Driverless cars banished to fake Michigan 'town' until they learn to read Opened on Monday, the University of Michigan's Mcity is the world's first controlled environment to test driverless cars, designed and developed by the university's "Mobility Transformation Center" (MTC) along with the Michigan Department of Transportation. The driverless utopia is a 32-acre simulated urban/suburban …
Self driving cars are already being tested on real roads; their safety record is pretty good:
https://en.wikipedia.org/wiki/Google_driverless_car#Traffic_accidents
Self driving cars are now pretty much inevitable, they will be safer than human drivers because once they are developed to an acceptable level of risk then all of them will be capable of driving to the same standard. Humans have varying levels of driving competence, ranging from those who kill themselves, or someone else, within a few days of getting a license to those who drive for fifty years without having an accident.
The USA has around 30,000 fatalities a year, so the humans are already way behind on safety; I can easily imagine a time in the future when insurance companies will prefer self-drive cars over human control, and that's what will force the issue. Already a young driver in the UK will expect to pay £2000 for their first year of insurance after receiving full license; if they can get mobile in a self drive car without the cost of learning to drive or a huge insurance bill, then why would they go to the trouble?
The same argument may also apply to hire cars; why would they take on the risk of people with variable driving skills, possibly in an unfamiliar vehicle with the controls on the wrong side, when they can simple rent out a self drive vehicle for less money? I know they make a lot of money on the insurance, but those who are renting self drive cars will undercut them in the market.
Humans driving cars could soon become a leisure activity for those who are prepared to pay for the additional risk involved, and it may even become illegal to drive oneself in urban areas where the risk of an accident involving a third party is much higher.
I can only imagine how much fun it must be to be on the dev team of these kinds of projects....
Ok developer number 1, your project car for todays exercise is a Shelby Cobra, your only constraint is that it must remain on the track at all times.
Ok developer number 2, your project car for todays exercise is a Double Decker bus, your only constraint is that it must remain on the track at all times.
Ok developer number 3, your project car for todays exercise is a Reliant Robin, your only constraint is that it must remain on the track at all times.
On your marks, set, go ................................Maria Karts comes to life.....( without the drivers of course)
"Why do people say "Robin Reliant", but not "Astra Vauxhall" or "Escort Ford" ?"
If you had ever driven/been in one you would know why.. My grandfather used to own one......which he then gave to my father...... which meant that myself and my brother would rather walk to school than get driven there.....
It's a kind of embarrassing car...it's only got 3 wheels ( 2 at rear , 1 in front) poor quality, absolutely crap handling, very negative street cred, you really don't want to be in a crash with one of these things...
A BondBug on the other hand is extremely cool...
I think JulieM's point isn't wondering why people pick on the Robin, but why people often get the name the wrong way round.
It isn't a 'Robin Reliant', but a 'Reliant Robin'. Reliant is the car maker (who also made, amongst others, the Scimitar, Sabre and of course the Kitten - a 4 wheel version of the Robin). For some reason people do seem to frequently call it a 'Robin Reliant' though but I have no idea why - can anyone shed light on it? As JulieM notes, they wouldn't call a Ford Escort an Escort Ford.
Reliant Robin: Comedic deathtraps to be sure but they were the car of necessity for old rockers 'cause you could drive them on a bike licence - instant family transport, no car licence required. Unfortunately guys I knew who had one tended to drive with excessive enthusiasm, being used to bikes. They'd consider it a mark of skill to be able to get it on two wheels in a corner without tipping it. Let's see a robot do that (just to bring it back on topic)
BTW Robin Reliant/Reliant Robin, Series (n) Land Rover as another instance of model before manufacturer in common usage?
Didn't really answer the point of the question... the point was that the car is made by reliant and the model is a robin... so it should be Reliant Robin. For some reason the mistake persists and because we are English and English is a wonderful language we don't care enough to avoid the problem
On to your other point... the Bond Biug was a pretty looking car but exactly like the Robin it falls over because the design is fundamentally wrong - the laws of physics say any car with this configuration will fall over at the first decent attempt to go round a corner.
To remind folk the 3 wheeler was (and actually still would be) a good idea because of the strange UK taxation rules that make it cheaper to tax.
Strangely I also don't understand the design for more practical reasons - the differential that is required is far from efficient or cheap to manufacture/source so why use it? The single steering wheel is also a completely custom creation. The approach taken by Morgan with their 3 wheeler (even the updated version of the original) and the German 3 wheelers...
Basically :
take the front suspension of any small rear wheel drive family saloon for the front two wheels
take a motorbike engine and transmission (maybe use a propshaft thats a bit longer than normal)
And then you have all you need... pretty near straight from a parts bin and perfectly stable.
