The fact that small boys still make “vroom vroom” sounds with their cars speaks volumes about the (lack of) progress in how cars are powered. My cars vroomed, as did my father’s and so did his father’s. Lightening McQueen, the hero of Disney’s critically derided yet widely loved ‘Cars’ franchise, is most definitely packing a big V8, even if his latest nemesis Jackson Storm has some kind of hybrid system.
I’ve previously written about electric cars, and even if my son’s first car isn’t all electric (it’s probably rolling off a production line about now) I’d wager he’ll own a few zero emission cars before he hits middle age. A bigger question though is whether he’ll want to, or even need to, get his driving licence when he hits the magic age of 17. I’m of course talking about self-driving cars.
To dream the autonomous dream
Seba’s paper is perhaps a good example of a headline-grabbing claim being used to draw attention to some less sensational points. The first is that the supporting infrastructures for petrol and diesel cars could collapse like a pack of cards if there’s a meaningful shift to electric propulsions: the costs of fuel distribution, spare parts, etc would be split over a shrinking pool of vehicles and investment would quite simply move on.
The second is that the current ownership model probably wouldn’t work with self-driving cars, and we’d see a move to ‘vehicles as a service’ (robot taxis). Even if people still bought their own private car they’d be very tempted to let it earn money as a Johnny Cab when they weren’t using it, rather than just having it sit outside of their house.
At the other end of the spectrum sits Berkley’s Steven Shladover, who thinks it unlikely that I’ll even get a lift from a self-driving hearse to my funeral (2075 is his prediction). Meanwhile vehicle manufactures have given their own predictions for when autonomous cars will hit the market, ranging from 2030 (Hyundai) to 2017 (Tesla, predictably).
So who’s right? Well to put an answer in context it’s worth telling the story of self-driving cars, a twin track tale of huge advancements in technology and even huger advancements in pay packets.
Eyes on the road
The following year the challenge resumed, with a new, tougher 132 mile course. This time round the robots were radically improved, with 5 vehicles completing the course and the winner, Stanford University’s ‘Stanley’, scooping the million dollar prize. The team behind Stanley soon went off to work for Google, who became one of the pioneers in self-driving technology.
Fast forwarding to the present day sees Google joined at the self-driving party by a host of other players. Some, such as the major vehicle manufactures, are working on complete vehicles, whilst a plethora of smaller start ups are developing technology for others to integrate into their cars (or even add on self-driving 'kits'). Ride hailing companies like Uber have also joined the party, seeing big profits in removing the driver from their cab services.
All of this work has seen the ‘eyes’ of self-driving cars well developed and falling in price. Most companies are using a combination of cameras, radar, LIDAR (laser radar) and ultrasonic detectors to allow the car to see the world around it, and combine this with detailed maps that let the car know where it is. With the hardware sorted the challenge is now to perfect the software that actually tells the car how to drive. This is the tricky bit.
Once a car is able to drive itself round a track, most companies are choosing to test their cars on public roads, albeit with a driver ready to take over if the computer makes a potentially dangerous mistake or ‘disengages’. Google’s ‘Waymo’ division is probably the furthest advanced here, logging nearly 2 million miles on the roads near their headquarters in Southern California.
During 2016 Google reported their cars were only disengaging from computer control on average once every 5,000 miles. This is impressive, but remember that the vast majority of testing has taken place in a sunny, suburban location – plonk the cars down on a rainy day in Bristol and they may not be so successful.
An alternative, rather intriguing approach, is being used by the electric car manufacture Tesla. They’re building all of their production cars with the sensors necessary for self-driving operation. The primary use for these is their ‘Autopilot’ system, which gives the car limited self-driving abilities (under supervision) along motorways and major roads. However, these sensors also constantly send data back to the Tesla mothership, giving them a huge amount of data on how their cars ‘see’ the road. This can then be used to train their self-driving software, and improve the Autopilot system via over the air updates.
A googleplex of cash
Ironically rather than keeping them loyal these huge pay packets have given many top engineers the financial security they need to strike out on their own in pursuit of even bigger paydays. Rather than develop their own self-driving technology many established companies are choosing to buy it in, with enormous sums being paid for small start-up companies.
The poster child for this trend is Otto, founded in January 2016 by Anthony Levandowski and purchased later that year by Uber for $680 million. Levandowski had formerly been a key player with Google, leaving after receiving an unbelievable $120 million bonus.
Engineers are obviously keen to keep the gravy train running, and the companies that employ them talk up progress as they don’t want to appear inferior to their rivals. Even staid, old school car companies are pouring money into self-driving technology and firing out optimistic statements, as they simply cannot risk being shoved aside by new entrants if the technology takes off.
