Traffic Control of the Future

petra13 writes "A high point of the Autonomous Agents and Multi Agent Systems conference this past week was Kurt Dresner and Peter Stone's paper 'Multiagent Traffic Management: A Reservation-Based Intersection Control Mechanism.' They designed an automated system where cars reserve a time to pass through an intersection as they approach it and are then sped up or slowed down to ensure their arrival at exactly the right time. This allows traffic to enter the intersection from all directions simultaneously, eliminating the need for traffic lights and considerably reducing delays caused by stopping traffic. On their website, you can find Java applet simulations to illustrate the system. Especially impressive looking is the six lanes of heavy traffic in all directions simulation. I would love to see this in real life (from a safe distance of course)."

Breakdown?

[a1cypher]

This seems really neat, but definately not practical. What happens when a car breaks down or stalls on its way through the intersection?

Re:Um

[a1cypher]

What about the same principle, but working with a huge traffic circle?

Security

[Stile 65]

I know people have already commented on the cars not changing lanes or turning, and the possibility of breakdown, but this system would be easy to exploit maliciously. If an agent didn't slow down the car, or misreported its speed/location, that could make for a lot of... er... amusement?

Re:Um

[testadicazzo]

Well, it's initial research isn't it. Clearly there are other factors to consider. But I think it's impressive as 'proof of concept' anyway: indicating that further research (i.e. handling turns, pedestrians, etc) is worthwhile.

As for pedestrians, It's pretty common in busier intersections here in Europe to provide overpasses or underpasses. Hell I've even seen them in Canada and Alaska, and a few places in the states. So where these are worthwhile this issue can even be dropped (and in fact these kind of high traffic areas are probably the domain of interest for the technology).

This was solved centuries ago

[Stubby]

if they are not going to consider turning lanes, there is a Much simpiler solution. A Bridge. If every vehicle is only going straight, an elevated bridge is the solution.

the other problem with this solution is average car length. An accepted Average car length is 19 ft. But the first semi truck that goes through this intersection gets t-boned.

This is barely a concept techonology. Every one thinks they are a Transportation Engineer because they drive cars, the problem is always much more complex.

Network management is not a solution to transportion problems.

Re:Breakdown?

[Anonymous Coward]

Exactly what I was thinking...

While it is impressive, there are other behaviours that need to be accounted for...

1)Variation in car length.

2)Variation in available accelleration rates.

3)Cars turning corners.

4)Lane changing.

5)Cars not equipped with this system...

6)People freaking out when seeing all the cars rushing towards them!

Given that they have about a hundred fold improvement in transit time through the intersection, they probably have room to take some of these into account. Personally I'd want to build a big physical simulation of this in a downtown core in order to see how well it works with variability, and to see just how well people handle seeing these cars whipping towards them...

TJ

Not sure it would work here :(

[Anonymous Coward]

It's been my experience that in a lot of towns in British Columbia (where I'm from) that the traffic lights are tuned to force you to stop at most major intersections. I was talking to a friend of mine in the roads department here and he said that the reason it's set up that way is to entice you to take a look at your surroundings and notice the local merchants, instead of just driving by and ignoring them.

I'm not sure if that's BS or not, but it sure seems that they aren't tuned to allow traffic to flow thru at a nice pace!!!

Real world applications

[KanSer]

The first thing I thought when I was watching the simulation was it would be nerve-wracking sitting in a car when it looks like it's just gonna plow into another car. Instead of doing it in an intersection make a bumpercar ride out of it and everynow and again put in some fuzzy numbers and send bumper cars filled with people careening into each other. People would call it fun and they'd pay you to torture them! Muahahahahahahahahaha...

Re:What about.....

[sysbot]

In China, they dug an underground intersection at most major intersections to allow pedestrians roam freely underneath and therefore reduced the problem with pedestrians.

Re:What about.....

[bizpile]

I would assume that they would use off-ramp type of turn lane and just adjust the traffic the same way for the cars entering the new direction of travel as they do for the intersections. No one says you have to make only 90deg. turns at the intersection (unless of course you plan on installing this system at existing intersections with no road modification). Howev,er there is still the problem of dogs, stalls, breakdowns and the like.

Bad Science

[ccoakley]

Have you ever tried to program a traffic simulation? There are few simulations more trivial, so don't be impressed.

Second, traffic simulations based on human behavior are always post hoc analysis. Twiddle the parameters until it looks right, then make up the behavior that fits the parameters. If you've ever had a chance to play with one, they are a lot of fun. Often the whole simulation falls apart with less than a 5 percent change in some parameters.

Actually, this is true of almost all behavioral modelling. If you've ever done any reading in modelling of software systems, you know how hard that can be (try proving the safety of critical sections in a multithreaded system). Behavioral modelling has all of the great concurrency of software modelling with less determinism (or at least it should...). What makes anyone trust behavioral modelling is beyond me.

As a side note, many things in the real world are based on this kind of crappy science all the time. Check out the San Diego freeway system. The I5 805 merge was just recently redone to improve traffic flow. It failed miserably. Staffing levels on military vessels are done by models. The ships are always understaffed initially until trial and error fixes them.

