ฝัง
- เผยแพร่เมื่อ 16 ต.ค. 2024
- Autonomous Intersection Management (AIM) is a new intersection control protocol that exploits autonomous vehicles' extraordinary capabilities of control, sensing, and communication to make traffic management at intersections much more efficient than traditional control mechanisms such as traffic signals and stop signs. This video illustrates the principle behind this new traffic control protocol and demonstrates its potential using Marvin, the autonomous vehicle developed at the University of Texas at Austin.
new cause of roadway death: passenger heart attack at scary ass intersections
at least the robot car won't go flying into someone else and then would just take that guy having a heart attack directly to the nearest hospital. :^)
so true... Lol :)
aux champs elysses...
That will take getting used to...
Lets start of with the good parts of this concept:
1. Autonomy relieves the driver (unless you prefer to be in control, in which case this is a con)
2. More efficient driver intersections (the seconds you save at an intersection total hours for the whole system - time back to the people)
3. Concepts about the future are generally engaging
Now the disadvantages:
1. If the system is centralized, hacking it could cause all kinds of traffic-based accidents
2. If the system is decentralized, in principle a DOS attack could mess with driverless cars, again causing traffic-related accidents
3. What happens to all the pedestrians, cyclists, etc? Building infrastructure at every intersection to support them isn't economically feasible...one every few intersections and you're effectively hampering their efficiency, as well as the efficiency of the road network. Allowing dumb pedestrians and cyclists like me to share the roads many intersections means [autonomous] cars do pretty much what we're doing today (stop and go signaling) which defeats the whole purpose of this 'traffic control of the future' concept. So, no pedestrians or cyclists in the future then? Good luck.
4. Cars use physics...things break. This system can be likened to a fluid, with all the cars being molecules whizzing past at a characteristic velocity. The last thing we want is for said molecules to disrespect physics, because there are always consequences for such kind of things.
I like the idea of using technology to improve our transit networks, and I like the core idea behind this project. That said, this system would only work in a vacuum -- too many elements of our expansive, built environments detract from your autonomous intersection management (AIM) system.
Best wishes,
-Float Circuit.
Most of the problems that people have with this system seem to fail to remember that self-driving cars have a suite of onboard sensors that should form the foundation of control and override all other inputs in the case of a potential collision. AIM is just intersection management, the vehicle's sensors will still double check AIM's work. They are the solution to most of the disadvantages that you listed.
1. If the system is hacked to cause accidents the cars themselves will still be analyzing the traffic around them to make sure all the vehicles around match AIM's model of traffic flow, any inconsistencies would cause them to fall back on their sensors to avoid collisions.
2. A DOS attack would cause the system to be unresponsive and any vehicles stuck at the intersection for a few seconds could just rely on sensors, and possibly P2P networking, to safely navigate the intersection, less efficiently than with a manager but likely better than today's best standards.
3. Pedestrian crossing buttons, many intersections have them already so these are obviously economically viable especially since the large light poles would just be replaced with a small pedestal. Press the button to cross and the system can stop scheduling vehicles that have to go through your crosswalk for a length of time, lights on the pedestal would indicate when it's safe to cross. Not to mention all that assumes we need to have crosswalks and self driving vehicles can't just naturally use their sensors to allow people to cross where ever. A cooldown timer could be implemented to prevent abuse.
4. Cars break very rarely, only some breaks cause a large and sudden change in control, and crossing intersections is a very small portion of driving. All this combined makes it extremely unlikely to have a catastrophic failure during an intersection crossing, and even if one does happen the affected vehicle can announce this problem to AIM which can issue commands to surrounding cars to mitigate accidents much more effectively than if there were humans in charge.
Kalleion I like your proposed solutions, thanks for taking the time to write an intelligent response back to me. Best.
We just need all the bicycles and pedestrians to have transponders or other ways of being identified including their intended path, possibly with multiple projections to account for different possible paths that they could take through the intersection.
Since bicycles and pedestrians, they are slower moving, there will be time for the server to reserve this space in the intersection well in advance.
Or totally separate cyclist and pedestrians from vehicles to avoid any delay and time, i guess for these intersections it makes sense to build them on high speed roads rather then dense crowded environments..
And if they have to be build on crowded places, then it needs to separate from pedestrians and cyclists since both groups are the most vulnerable..
So its all a matter of better planning and testing if it works..
