As a follow-up to my post about driverless cars, here are a more links on the topic:
- The ethics of driverless cars.
- Should driverless cars have to obey speed limits?
- Google exec: driverless cars coming ‘even faster than I imagined’.
- GM says almost-driverless car coming by 2020.
- People say would rather buy a driverless car from Google than GM.
- This guy says a Google driverless car is nonsense.
I wish Google would do a “heads up” test.
Buy some abandoned town or suburb.
Put about 100 Google cars in there, and program them for daily routines.
Let them go completely driverless (have a remote control to deactivate them).
Put up a bunch of cardboard cutouts to represent people, kids, on the side walk…
Then see if they can behave without going out of control.
I think driverless cars are one of those things where the 90% case might be okay, but the last 10% is very very hard to solve. Driving around sunny mountain view during the day? Fine. Driving around mountain view at night? fine.
Try driving around Capitol Hill in the snow? Not going to work.
https://seattletransitblog.wpcomstaging.com/2012/01/20/in-defense-of-seattle/
Woody ex-urban/mountain roads in the fog?
Those are the use cases that will take years to sort out.
Well, exactly, that is why there should be continuous real time testing in a controlled environment. Let this test go for a year, in a place with varied climate to see what happens.
As it is they just keep injecting these cars into regular society, and yes, the Google video proves that you can go from a driveway, to a Taco Bell and back again without touching the wheel, so at some point someone is going to just going to fall asleep and hope the car takes him home from a bar. Will it? And what about when multiple robocars encounter each other?
An all day, real time test with actual cars would help us know if these are ready (or even more than ready!)
To be fair, practically no humans can drive when in snows in Seattle, and computers are perfectly capable of handling a vehicle at its limits. They certainly have the potential to do this much better than human drivers; I’d bet they’re already better at this than the average yahoo. They’re also better than people at seeing things, including in fog and snow, because they have sensors designed to do that, positioned in better places than the driver’s eyes, all looking everywhere at once. This is all the standard stuff machines do better than people do.
When two robocars encounter eachother they’ll avoid colliding by following the rules of traffic independently, just like regular drivers do. Maybe they’ll give eachother a little wave as they pass, like bus drivers do to eachother. Or maybe they’ll join together to KILL ALL HUMANS. *shrug*
Today’s prototype self-driving cars quite readily give up and hand control back to humans when they’re not sure they’re doing the right thing. I the hardest problems will involve reading comprehension — how to interpret temporary, ad-hoc signs and directions in construction zones, how to follow someone directing traffic, how to recognize closure of a route the map data says should be available, which road a do-not-enter sign applies to. How to find a legal, sensible place to pull off the road and stop if necessary. How to determine which parking lot belongs to which business, or which driveway to which house, in the absence of full data. Of course, humans are hardly perfect at this — I’ve seen a Mini go through the diverter east of 15th on the Ballard Greenway; a minivan pull off of Green Lake Way then over the curb, onto the grass, and onto the gravel trail; cars on the Burke-Gilman; people driving the wrong way on all kinds of streets. Some of that’s even intentional or malicious, and self-driving cars probably won’t intentionally break the law or act maliciously. And I don’t think self-driving cars will be hopelessly stupid, but they may make some weird mistakes. They’ll probably be reasonably good at avoiding collisions while doing so, however.
It would take a super-robocar to drive when there’s pockets of ice.
Sure, robocars won’t be able to drive on ice either, but making a robocar that knows that is important.
It’s easy to make a robocar that drives in sunshine and everything is fine. It’s all the error cases and edge conditions that are harder.
Sometimes, crossing streets or driveways as a pedestrian involves making eye contact with drivers to make sure they see you. This will get significantly more complicated when you have a mixture of half human-driven cars, half robot-driven cars on the road.
Do you think it makes sense to require cars to have some sort of indicator light so that people outside the vehicle know that it is being driven by a computer?
Exactly the problem. The last 10% is *never* going to be solved.
Robocars are pretty good at detecting and not running into things.
When you’re walking (or biking) toward a driveway the worst moment is before you can make eye contact, when you see the hood poking out but have no idea how far the driver is going to go. Robocars have sensors better positioned than the human head — they don’t have to lean out that far to see you, and once they see you moving across they’ll stop and yield as the law requires. When drivers pulling out of driveways come to a full stop at the sidewalk the need to make eye contact is not so great.
Navigating through traffic is more than just avoiding collisions. It is all about interacting with other drivers and responding to the unexpected. I’d say a Google car that adapts to any situation and gives an appropriate response in 90% of cases appropriately is way too high.
And what is it about cost? I have heard several times that car prices shouldn’t be appreciably higher, be we are talking about a robot technology decades in the making. High r&d, sophisticated software. How can it not translate into high cost?
I have the perfect location to do these tests. It’s got lots of abandoned neighborhoods with remarkably good street infrastructure. And they kind of know their cars.
It’s called Detroit.
Pay them something to do this. They could use the cash.
It was called the DARPA Urban Challenge, held 6 years ago. Multiple robocars and even more human drivers all mixed together in an urban roadscape. There were no problems. This was SIX YEARS AGO. It really is a solved problem. As soon as the legal problems are solved and the price drops enough, it’s happening.