Unfortunately, people think that computer simulations can solve all problems, even when there is no theoretical reason to believe that the model will even approximate reality.

Two Phases

[DumbSwede]

I've given this some thought over the years, for instance when sitting at an intersection when a light is red and no other traffic. I think what is needed is a 2 phase implementation. First mandate that all cars will be equipped with some sort of transponder that traffic control systems can read (robustly). In phase one, new cars would be mandated to have such equipment by say 2008, used cars by 2010. The benefit during the transition period, traffic lights that optimize for transponder cars, and decrease their average wait time. Phase 2 mandate new cars be drivable by a combination of remote and autonomous control by 2015, used cars to be upgraded by 2020. Advantage: several highways allow hands off driving for upgraded cars in transition period. Mandated transponders on older cars alert remote controlled cars to a hazard in the area.

Starting 2020 driving on a highway or in most city-centers without being on autonomous control would be a crime. Starting 2010 driving a vehicle not sending valid transponder signals would be a crime.

Of course the transponder signal will drive privacy advocates nuts, but I don't think you can get to a robust autonomous driving system without it. I suspect transponders are coming anyway for other reasons, so best to make lemonade out of lemons. Yes you will be taxed for in city driving -- sorry, it's coming anyway. Yes cops will know where your car was in any 48-hour period -- get over it. Yes you will no longer be able to speed - who cares as long as I can blog /. while I ride, and my average arrival time is lower due to everyone optimizing the available traffic ways.

As to unexpected hazards like pedestrians, cars will have built in radar (already practical) that reacts much quicker than even the most alert driver. Drivers will have to be insulated legally from any liability for hitting a pedestrian when said pedestrian jaywalks in an autonomous driving zone, as will the autonomous driving system manufactures.

Will children and pets be hit by robot cars? Yes, but congress will have to mandate legal protections as long as aggregate fatalities fall. Gross negligence in equipment manufacture could still be prosecuted, but any system certified by government for use should be immune from legal persecution as long as the accident falls outside of the parameters the government mandates it be able to handle. The legal challenges are the true roadblock, even if aggregate safety is improved.

While we're at it, lets lower the sound level of emergency vehicles, but have a signal override your loud radio to inform you that there is an emergency vehicle approaching. Same for trains. This could lead to some additional pedestrian accidents, but not if pedestrians are trained to use existing traffic systems better. Children could (should?) be equipped with transponders to alert the system to increase safety margins (i.e. slowdown). Of course transponders on children is another hot button topic, but I'm not referring to some 24/7 implant, but a device they carry when in downtown areas, same for the handicapped and the elderly, even your average citizen if they wish to enhance their own safety.

Transponder abuse must be a severely prosecuted crime for obvious reasons, both for sending false signals or for stalking individuals by tracking their signals.

You can fight these changes, which I believe will come, or you can live in a less technologically advanced nation. Other countries come to mind: "autonomous driving mandated ... in Japan" (ongoing /. joke). We will not have robot servants, we will not have autonomous highways, we will not have other unthought of applications of technology if we are not willing to allow our physical presence to be tracked in real space (and this means everybody). How that information is used and stored is where we must concentrate or efforts in the privacy fight.

Granted sufficiently intelligent systems would not need transponders and

Re:What about.....

[Anonymous Coward]

No, neither turning. For the reservation system, I want a black windscreen (Douglas Adams type) - so you can't see all the near-misses around you, brought to you by the Sirius Cybernetics Company.

On a serious note - ONE dog, or ONE loose wheel, or ONE faulty sensor - and whamo! You will have a deluxe mess all over the road. Could never be under human control - people change lanes, get lost, change their mind, U-turn anyway, etc.

For a realistic look at Austin TX traffic make a 3-3 road join a 2-2 road with a period of 240 (yes, 2 full minutes in this village!) and set the probabilities for NSEW at 0.1 each. Watch the fun. We watch it daily in real life. This is a UT simulation - and no solutions like shortening the cycles have emerged in this villlage (can't call it a city).

Or we could just built mass transit systems

[hal9k]

I've got an idea - create a mass transit system where cars link to each other and contribute to the overall propulsion of the train. This way we can all go in straight lines together and unlink when we need to go home. Kind of like a bicycling team, but the one who leads is the one who expects to travel furthest. People drive up to get in queue, and link at a speed exactly the same as which the train is travelling. When they need to get off, they begin to drive at a calculated speed and then come off the train. That way we dont need a bunch of lanes. Will someone make some java applets now?

We call it mathematics

[ishmaelflood]

Working in feet and seconds

s1(0)-s2(0)=7/12

a=-32.2 (ie a 1 g stop)

First car
s1(t)=v0*t-16.1*t^2
s2(t)=-7/12+v0*t

They collide when s1 =s2

so v0*t-16.1*t^2=-7/12+v0*t
so t is sqrt(7/(12*16)) or roughly 0.2 s

so the car in front will have slowed by 6 fps, or 4 mph.

So the OP was wrong, with a 1g stop, but not by much, and if she'd assumed a more realistic acceleration, she'd be right, or wrong by less.