Kalleion
The problem with your response
If the system stops responding they can use their sensors
But what if it is only that car that can t connect to the server
The system wouldn't know it is in the intersection potentially resulting in a collision which could lead to a massive pileup as cars may be following too close to stop
Also this won't work until everyone has autonomous cars it won't be possible
And if an oversize load needs to cross through the intersection
You say cars don't break often but computers do
Computers controlls these cars and if the onboard computer fails then there could be serious consequences especially with level 4 self driving cars and above (no manual controls)
Finally the cars sensors will detect pedestrians and just stop.
We all know how that will end
imo roundabouts are better. They reduce accidents pretty well and require no complicated and failure-prone technology to function, they are completely free to set up, consume no electricity or other running costs, and need no maintenance. On top of that, they function like a natural pressure valve to let the input with the most traffic get more 'share' of the throughput, while not holding an exclusive lock on it. But wait there's more... in low traffic periods they cut down on wasted time, fuel and brake wear because you don't even have to stop at all. The icing on the cake is, you can use them to do a u-turn :)
***** But they can easily produce deadlocks...
***** how so? If you can cite an example of how that has ever happened I'd be interested to see why...
***** hard to explain here, just google it. queuingtheory.files.wordpress.com/2012/05/traffic-circle-from-hell.jpg
Only problem with them is consistent moronic driver error. People in inner lane trying to turn right, people not yielding to all lanes, people can't stay in their lane, people not knowing how to use blinkers (even if by some chance they would be willing to even at a normal intersection), people driving clockwise to turn left (yes I've seen this happen.)
So using AI and computer and getting rid of the human driver is the answer then?
Cool stuff, but from a security perspective it obviously needs work. It seems to be designed as an open system, so adhering the protocol is enough to communicate with an intersection server. The first thing that comes to mind is a DOS attack by requesting passthrough privileges all the time and therfor blocking the intersection for basically all real cars.
Well in that case, you can expect hacking penalties to rise astronomically. The whole thing sounds like a bad idea to me.
Yeah and if you are close enough to communicate with the short range radio of the intersection you are close enough to just park a car, bus or truck in the intersection itself same effect nobody going nowhere till the offending obstacle is removed. Just in this case it's a rouge transmitter rather than a big lump of steel but same effect.
where is the people? ( so called "pedestrians"?)
+Gabriel Hernandez Valdivia There will be no people in the future, only robot cars ;)
And cyclists.
You hit it on the head. Such an intersection would be impossible to navigate unassisted.... I hope we will still be able to get around when not in a car!
Are we forgetting about walkovers or tunnels here? Also there are a lot of intersections which are not near any villages or cities so there will never get pedestrians near these intersections.
Gabriel Hernandez Valdivia You could still have a push to cross button.
What happens if the server goes down, even for a fraction of a second, while dozens of cars are in the intersection?
+toll_booth What happens if a person gets a seizure, even for a fraction of a second, while dozens of cars are in the intersection? Stop pointing out flaws and start helping out to find solutions
Michael Lacks The first step in solving a problem is admitting that the problem exists.
+toll_booth This is very important. It's important to discover all of the possible flaws to ensure you have solutions for them.
There shouldn't be a problem if the server goes down.. the cars simply will not proceed. No car is allowed to cross until it has been approved by the system which means their trajectory is safe. The new cars will simply stop when they arrive or communicate with each other according to a networked algorithm to proceed in a possibly less efficient manner.
Tim-J.Swan What about cars already in the intersection? What about the momentum of a car that is just about to enter the intersection and cannot stop in time? You cannot beat physics. Ever.
Been looking for a video like this. Well done!
What happens when a mechanical failure DOES occur? Boom goes the dynamite...
Mechanical failures can happen also when a car is controlled by a human. A car nowadays is already able to detect failures and inform the driver with an indicator on the dashboard.
If you mean a system failure, it will probably implement multiple systems that work independently from each other. You could make one system that works purely on GPS and communications with other cars and then implement also another system (comparable to Google's car) that only works with camera's and collision detection. Also cars are still controllable by a human at any time so you have even another layer to fall back to.
Yes, certain systems will cost lives, especially in the beginning. But overall I think that overall it will save more lives (priority vehicles can get faster to their destination, less polution, less car accidents caused by humans etc.)
I'm not saying humans can't work it out, I'm just saying it'd be mighty easy to cause a major disturbance with that many vehicles traveling that fast in such close proximity.
