trains can also help smooth out the dangerous concertina effect which sees human drivers overreact to cars ahead slowing or braking. In extreme cases, a minor change of speed by a single vehicle can create a ripple effect that ends miles away with stationary traffic or even accidents.
Volvo and the SARTRE Project demonstrated autonomous car trains operating successfully over a 100 mile-plus stretch of Spanish motorway as long ago as 2012. So the technology is already quite mature. As with many driverless technologies, it's just public acceptance and legislation that are holding back the technology.