As I think I have mentioned before on these forae, the Bond Bug was actually an amazing vehicle. The top model was capable of 115mph (for internal consumption only - never let insurance companies know!), and the factory test drive demonstrated to me that it could do handbrake turns, doughnuts, and 3-wheel slides on corners - all because the COG was barely above the wheel hubs.
Because it should be 'Reliant Robin' ....
while I'm here I fail to notice anything other than level roads, no hills in driverless car world? How will they handle blind summits with parked cars either side or angled junctions at the bottom of a hill? How about getting confused on an overpass and trying to turn onto the road below? ouch!
Maybe however these won't be able to use the 'Sorry I mate I didn't see you' (SMIDSY) method of 'apologising' for maiming cyclists and motorcyclists....
I don't see a version of a one-lane country road with the occasional laybys and small herds of sheep wandering slowly across. Nor the historic town centre with a bloody great half-timbered building that turns two lanes into half a lane with a pelican crossing at each end. And where is the unbelievably tight parking areas where a Chelsea Tractor and a Smart Car have managed to eliminate two other possible parking areas? When the driverless car can handle all of those (and Parisian taxi drivers), then I will gladly hop into one.
Dear Mr or Ms Grinder, no actually, I am not one of those who validates by personhood by my vehicle. I would actually love a car that drives itself, as then I could read on the journey. But tested a driverless vehicle on American-style roads is not setting a big enough challenge. I have known excellent American drivers go white when driving through a quaint Cotswold town or, indeed, Scotland. When I see a Google car zipping safely around Rome, I will happily climb aboard.
It depends on where in America you drive. It's a big place. Cruising through the plains states will be a cakewalk, yes, but the Google cars are currently road-testing in California, which has plenty of potential for crazy driving. Consider Los Angeles or San Francisco during rush hour.
"...no actually, I am not one of those who validates by personhood by my vehicle..."
Well then you're accusing people smart enough to build a self driving car of being naive enough to think the world is entirely covered by trivial road layouts (while presuming it's designed entirely by Americans who are so insular they've never visited another country). Even if they really are that dumb, every time we have an autonomous automobile post, everybody points out these problems. I'm sure they have a full list of test cases by now.
Not just Scotland. Anywhere outside of what appears to be the tunnel vision of these researchers.
"Our cities will be much better to live in, our suburbs will be much better to live in."
And it's not just graffiti covering road signs, trees and bushes do as well, I don't suppose that has occurred to them. Then there's the concealed entrances to farms and the like, usually full of a whopping great tractor and trailer just pulling out into the main road.
So, all right for townies, maybe not for the rest of us.
"I suspect the researchers are well aware of the complexity of real world driving outside the US, but one must start somewhere."
Maybe. Maybe not. IIRC only about 20% of USAian have a passport and only about 13% ever actually leave the USA to see any of the "rest pf the world"
There is no reason an automated car can't cope with all these situations far better than a human.
A tractor pulling out is much more likely to be noticed by an permanently on, un-fatigued always scanning device, that can respond in minute fraction of the time. WHile you pause for a quarter of a second after seeing the tractor the car has already started braking, calculated the trajectory and speed of the tractor, started sounding the horn, turned on the flashers to warn drivers behind and possibly started a diversionary course.
As a human you have to negotiate new road conditions as you encounter them for the first time and make a decision based upon what you think is the best option. A new driver often panics in these situations. After driving for many years yo have experienced a lot of situations and so things like this don't faze you. Now an autonomous vehicle can analyse the decision of, potentially, millions of drivers in all different scenarios and update its decision making with the best outcome in each case. It can have thousands of years of driving experience in a few minutes at the next synch interval.
The hardest issue for vehicles is potentially having to do an illegal manoeuvre for the good of the rest of the road traffic, such as when queuing traffic backs up and uses a shoulder so that they don't block a lane and cause a tailback. Would the developers be allowed to program in non-emergency illegal manoeuvres?
> And where is the unbelievably tight parking areas where a Chelsea Tractor and a Smart Car have managed to eliminate two other possible parking areas?
Of all of your questions, I would think that this is one area where we would expect self driving cars to particularly excel. It is a purely geometrical problem which we would expect them to get right every time.
"Of all of your questions, I would think that this is one area where we would expect self driving cars to particularly excel. It is a purely geometrical problem which we would expect them to get right every time."
What has geometry or parking in a tight spot got to do with my autocar stopping, dropping me off then buggering off somewhere else to park and await my summons to come and pick me up again? Or even just driving 'round the block a couple times? Sheesh, some people! :-)
How would they cope with the Magic Roundabout?