Still safer than humans driving.
True. You take out the human element but not the car element. Cars do not function perfectly and accidents are also caused by tire blowouts, mechanical failures, and unpredictable rode conditions. If cars are traveling at roughly the same speed in the same direction around you (like they do now) the risk of death goes down substantially compared to the system shown above (head on collisions would be more statistically likely). Our current system accounts for individual failures and errors, the above system does not - no system is perfect.
Mechanical errors would occur much less frequently than human errors anyway so it doesn't matter. Automation would still save thousands of lives even if some people still die.
This will be awesome! You can get adrenaline rush after adrenaline rush as you pass through intersections commuting to work. Screaming my head off as I pass through an intersection with 20 lanes of non-stop traffic flowing into it!! With centimeters of clearance between cars!!
This is a cool idea, but sounds very optimistic and has a lot of assumptions. No matter how far technology comes, mechanical failure and software bugs will ALWAYS exist.
tryhardnoob
Yet still ok with the invention of Computers, iPhones, TV etc
@@alohatigers1199 well it depends on what you call ok, computers and smartphones have bugs and exploits too, some of them with problematic consequences. With a vehicle, lifes are on the line and we need to have air industry type mentality to the issue, by not relying on assumptions and taking into account everything that may happen. That requires a pessimistic and cynical attitude in formulating a solution, which in in my opinion, isn't fully present
Yes. You're right. This is the reason why all the flight traffic is completely manual today. Hooops. I'm wrong. The flight traffic and navigation is completely automatic. Yes, you're right: less risks when you're in the air, 30000 feet high, - 60 celsius degrees, 600 km/h.
Most of the times, our minds are our limits.
But still much safer than a human driver.
I never thought about testing 1 real car in a mixed-reality environment; That is very clever!
In Europe we invented things called roundabouts quite some time ago.
Is that computers are called in Europe?
Do you have a lot of roundabouts in downtown London, Paris, or Rome? I'm not being sarcastic. In the USA roundabouts tend to be large and wouldn't fit in every intersection of a large city. I'm also curious how your roundabouts handle pedestrian traffic? Ours use standard crosswalks, which I don't think would work in a totally automated intersection or roundabout.
Roundabouts aren't as efficient as everyone's pointing out. Roundabouts are only efficient on less frequented roads and were primarily designed to prevent cars from stopping prior to entering the junction. When they're built on a busy junction, they require the same amount of traffic coming from all directions to work properly - which especially in rush hour isn't possible.
Imagine a 4-way roundabout in central europe on a busy junction in the rush hour. Imagine there's a lot of traffic coming from the west, most of the cars want to drive straight through to the east. There's also a similar amount of cars coming from the south who want to take the same exit. There's almost no traffic coming from the other directions. What will happen is that the traffic coming from the west may keep their speed and drive through while the road on the south clogs up, since they're only allowed to drive when someone turns into their road. Conclusion? In a situation like that the roundabout works exactly like a traditional junction with yield-signs on the southern and northern end.
Traffic lights don't just regulate the junction, their main task is to regulate the traffic on the road to also make it more accessible for connecting roads with yield-signs. In the rush hour, the yellow phase of a traffic light might be the only way for a person living on that road to leave their house since it generates a gap in between the cars. Traffic lights also ensure that everyone gets the same amount of priority they deserve.
we have a TON of roundabouts in Australia.
building them appears to be the favourite hobby of town planners here.
They come in many sizes. Most of the ones you come across are basically the same size as a standard urban intersection though.
Roundabouts are most frequently seen on moderate traffic roads.
Low traffic roads don't bother, and the highest traffic roads use traffic lights usually. (though there are some roundabouts even on these roads. Why some intersections get lights and some get a roundabout I'm really not too sure.)
In any event it's the medium density roads that tend to get the bulk of the roundabouts;
The 2 lane roads that carry a lot of traffic and have moderate speed limits.
You get a decent number on 4 lane roads as well, but if a 4 lane road crosses another 4 lane road you're more likely to get a set of traffic lights.
And yes, downtown London and Paris DO have quite a few roundabouts. Really, really large ones in fact.
Though you have to remember these are very old cities with lots of very narrow streets where you almost can't do anything to a road without major consequences.
Even so those kinds of cities do have roundabouts, and quite a lot of them.