I see a real near-term potential to use slow-moving driverless cars for rail station access — the so-called “last mile” problem. Vehicles could be programmed to serve specific areas, and would provide a great test case situation. Imagine coming home from work, hopping in a driverless car at the rail station and being dropped out in front of your home, with the car driving back to the station. in the mornings, the driverless car would leave the rail station empty, drive by so that you can get in, then return to the station. Lots of feeder routes could be eliminated, and yet the improved access would really help build ridership on trunk lines. Finally, I see this as an eventual private sector strategy for lots of local non-work trip making, so either transit operators figure out how to support it now or lose out on a great opportunity to improve their systemwide farebox recovery.
You could have a major station, with lots of apartment complexes surrounding in.
On the interior you would use small vehicles, maybe even just golf carts, that go max 20 mph.
On the outside of it, you would have regular sized cars, which enter directly onto highways.
In this way, people would never interact with regular cars, only the golf carts.
This technique could eventually be used to greatly reduce the amount of parking needed at suburban transit stations, along lots of existing parking lots to be redeveloped to better use.
At any rate, it certainly makes present-day investments in $50,000-per-stall parking garages look incredibly silly.
The parking garages could also be storage places for publicly available autonomous vehicles. In a transitional period, maybe the underground garage spaces are used for queuing of autonomous vehicles, while above-ground spaces ‘with a view’ are reserved to human piloted vehicles.
I had high hopes for the Ethics link, and boy did it fail to live up to them. Lazy, half-baked application of neat-o concepts from Philosophy 101. It felt like 2:00 AM, in the dorm room, with a bong, all over again.
Weed’s legal now, so expect a lot of that.
How do driverless cars respond to flaggers at a construction site? to police directing traffic at an accident scene? Can driverless cars follow a detour around a road closure? Do driverless cars know to stop at non-functioning traffic signals?
Do they know to pull over for emergency vehicles?
Follow directions for loading onto a ferry?
I think people are focused too much on the cool technology and not enough on the regulatory environment. Automated driving, if it can solve the last 10% problem, is coming. But how it affects our larger lives is something that we have a lot of control over. It is up to humans to decide whether automated cars get priority over people using other modes, or whether we prioritize moving people and regulate automated cars to support that goal.
With minimal regulation or regulations that assume automated driving as a norm, I would expect a worsening of most of the bad aspects of non-automated cars. We’d get more sprawl (since people will be able to work or sleep while their car drives longer distances), more useless trips (from sprawl, and from cars circulating so they don’t have to park) and thus a new equilibrium with high congestion and more traffic, and an urban environment that is hostile to pedestrians and cars. Except by getting worse, nothing fundamental would change.
With good regulation, automation could enhance connections to transit, allow free flow of traffic while still leaving room for other modes, free up parking lot space for more profitable development, and lead to a renaissance in urban living. But we need to make that happen.
One thing to consider is that automated driving is unlikely to be any cheaper than current cars, so many people in cities will still choose not to drive for economic reasons alone. Arguably, auto-taxis might be more expensive than driven cabs, or at least not cheaper. There are still geometric issues with everyone car-commuting at the same time; automated spacing can put more cars in lanes but not enough to take away the advantage of buses and trains on busy routes. I really think that insurance/liability issues will in fact make non-automated driving too expensive over time except for rich people, mostly hobbyists, and in time having a car of one’s own will be seen as an inconvenience when it’s easier just to call up an auto-car. Unless great pains are taken to prioritize automated cars throughout our built environment, I don’t think they will replace transit.
People buy cars because the like them and mostly like driving. Except for situations where traffic sucks, most folks would rather be in the driver seat. You can go to the Auto Show right now and see what I mean. The richer folks I know and know of, drive themselves and spend their money on high end driver’s cars, not chauffeur driven limos.
The future of automation is mostly in features that are being added to cars now. To make the car more idiot proof, but keep the driver in control. At least thinking they are.
Depends on the person. Many do buy cars because they like driving. Many others buy cars because they feel they have to.
There are some great questions here – but please be aware that this technology has been under development by Google for over 4 years, and by the automakers for considerably longer. The developers are very aware of all of these questions and have long lists where they are slowly but surely ticking off the problems as they solve them.
Google know that they will never get that final 10%, but they are aiming for something like the last 9.99999% – which will make their vehicles significantly better than the average human driver. As I heard Ron Medford say at the TRB Workshop on Road Vehicle Automation at Stanford in July, Google have been putting in the hard work and aspire to produce a fully self-driving vehicle. They are already in beta testing mode on public highways and have achieved 96,000 miles without safety critical human intervention. They have a long way to go – but please don’t count them out without first doing a deeper review of what is happening in this space.
To really understand how far development has come and to understand the new paradigm that fully automated vehicles bring, then may I suggest that you try a number of forums that I am involved with such as DriverlessCarHQ on FB, or the sub-Reddit /r/SelfDrivingCars/
When Google say that they aspire to have fully self-driving car technology in public hands by 2017 they are not joking. It may not be ready for extreme weather on first release, but with advances in sensor technology, falling costs, and the way that the artificial intelligence is programmed to learn, then it will be a very rapid progression to widespread usability. The business case is compelling in my opinion – as I have argued in my blog on a post titled ‘The Inevitable Rise of Autonomous Vehicle Fleets’.