The Mercedes-Benz S-Class now has a model that can drive around totally unaided 鈥� as long as you are in one specific car park in Germany. It is an impressive achievement, but how far have we really come since New 杏吧原创 proclaimed 15 years ago that a fully driverless car 鈥�may not be far off鈥�? Cars have accumulated automatic driver aids, but a fully autonomous private car that can operate wherever you want to go is no more available today than it was in 2007.
Perhaps the most ambitious driverless experiment so far is from US start-up Cruise, owned by General Motors, which began offering a taxi service in San Francisco this year. Customers are the only occupants of the car and no employee sits ready to grab the wheel. While these cars have randomly stopped and blocked traffic, and had a , things have otherwise gone relatively smoothly.
City streets are by far the most challenging task for a driving AI, making motorway miles seem relatively straightforward and proving trickier even than winding country roads. As a result, Cruise initially only offered rides at night, when the streets were quieter, and limited operation to San Francisco, where it had spent years mapping the city.
Advertisement
It is, nonetheless, proof that driverless cars are possible, and it may be a vital step towards convincing customers, legislators and manufacturers that the technology鈥檚 time may actually be near.
There are varying levels of car automation though, : level 0 is a conventional car, while levels 1, 2 and 3 include increasing amounts of driver aids. Level 4 is full automation in 鈥渓imited conditions鈥�, and level 5 is total autonomy, all the time. At that final level, manufacturers could even do away with the steering wheel and pedals.
While Cruise certainly meets level 4 鈥� as do 鈥� there have been no cars on general sale with this capability.
鈥淕etting it to work everywhere, and getting it to work in all weather conditions, that鈥檚 where a lot of the challenges come in,鈥� says at Durham University, UK.
Taxi services can sidestep these problems by controlling when and where cars operate, suspending operations in rain or snow, and shutting them down entirely for fixes if problems are spotted 鈥� all without angering owners.
Breckon says creating AI software to navigate on US city streets, or US streets generally, which are wide and built for cars, is a very different challenge to developing software for narrow European streets, or bustling Asian roads. And because making different models for every market is prohibitively expensive, the software needs to be capable of all of it, in all weathers. Anything less could mean cars struggling to cope and handing control back to a disgruntled customer who has just spent a lot of money on a 鈥渟elf-driving鈥� car.
鈥淭he automotive industry is very aware that if they come to market too soon with something that鈥檚 not a positive experience then, potentially, they鈥檙e going to lose ground in the future,鈥� says Breckon. 鈥淭o be the first to market, I think there鈥檚 a lot of risk.鈥�
Add to this the uncertain legal issues around who is at fault if a driverless car kills someone or damages property and unknown market demand, and launching an autonomous car is a huge financial risk as well as a thorny technical challenge.
There is also the problem that the sensors and computers needed for autonomous driving drain the car鈥檚 battery, says Breckon, which can require additional battery packs, adding weight, reducing performance and range, and increasing cost. These are all more of an issue with a private car than a city-based taxi.
Private car automation isn鈥檛 being entirely neglected, however, as shown by Mercedes-Benz鈥檚 announcement this month that its S-Class model will now offer level 4 automation, albeit only in one car park at Stuttgart airport, where it can park itself with help from special sensors dotted around the area. That limited functionality is the first such approval of level 4 automation for public use by Germany鈥檚 Federal Motor Transport Authority. Mercedes-Benz declined a request for interview about the prospect of a wider trial.
Chinese start-up Jidu 鈥� a partnership between Baidu, often called the Chinese Google, and car-maker Geely, which owns Volvo and Lotus 鈥� also that it would launch a level 4 car called the Robo-01 this year. That deadline has since been extended to next year, though, and a recent press launch . Neither Geely nor Baidu responded to a request for comment.
Other companies have faced criticism over claims. Tesla named its driver-assist software 鈥淎utopilot鈥� despite its meeting only level 2 of the SAE categories. It is now facing legal challenges from US state regulators and private customers who claim that the company鈥檚 use of phrases like 鈥渟elf-driving鈥� in its marketing material . Tesla hasn鈥檛 responded to a request for comment.
Given all these hurdles, general sales of a level 4 car are likely to still be more than five years away, and companies may never actually sell them, says Breckon. 鈥淭here鈥檚 an analogy with autopilots in planes, and a lot of small, private planes that people own individually don鈥檛 contain autopilots,鈥� he says.
Another option is that firms only lease automated cars. This would allow manufacturers to add clauses banning modification and retaining control of what the car can do, where and when, in a similar way to running a taxi service.
鈥淚f, suddenly, there鈥檚 a spate of pedestrian deaths, they can actually just turn [autonomous driving] off,鈥� says Breckon.