We have a lot of tiny urban roundabouts. Some can even be nothing more than a one-meter-wide circle painted on the ground. They still make the towns much more navigable.
I don't know about you but my whole life I've been driving my Cadillac Fleet Wood Brougham D'elg land yachts and I don't think I'll stop for both comfort and the peace of mind knowing I am in control. The whole reason I love driving so much is because I can drive. Technology like this would leave us car folks a little bummed out if anything. Still, for your general non-car-caring public, this would be a great alternative to waiting in traffic.
So by attacking the server you essentially shutdown the whole intersection system?
Translate this to all the intersections of a city and traffic will be disrupted for several hours.
Such a system should be decentralized.
i agree. they need to be robust and stand alone with maybe network connections that can send traffic data out. cause that would be pretty useful.
if someone were to disable the server with malicious intent. i think that would breach grounds for attempted murder. hackers should only deserve the death penalty.
Well, you can already just dump a block of concrete in the middle of an intersection and cripple it until it's removed.
Worst case scenario - the servers are unavailable and the cars fall back on the same systems which they use currently
Just go back and watch Fate of the Furious. You can see how one hacker from a highly-advanced technological plane flying high can remotely hack cars in New York from thousands of miles. That could easily become a reality and even scarier when the hacker can take control of his codes and do what he wants to do.
The fact that people are approving and others are disapproving and critiquing is fantastic. That's a sure sign that this is coming soon.
I love driving, and I drive a lot but I can`t STAND dumb drivers and wasteful red lights.
It`s time to change our slow automatic system where you risk paying hundreds of dollars for rolling through a red light or stop sign where there is absolutely no danger.
this is so cool but fail safes of some sort would need to be put in place
+FTG Yeah thats exception tho
With better software engineering and better efficient hardware it can improve even further.
AIM seems like a valuable addition to the toolset -- how does it work under challenging circumstances? E.g. when the traffic exceeds the volume that can gracefully use the intersection, how well does the algorithm do? Can you give some throughput comparisons with traditional signal policy? How does it do on bursty traffic? How does it do in fog/heavy rain/snow that somewhat degrades radio signal? Can you summarize what the system does when it detects noncompliant traffic (bike, pedestrian?)
so many people are against this brilliant idea. Such a shame.
This is how science/engineering works, my friend. An idea is proposed, and others react with critiques. My guess is that the engineer(s) that posted this are delighted to get the feedback.
I want to study about the whole trajectory planning for a case like this, what might be a good place to start.
imagine you are travelling for the first time in an autonomous car and you reach a busy intersection with such speed .You would probably shit your pants
And municipalities could/will charge for subscription fees allowing highest priority to highest paid subscribers. :)
drivers already pay for road usage though tax. don't give them the idea that they can double dip, because they would most certainly try.
Good idea.
There would definitely have to be laws made to prevent cities from doing this, but it would make sense for emergency vehicles
Love the idea, assuming everything works perfecting you can actually have all the cars never stop. Only adjust speed so the cars are timed to move through gaps between each other. However it can't work with current car and drivers, it would have to be a closed area to autonomous cars only.
Very simple solution. You can always create a way to put the system on old cars just like everything else we've already done to old cars. Some old school cars that never had AC are now custom made to have AC and the skies the limit. Just like some cars today already bring a GPS, we still have GPS on old cars because we can just by in separate. Cameras and sensors is all that it takes to connect to the boxes and then connect to every other car so it can move like an ant colony all working together as one collect brain. Hope this helps clear it up. Take care.
A radio frequency jammer would be fun to use at one of these intersections.
"Intersection Manager" basically a server that can be compromised (hacked) like any other network computer.
That computer does what everyone Should do
Great stuff but one thing is rarely discussed in these scenarios; A car with bad intentions, having a scrambler or worse, an antenna array designed to hijack and mislead vehicles for nefarious purposes. Such persons and vehicles could do a lot of criminal damage for bad intentions. Remember students building am/fm base stations and driving around in traffic to hijack the radio signal of the other cars in traffic to prank them? Same principle but with potentially much more harmful results. The traffic management computer will probably need to have some kind of encryption that is decrypted in real-time with the vehicle in communication in efforts to prevent misuse.
So I wonder if the red car speeds up. the blue car slows down, or if the load can be shared and how this affects the ten cars on either street. Also, would eliminating left turns altogether save us time?