The magic roundabout is just a series of interlinked roundabouts. My mrs face was funny the first time she saw the one at MK (she's not British and had little previous experience of roundabouts prior to import). Once I explained how it worked she was fine.
Road layouts are not going to be the problem for these cars. The problem is going to be stuff like sheep, cyclists, or other drivers that don't follow road rules. It'll struggle with fallen trees or single track roads where the housewife has overshot the passing space and refuses point blank to reverse 5 meters into it instead of forcing the JohnnyCab to reverse 50 meters around a bend, etc etc. How will it cope with situations where drivers don't follow the rules of the road? The challenges are not insurmountable, but will need a lot of real world experience in order to design effective tests.
Did they also build potholes?
Manholes with stolen covers? Occasionally with a tree branch inserted to highlight it?
Bus stop grooves? Speed bumps? Really vicious near-square ones?
How long is a googlemobile suspension expected to last anyway? Or do they intend to make it go soooo slooow so as not to make any difference?
This "city" on the University of Michigan (a state chartered university with it's own police force) campus, is within the City of Ann Arbor (with it's own police force), in Washtenaw county (with a sheriff and deputies) in the State of Michigan (with state troopers, etc.) in the US (with the FBI, et. al.)
There's deer within the city limits and on the University of Michigan campus. Not by design though. Are deer a reasonable proxy for (libertarian) sheep?
There are pedestrians and bicyclists - but the vast majority of them are to the west a few miles or so.
There's townies too. The townies and the gownies sometimes make resentful noises about each other. The rest of the region is alternately (and sometimes simultaneously) envious, bewildered, and confused by the whole mess.
There are truck eating bridges in town - a north/south railroad is elevated by bridges over a few of the citie's east/west roads and apparently a truck gets it's top peeled off ever month or so.
Sorry, no double-decker buses - although both the city and the university do run buses past the facility.
"Are deer a reasonable proxy for (libertarian) sheep?"
I don't know. Do they balance on the tops of wall & then suddenly jump down in front of you?
This was one of my first encounters after passing my driving test. Fortunately the suddenness was relative; I realised it was preparing to jump & acted accordingly, but would an autonomous car?
No, they'll jump over the wall without warning. If a sheep is balancing on the wall at least you know to slow down as they might jump off. Deer will just jump in front of you without warning. I'm fortunate I've (knock on wood) avoided such collisions so far, but some are unavoidable, even with robo-car reflexes.
Having driven past many different types of animals I would say they all pretty much are the same - they are all liable to jump off walls and rush across the road at importune times. I'm including humans in that list of animals after seeing some of the dumbest reactions ever from them.
Both self driving cars from other manufacturers and themselves along with something imitating real drivers - could be done using remote control and a fixed computer program (press accelerator now, turn left now, stop now....) Some of the cars should have defective lights and lazy 'people' who don't indicate (I don't :) )
Detroit might win for empty, but I'll put the roads in Philly up against the roads in Detroit any day of the week for title of worst. Couple years ago we made the mistake of taking the main road through there when leaving a convention and my friend wound up with two damaged wheels.
...some people with a clue. Now if Blighty had one, they'd not allow AVs on the roadways until they are truly ready for prime time and have all appropriate safety systems functioning including extensive durability and operational testing. Then and only then should AVs be allowed on the roadways to discover what the engineers did not get right such as accident avoidance for sink holes, collapsing roads or bridges, flooding and much, much more where AVs are unable to detect and prevent an accident or disaster.
If the disasters truly were impossible to detect, such as the bridge collapsing while you're on it, then you're screwed either way. Where the big question mark is when something potentially avoidable occurs with scant warning from an unusual direction, say from above (a large tree branch suddenly falls on the road, for example), and options are limited (say there are cars behind and to the side). The test track would be a good place to test the limits of machine perception, but the tests themselves need to be able to put machine perception to the test accordingly. They have the setting; they still need the parameters.
"when something potentially avoidable occurs with scant warning from an unusual direction, say from above "
You may not be looking in all directions at once. The car is - and it doesn't get distracted by kids in the back seat or the cute ass/large cleavage of the girl on the sidewalk.
It will be paying more attention than most of us in that same situation. We recently had someone get their bike crushed because the driver of the bus was paying more attention to the lights than to what was around the corner. It didn't need any unlikely-to-occur often event to cause that accident - just the usual inattention most of us have regularly.
The thought is that the human, despite being half-blind, could still make out the blur, figure out it's an obstacle in spite of the poor vision, and take action. Humans do it by a combination of subconscious learning, intuition, and heuristic thinking. Sometimes we don't even think about it but we somehow recognize something is a threat and react to it. These are thought processes that we can't even explain ourselves, let alone translate into machine processes. How can we teach a computer how to quantify something that vague as a legitimate threat when we can't even quantify it ourselves? How do we teach a computer "gut checks"?
