I have news for the doubters: you're too late. Self-driving cars aren't a future menace, or even a future product. They're here today, in commercial production, and they will inundate the market as surely as the internal combustion engine displaced the horse and carriage.
What CES demonstrated was not that car manufacturers have ambitious, blue-sky plans for futuristic technology, but how fast they are moving towards mainstream adoption.
There are two parts to that. The first is the reality that autonomous robotic vehicles are a stable technology which is improving rapidly to overcome technical constraints, and the second is the speed with which obstacles to consumer deployment of the tech are falling.
When Tesla's Elon Musk predicted a self-driving Tesla would be able to drive across the USA in 2018, he may in fact have been erring on the conservative side.
Actually, humans are pretty terrible drivers. In practice, driver error causes 94% of road accidents. And globally, there's a road death every 25 seconds. When you tally up the factors that cause road accidents, it's surprising that humans think of themselves as competent at all. Psychologists call this the overconfidence effect, and it's been studied extensively. It's often referred to as the Dunning-Kruger Effect, after research which showed many people overestimate their capabilities in any field, but it's been studied specifically in driving: in the 1980s, Ola Svensson published research in the US showing that most motorists consider themselves to be above average and lower risk than average.
An onboard computer never gets bored, never gets distracted, and never looks away to change the radio station, send an SMS, put on makeup, or read a newspaper. It doesn't get angry, or reckless, or tired, or drunk.
Reducing driver error, and ultimately taking the driver out of the equation entirely, will save thousands of lives every year. And most traffic accidents are not fatal: the financial saving would be enormous too, which is good news for everyone who isn't a motor insurance salesman, or a professional driver.
Full robotic vehicle autonomy is probably some way in the future, but the road to self-driving is clearly lit. It starts in simple, predictable environments where robots can operate vehicles with a minimum of anomalies. Gradually, their capabilities to detect unexpected conditions grows (external anomalies like collision avoidance and internal anomalies like mechanical failure) and the application broadens.
In fact, autonomous vehicles have been a reality in many environments for years. Mines operate self-driving mining vehicles of many forms, from fully automated to varying degrees of remote control. Autonomous mobile robotics are increasingly popular in warehouses, where the advantages are not only efficiency and reliability but also ongoing BI – a human low-wage warehouse runner doesn't have time to stop for breath, never mind optimising shelf layout on the fly, where integrated warehouse management systems can constantly tune the layout to ensure popular items are optimally placed.
Humans in automated e-commerce warehouses are usually relegated to the packing endpoint, where it is still more efficient to allow a dextrous human to jigsaw items into a carton for shipping. From a process optimisation point of view, the key realisation is that humans are expensive components being optimised out of the system: the goal is to reduce the anomalies that automation can't handle.
Self-driving is just an iteration of the same discipline. In the case of driverless cars, that means steadily expanding the range of conditions that a vehicle can handle until it can reliably manage increasingly complex tasks – parallel parking, driving on freeways, and eventually complete trip automation. As that happens, the upside accrues – fewer accidents, fewer deaths, lower costs.
One of the areas where self-driving cars are already on the verge of outpacing human drivers is raceways. That's a pure performance, highly constrained environment where the combination of lightning reflexes, zero self-preservation, and raw data crunching will certainly surpass human capability, just as it did on the chessboard and the TV gameshow. That's a point we may have already crossed, with Audi's self-driving RS7 race car.
Other technological driver aids – anti-lock braking, electronic stability control, lane-assist, self-parking and many more – became rapidly commoditised. At and around CES, we've seen the signs of that coming to robotic vehicles. NVidia's in-car compute module is aimed at the high-speed data processing required for driverless cars. The complicated spinning radar units on top of early prototypes are giving way to more compact and durable solid-state LIDAR (laser range finding).
And, tellingly, existing vehicle platforms are built with driverless capabilities in mind. Tesla eschewed additional LIDAR entirely, using nothing more than existing onboard cameras and distance sensors on its cars to deploy, as a software update, self-driving capabilities now actively deployed on the road.
Driving is hugely complex, not in terms of vehicle control but in terms of the infinitely varied conditions that a driver may experience, like erratic fellow motorists, adverse weather, verbal instructions from a police officer.
Those are challenging, to be sure. What detractors don't see is just how fast they are being overcome. A year ago, Google's self-driving cars were initially unable to cope with rain or snow – now Ford has demonstrated driverless vehicles capable of handling snow. Google's telemetry data shows the rapid reduction in anomalies forcing drivers to intervene, from one every 1 263km in 2014, to one every 8 500km now.
Chris Urmson, director of Google's self-driving car project, posted an excellent breakdown of the lessons learned by their research, particularly in identifying the areas where self-driving vehicles struggle to handle anomalies, both sudden surprises and driving conditions which defy sensor analysis.
Most of those anomalies happen in high-density areas – the open road will likely see driverless vehicles permitted first – truck companies like Volvo and Mercedes-Benz have vehicles nearly ready to go, aiming at allowing humans to do the tricky last stretch of a drive, but leaving the boring open road to the machine. One possibility is that long-haul trucks may drive themselves to depots, where human pilots will (for the medium term) complete the journey.
Note that the key hurdle there for trucking is not technological, but regulatory. And many countries are actively passing regulations and setting performance benchmarks for self-driving vehicles. It's likely we'll see heavy lobbying from unions, insurers and other affected groups, but lawmakers globally are taking a proactive stance, encouraging innovation without compromising safety.
Lastly, a common resistance to driverless cars is that driving is simply fun and people won't want to give it up. We like driving: we like road trips and racing too much to quit. But there's no suggesting that autonomous vehicles would result in a ban on human-controlled driving. There may be zones where autonomy is required – congestion-controlled city centres to start with, most likely – but even in the final stages of automation, leisure driving will remain, just as horse-riding became a leisure activity and continues to thrive in that redefined role.
There are many ways in which self-driving will roll out, but it will look something like this: driverless cars happen, then become demonstrably safer and cheaper, until manual vehicles are deprecated, then penalised (taxed, insurance caps, curtailed access, etc), and eventually dwindle in common use.
That evolution will happen gradually, and how it will play out is very much up in the air, but make no mistake: the development from both ends has now met in the middle. We're halfway down the path, not starting along it.
Our comments policy does not allow anonymous postings. Read the policy here