Finally a cool video without everyone in the comments saying "nOboDY: yOUtuBe reCoMMenDatIoNS iN 2020!!!11!"
What is die the expected improved capacity? Compared to an equally sized intersection managed the conventional way?
Wow we have come a looong way in 6 years!
Hi, could you let me know where i can find the project website? thanks
And IF the car can accelerate and turn at the expected trajectory. How does the system deal with cars with different torque, horsepower capabilities? Or acceleration, turning, traction, and braking characteristics?
How does it measure and deal with the variability of these (engine warm vs cold, variability in air/fuel mixture depending on altitude and atmospheric conditions, gradual loss of turning ability as tyres wear out, loss of breaking ability as breaks wear out, traction due to gravel...
My thoughts on this kind of idea were that there would be groups of cars bulked together- just inches apart- kind of like how modern intersections operate today... But these segments would be in multiple forms, sometimes different than today's sections of vehicles going through intersections, essentially as a form of bulked threading going through the intersection. And of course we cannot forget the maxim of 'the faster the vehicles go, the less time they spend in the intersection'. So I guess there would need to be various modes for different speeds of roads intersecting, and there would need to be various modes for different traffic densities intersecting.
^But what I meant by this (on the most simple basis) is that the organization of groups of cars created today that are stopped and have to wait (often) would instead be grouped ahead of time, depending on lead-up to the intersection and other circumstances... and of course, all that yellow light mumbo jumbo bumbo stuff would go away. You'd see really close calls with changeovers. No more Dom Jolly crossing the street in a three cornered hat or a snail costume (which hopefully is what modern intersections would hopefully seem like to someone used to such speedy efficiency).
........Would this make the road flyover and underpass obsolete? Also, if this became a reality then you would need to blindfold someone with a reasonable tendency to PANIC when you see vehicles flying right towards you at quite a clip.
I don't know if anyone noticed, but Marvin made a *really* hard stop at the stopsign. Not only would this cause passenger discomfort, but it could lead to a potential loss of control in a slippery environment. Think of Seattle streets after an ice storm. Do Marvin's architects handle all of these circumstances and if so how?
Will there be no traffic lights in the future?
Ok, and how does the AIM server signal conventionally manned "humanomobiles" whether their estimated trajectory is cleared or not? Traffic lights re-purposed to single-car signaling?
Always giving non-automated cars the right of way does only scale well if those make up a small fraction of traffic. The transition to fully automated traffic will start from the other end though...
Less reaction time doesn't really matter when you are a computer with a reaction time on the order of milliseconds. If road conditions suck and the car loses traction the computer would notice it and adjust for it before a human driver would even have realized it. A robotic car could certainly avoid other cars, and much more effectively than a human, considering the car has 360 distance sensing view and ability to communicate with other cars.
what is the software that used to building a app like this
Do we really need an intersection server? That is not going to work so well during a power failure or system failure. Is it possible to do a virtual server, that resides on the self-drive cars only? Currently, people still need a way to pass through intersections when the traffic signals are not working. Driver's training says to treat it as a 4-way stop sign. Erring on the side of safety rather than efficiency. Perhaps in reality, people slow down and take turns, sort of like how the traffic signals work.
Roads will work so much more efficiently and traffic congestion would be reduced, once slowing down for intersections is minimized. Another beauty of the soon-coming self-drive cars.
One thing to think about: There are hundreds of intersections that are hardly ever used. Somehow, the AI cars will have to keep track of which intersections have a server, and which ones it will just have to scan for itself. For example, a stop sign in a subdivision/neighborhood.
Autonomous Intersection Management (AIM) is a new intersection control protocol seems good for individuals
Hey csatc2002, can you put a link to the video you used? Thanks!
What if the road is wet. Or iced over?
How would someone walk across that intersection? Do pedestrians, cyclists or other humans just shut the whole thing down?
waiting for this to happen!
Brilliant work!
No thanks. I'd rather wait 30 seconds than travel through several near misses. 1 software glitch, 1 terrorist hack, 1 tire blowout etc.. And boom huge multiple car crash. No thank you. Sure reduce the red light time according to volume of traffic but not everyone going through at the same time.
Daughter/Son: "Hi dad, I'll be staying out a bit later tonight with your car"
Dad: "Ha, sorry, I just sent a text to the car to drive you home. Sorry hun, maybe next time"
Over/Under-pass most likely. Or teleportation, which would be much more FUN!