There should be one set of tests checking that the car can actually detect reliably what's around it, and another set of tests (fully virtual) checking the behavior of the car given surroundings.
Building an empty city for tests feels overcomplicated.
How many projects have been signed off on a list of "complete" specifications, only to fail miserably when actually put into production (that was never included in the specs ! that never should have happened !) ?
Enough to validate real world testing. If your simulator says everything is fine, I'm happy for you. I don't want these things in the streets I drive on until they have proven to be actually capable and reliable. By proof, I obviously mean set in a street and made to go from point A to point B without running over anything but actual road.
We have enough trouble with incompetent/distracted drivers, no need to add incompetent autonomous vehicles to the mix.
I'm guessing you're a fellow motorcyclist from your use of the disparaging term 'cagers'? I'm resigned to the fact that autonomous vehicles will happen and can see the benefits however bear in mind that there will come a point in time where critical mass dictates that human-operated vehicles no longer have a place on the roads. That'll be the end of motorcycles.
>I'm resigned to the fact that autonomous vehicles will happen and can see the benefits
Personally, I can't think of aynthing that I would enjoy more. I have to drive the same damned motorway quite a bit and it is boring, time consuming, dangerous due to geographic position and weather and simply drives(pun intended) me crazy..
Now if I could get in the back seat, wth a cold beer and a good film I would be a happy camper..
As for bikes, gave that up years ago when my back decided that the draft would become a pain...literally... I still like to admire bikes but the pleasure has all but gone .. ( My wife refuses that I buy a quad, grrrrrrr, and yet she loved being in an offroad buggy whilst in Spain )
I feel your pain (well, not the physical bit - yet). Motorways are a drag and chore driving is usually horrid but even in cars, when you get the right vehicle on the right road the sheer visceral thrill can be rather pleasant.
I'm fortunate to be able to choose when I ride my bike these days (I'm with your wife re: the quad I'm afraid, personally I'd consider a trike)
I agree with that, but that will happen many moons from now, car and motorcycle enthusiast will lobby hard against the rules and delay it several years, then the roads have to be adapted for such technology, gov't incompetence will delay that another few decades. I don't think I will be around anymore when that finally happens.
One of the subjects they covered on my artificial intelligence module at uni was the concept of a microworld. A microworld is some sort of simplified reality built for testing and developing AI systems that are designed to interact with it. It could be completely virtual, or it could be a real world environment with all the "sharp corners" taken off (A microworld for a pattern-recognition AI could for example be tested in a microworld with "sphere", "cube", "cylinder", etc but never have to deal with things like "person", "sofa", "kumquat", "squirrel" and so on).
The problem with microworlds is that systems developed in them tended to work brilliantly in the microworld, but when you tried to graduate those systems to more complex environments their performance tended to fall apart. This phenomenon became known as the Microworld Ghetto, if your AI was born there it would never escape the microworld. The complexities of the real world aren't something you can "bolt on" to an AI system, they have to be engineered in from the start.
I think Google's approach of testing in the real world with a human in a position to pull the plug in an emergency is far more likely to yield a successful outcome than building a driverless car microworld will. The photo of the track doesn't show any buildings or blind crests or poorly laid-out road markings or worn road markings or any other kind of obstruction to vision that you'd expect to encounter on a daily basis. Those are the things a driverless car needs to be able to cope with to work in the real world.
"The problem with microworlds is that systems developed in them tended to work brilliantly in the microworld, but when you tried to graduate those systems to more complex environments their performance tended to fall apart."
I would think BOTH approaches have their merits. After all, isn't it said a bird has to learn to stand before it can learn to fly? Why do Microworlds fail? Is it because the parameters don't take enough things into consideration or is it because the jump between graduations is too great? I mean, OK so you start by making sure the car works on an ideal world, then you start adding imperfections and make sure the car can compensate for those and work up from there. Why do you say this approach doesn't work?
I had to read entire textbooks on the subject so I can't really go into a hugely detailed answer but the short version is that the microworld is a simplification that's supposed to allow the engineers to focus on what they think are the fundamental problems without having to worry about the messy imperfections of reality, but the truth is that many of those messy imperfections are actually part of the fundamental problem itself, and once the system that can handle the microworld has been perfected they can't adapt it to cope with the mess they'd engineered out of their simplified model in the first place.
In short, AI is hard, and you can't make it easier by making the world it works in simpler.
"Oh, and you can't get heavier than air objects to fly. And super-sonic speeds, never. As for instant global communications, no way. And you can't get computer controlled cars to go about without killing everyone...."
Well, we know SOME things are physically impossible such as a deterministic infinite loop detector.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