Any meaningful assessment though seems to suggest that the technology can drive a car in 'most' driving conditions, as in it the leading players could field a car that would drive itself in (say) 95% of driving situations. The trouble is that a car that only works 95% of the time won’t make much of a taxi, and the last 5% is very hard. This is the realm of bad weather, knackered roads and stupid humans who don’t play by a robot's rules.
Take my commute home from work. First off, if I’m driving it’s usually raining - I bike it otherwise. Exiting my workplace brings you to a roundabout clogged with traffic joining from the right. Getting out therefore involves gently pushing into the traffic and hoping some kind soul will wave you in. After this it’s 5 miles of nice wide stop-start roads, followed by half a mile where the road is made single file by parked cars – again there’s lots of flashing and waving with other drivers to proceed.
I’ve no doubt that technology such as Google’s could drive the vast majority of that journey better than I could, if it’s not raining. At the very least a machine always fresh, whilst I’m tired after work. However, the small sections that involve pushing, flashing and waving will present more problems. A flash of the headlights can mean anything from ‘go on mate’ to ‘I’m very angry with you’, with interpretation depending on context, culture and, above all, experience of being a human.
I should also say that parking in front of my house involves leaving one wheel on double yellow lines (the wardens don’t seem to mind). I seriously doubt a self-driving car would be happy jutting out onto yellow lines, at least not without the owner signing a legal waiver. Rules are similarly bent in many UK urban parking situations.
Assuming this final 5% continues to be a problem then there’s a few possible solutions. The first is simply to keep a driver in the car who can take over if autonomous driving runs into a problem. This is what’s known as ‘level 3’ in the 5 levels of autonomous driving, and unfortunately one of the most problematic. Testing has found drivers simply switch off if the car is driving itself, meaning that if they need to take over in an emergency they may not be able to react in time and even make the situation worse. Requiring a licenced driver to be present in the car also destroys many of the benefits of a self-driving car.
Another option is for the car to ‘phone home’ when it encounters a problem, allowing a human to take control of the car and resolve the problem. Earlier this year the Japanese car company Nissan advocated this type of solution. This removes the need for a driver in the car, but has some fairly obvious shortcomings. The remote driver wouldn’t be able to connect and react in time to handle any safety related issue. It would rely on the availability of high-speed mobile broadband, which in the UK at least has far from universal coverage. Finally, human drivers would need to be content to sit behind a stranded vehicle whilst it phoned home and sat-in a call queue.
The final option is to remake the roads for robots. Road markings and signs would be replaced with machine friendly versions. Traffic police, workmen and any remaining human drivers would be given equipment to allow them to communicate with self-driving cars. Cyclists, horse riders and pedestrians would be barred to segregated paths. This would provide a controlled environment for self-driving cars to operate, but you have to questions whether such a huge makeover of public space is either feasible or desirable.
The kind of questions we should be asking then are: what kind of level of change and disruption on the roads are we willing to put up with in order to roll out self-driving technology, and who will pay for any infrastructure changes deemed necessary for these cars to operate (the public or the tech companies)? And how will we handle the inevitable media backlash when a self-driving car kills its occupants or runs someone over?
These kind of questions seem to be lost in the current media hype, with the most frequently raised issues being moral judgements such as should a robot should run over a group of nuns or drive into a wall and kill its occupants if there were no other options.
So can I have my robot car now?
Although they’re unlikely to be able to handle every driving situation in the near future, even autonomous vehicles that only operate in certain conditions or environments (‘Level 4’ vehicles) are very useful. Such systems are already here offering driver assistance, for example parking the car or driving from a parking space to your front door to pick you up. More is to come, and the economic impacts could be huge.
Motorways - where driving conditions are predictable and pedestrians and cyclists excluded - are perhaps the most obvious place that such technology could be deployed soon. For the car driver the benefits are obvious – drive onto the motorway in Birmingham, select autodrive, put your seat back and wake up in Scotland.
For larger vehicles the impact could be even more profound: lorry and coach drivers would only be needed for a journey’s local legs, jumping out of the cab of outbound vehicles at the edge of town to take over inbound vehicles for the last mile. Such a scenario is bad news for lorry drivers, but could reduce costs and, by making freight distribution a 2-step process, help remove big, dirty vehicles from our city centers.
Ironically by making long distance car travel easier, faster and more relaxing self-driving cars may be more of a threat to train drivers than taxi drivers. This wouldn’t go down well with my son, as his love of cars is second only to Thomas the Tank Engine. What he has no love for though is long motorway journeys, but I dare say that by the time he has children of his own families will be chatting, playing and sleeping their way down the M1 rather than keeping all eyes on the road.