Since obviously there isn't going to be an instantaneous switch to autonomous vehicles -- there will be both autonomous and human-driven vehicles for some time -- maybe a hybrid of this system and a regular traffic light could be useful. Human drivers would follow it like a regular traffic light, while autonomous vehicles would have the freedom to run the red, but must always yield to human-driven vehicles.
The *yielding* can be extended to pedestrians and bicyclists as well; the system can use data gathered from sensors and cameras to try to predict the trajectories of humans based on physics. Since physics alone can't perfectly predict what a human might do, it's probably best to use a dynamically updating system rather than "reserving" trajectories. A human could suddenly stop, for example, so the system should update the autonomous vehicle trajectories to accommodate that within milliseconds.
This beautiful....until the EMP goes off
Indeed
Crashes in the server would be disruptive but not very dangerous. But malfunctions could give the vehicles a false sense of security. I wonder what the failure rate would be and how could the autonomous vehicles would be at compensating
They do. They underscore the implicit message of large intersections that people not in cars are second-class citizens, whose needs, comfort and convenience don't matter, at least in comparison to the all-important car.
Benjamin Alexander Don't know how it is where you live but in most of America we live in suburbs and commute into the city for work.
Where are the pedestrians, cyclists, and non-autonomous drivers in this video?
They would probably have to have red-cycles where no traffic enters the intersection to allow other users to cross, and this would only be efficient if there weren't many non-autonomous vehicles
How do you deal with pedestrians, fallen trees, flooded roads, etc?
Autonomous cars can eliminate human error without this intersection. Therefore, the only relevant safety question is "How severe are mechanical failures?" The authors simply gloss over this question.
As shown, AIM has many cars on intersecting paths traveling at moderately high speeds. Mechanical failures will result in dangerous broadside collisions with little to no braking distance. Today, they would generally result in less harmful rear end collisions with significant braking distance.
No worse than the mechanical failures in car brakes, mechanical failures in existing traffic signals, etc. And honestly, it shouldn't be too hard to get the mechanical failure rate below the rate of human failure due to distraction, inebriation, lack of sleep or any number of other things.
I don't see how this will be possible in my lifetime. It'll take decades to get non-autonomous vehicles off the roads. And this system would only work if all vehicles are autonomous and communicate with the server. If you include any human-controlled vehicles, bicycles, or pedestrians without additional considerations the system will fail.
I'd like to think you could build a system that could merge the old and the new together. Have the server control the traffic lights, while still communicating with autonomous vehicles to help optimize the traffic lights. This would allow bicycles to share the streets with vehicles and allow pedestrians to continue to use the same crosswalks.
And, if I understand, placing a server at every intersection seems costly and realistically prohibitive. You're actually buy and install a server at *every* intersection on every back road? No. It seems much more realistic to have this 'AIM' be a kind of distributed network handled by all vehicles near that intersection, right?
Rather than be server based (which could be crippled somehow disabling thousands of cars), wouldn't it be better for cars to radio each other locally to negotiate maneuvers?
It would probably end up being a mix of things. The AIM would offer a path, but the cars would navigate it.
Think of it like air traffic control: despite ATC saying they're clear to land, and despite the on-board components saying they're clear to land, the pilot can refuse that clearance and pull up once they see a tram parked on the runway. And if the ATC isn't available or is compromised, they can radio other planes in the area.
Yes
+sirdeadlock Exactly this is my thinking as well which is why I think these things where people say stupid stuff like what if the server is hacked and tries to intentionally crash cars is stupid. The car will still have it's own sensors and if it detects a collision threat it is just going to reply with a NAK (Negative Acknowledgement) packet rejecting the clearance offered (Probably with a flag set to indicate the rejection reason). Basically the NAK packet would serve the same function as simply using the word "Unable" with ATC.
Fair enough; that makes sense. Does this mean that, in exchange for autonomous driving, we will all have to have more or less identical cars? I always like to think its fun to have diversity in the ecosystem, but I suppose have a "fast" car is a moot point if you are not the one driving it.
6 years late to the party. But still.
Yes, an inadvertent mechanical failure in the middle of the intersection would invalidate an awful lot of trajectories all at once. Weaving looks cool though, and seems probably close to optimal for minimizing transit delay for an individual car in low to moderate traffic conditions. But if your cars are already communicating in a mesh topology, there might be other ways.
For example:
In steady state flow, one could separate cars into "packets", clusters of cars travelling very close together at equal speed. (You could even have them draft one another and cycle the head car to save fuel.) But regardless, the advantage comes at intersections: they can be programmed to understand and route the car packets such that the groups travel through a much faster light cycle than currently available, with very little need to slow down or stop. It is similar in idea to weaving, but operating on larger groups. And because light cycles still occur, any mechanical failure can be reported to the neighboring cars, who can then respond sanely, because all of the moving cars at the junction were travelling in the same direction to begin with. And finally, because light cycles still occur, it's not entirely impossible for a limited number of humans to continue navigating such a system, especially if the car could warn the system ahead of time that a human was approaching. However, this relies on retaining the paradigm of speed limits; if one were to abandon them in pursuit of even shorter total transit delay, the system becomes prone to bottlenecking when a slow vehicle must use the roads.
Alternatively, since we're re-engineering the entire road system, why have cars at all? Modular autonomous rental vehicles could be very profitable for cities or corporations. They could simply combine together in much the way trains do, but dynamically while in transit. It is possible that a limited amount of mechanical failure could be tolerated by such systems, especially if engine-only cars were added for redundancy to handle precisely such emergencies. Intersection handling would simply use railway logic. This also means that speed limits could be set as needed by the public or private organization(s) involved in operation. Open and well-planned protocols would be needed to prevent abuse of this system. Nonetheless, the financial incentive to convert car transit into a service would be considerable if the idea could be sold well enough to the general public.
Of the two ideas I just now randomly crapped out, I prefer the first one. It would enhance the capacity of the traditional 4-way intersection (one could even envision maneuvers such as safely using the entire roadway to perform left-hand turns, since all the cars are doing it in tandem) while not entirely sacrificing the ability of humans to continue using the roadways. Of course, humans would then become the new bottleneck, since they would still require lane control and, obviously, couldn't be expected to join car packets under manual control. There may be workarounds to this as well (perhaps coordinated passing maneuvers when a cluster of vehicles has a higher speed specification than another.)
Well. That sure was a nice thought experiment, wasn't it?
where can i download AIM?
this was dated 9 years ago, nevertheless, it still amazes me..
In a manner of speaking, yes. If you're actually curious, read more here: cs.utexas.edu/~aim/ "Autonomous Vehicles Meet Human Drivers". AIM speeds up intersections by a factor of several hundred - but only when all cars are autonomous. When even 1% of intersection interactions are done by humans, AIM's performance degrades significantly. When 10% of interactions are human, AIM's speedup is virtually zero.
Duncan Bc Canada would be a good to do a small scale AIM project, square pattern with mix highway and city roads; check it out.
Also make a accelerator Universal control/wiring harness and a Deceleration wiring harness for all makes of cars. The person just steers the plotted trajectory.
I read an excellent article about car-herding, where the autonomous (unoccupied) car could be stopped and forced to take another route by mischievous pedestrians, if the safety protocols of the car were to avoid impact with them. certainly, there would have to be complete separation of cars from human elements if swarm activity were permitted at junctions. Humans have no hope of safely traversing such a system unless the communication protocol allows those human elements to be given a wide berth
Who's liable when the system or systems fail? Who will be sued?
DrSmokeTrees No answer? Who's liable?
no one knows
If it is a vehicle that failed, either the owner of that vehicle, or the manufacturer if it was something that would be subject to a recall. If the server itself failed, that would be analogous to the failure of a traffic light resulting in causing an accident. However, odds are that a failure of the central server would result in the vehicles automatically falling back to a backup system. That system could be something as simple as the vehicles treating the intersection as a 4-way stop, to something as complex as the vehicles using protocols to create a virtual traffic signal.
If your car's break malfunction and you crash because of this mechanical malfunction, who's liable? Why would mechanical malfunctions be more of a problem for liability in the future than nowadays?
Besides that's a system that can have built-in failsafes, making crashes really improbable even though a fail could result in a little congestion. Self-driving cars will be program to trust their sensors to avoid collision. Even if it has a lane reserved it will stop if it sees something in its path.
The automated cars already handle people getting in the way and deal with it according to the rules of the road. If the car finds an obstacle it just cancels its intersection reservation and requests a new one when it knows when it will be able to get there - not all that difficult to figure out.
Planes are held to much higher security clearances.
I forget exactly where but I have already seen a video where some "hackers" got hold of common vehicles already on the road and were able to do stuff like disable brakes, steering, or make the vehicle steer by itself erratically, display no speed, etc...
CAN networks are very simple and once you're in, that's it. Especially considering more cars have USB ports now, it's not that hard...
good one on future traffic control
Another flaw to this AI : What happens if output lane is congestionned ? If the car has to brake while under the crossing, this seems dangerous
Wouldn't the intersection manager notice stalled movement and reroute traffic? In a practical application there would likely be multiple cameras monitoring the intersection for that kind of thing. And being computers, it would respond preemptively rather than reactively.
Its depressing that people are thumbing down the nich's comments
so this would only work when everyone has to use self driving cars what will be far from now but when we are there it would be pretty great
How are bicyclists and pedestrians going to navigate this?
Dave Holzman Umm, push button to cross, the same as we have right now.
Watching this 5 years laterXD
"and harmful emissions can be dramatically reduced as well"
Thing is, once this gets implemented which won't be for a while, there won't be NONE of that. But still, does reduce congestion. But you gotta consider what if the power goes out? Something hacks it or weather breaks it? Like yeah u gotta have a backup. And also what about bikes and pedestrians. Still pretty cool idea just have to work out those things
+Tommy Sandal Also AS LONG AS THERE R NO MECHANICAL FAILURES so ya
+Tommy Sandal Well anything depends on no mechanical failures. "Your plane will land safely as long as there are no mechanical failures". That can literally be said about anything.
ByteMe That's what I was trying to get at
If the power goes out and knocks out a traffic light, the intersection becomes a four way stop sign. I imagine the same thing happens here.
What if a Hacker programs a Car so that it sends out fake requests essentially blocking traffic alltogether by making the server reject every other request.
The problem with this is you end up dumping too much traffic onto the outlying intersections. Unless you upgrade everything to handle the flow. But you will always create a bottleneck somewhere down the line. Traffic lights suck, but they’re like traffic flow regulators..
Is this SUMO tool???
"as long as there are no mechanical failures". Good thing that never happens in the real world either.
Actually, a computer could quite easily be programmed with a learning algorithm that would be able to predict the slide before it happens and act accordingly. In fact, computers are more efficient at learning than humans, as it only takes one failure/success event for the computer to improve its technique one "increment" where as humans are slow learners, and take multiple events to improve an equal increment. Just because it hasn't been done yet, doesn't mean it CAN'T be done.
I suspect a system like this would be more reliable than human drivers; we react to surprises rather poorly, in general. You might grind an entire intersection to a halt as each vehicle autonomously avoids a collision and then the system works to sort itself out. But that's not a lot different from the way it works now; the server acts like a police officer waving traffic through, regardless of whether there's an obstacle in the way.
I hope, the future is coming soon
Is that a vehiCROSS?
This system makes several assumptions.
1. That the general population will even be able to afford, let alone be willing to operate, self driving cars.
2. That you can provide a totally secure and totally un-iterruptable wireless data connection instantaneously to multiple vehicles moving at high speeds. (anything over 30 mph is high speed for fleshy meatbags like us)
3. That the system will never lose power, will never be tampered with or hacked into, and will never experience interference from natural or man-made causes.
2014... ..?? feels like a distant memory now.
What happens in a blackout?
this wouldn't work unless almost all or most cars are autonomous, right? so that means it probably won't happen within the next ten years. Many people either don't want a new car or can't afford one. I know many people that still drive cars from 2004 and earlier.
Yep. But it's interesting and probably useful to start speculating and trialling now
Like this ;)
Or build a tunnel and britch. All traffic that goes straight doesnt have to stop. And than add quater circle for going left or right.
The only problem is that non autonomous cars may still be on the road, and would not be able to go through these sorts of junctions.
Build pedestrian bridges. Those already exist at plenty of large intersections.
What about pedestrians, bikes, and hackers? Bad idea
When a hacker will take control of this fucking autonomous cars it will be carmagueddon.
That's an issue with mis-labeling or mis-loading the machine and has nothing to do with the machine itself, only the human who serviced it.
Just saying.
yey now you can smoke weed and drive at the same time :D
and what about glitches?
and what about hackers into the system?
I put a dollar in the vending machine to buy a soda, the machine gives me the wrong flavor and screws up my change.
Fix that before you invent this crazy crap.