My big multilane roundabout chat

[webny99]

The thing with AI is the output is as good as the input. By asking highly specific questions and calling out errors, you can essentially control what you get out of it.

The problem is that you can only tell for sure if the input is "good enough" by either already knowing what the correct answer should be, or doing the research to double-check the output...neither of which saves any time or effort over just skipping the AI to begin with.

If your intent was to save time and effort, sure. But in this context, it seemed to be more of a thought exercise/sounding board, so I don't think saving time and effort was the endgame.

AI is no different than overhearing a conversation, or reading a forum post. You can't verify it, you don't know the source...but that doesn't mean you can't take what you learn/hear/read and do your own verification or research. If it sparks conversation and independent research, it has proven itself useful.

Agreed with one caveat...

The issue with AI is people taking what it says at face value, not the AI itself.

The main problem (applying this to your earlier point as well) is that it is professing to be a trusted source but isn't necessarily actually a trusted source, and that's a slippery slope that I would consider to be an AI problem as much as a user problem. It's mostly fine in this context, for example, but could be highly problematic in others.

[jakeroot]

The issue with AI is people taking what it says at face value, not the AI itself.

The main problem (applying this to your earlier point as well) is that it is professing to be a trusted source but isn't necessarily actually a trusted source, and that's a slippery slope that I would consider to be an AI problem as much as a user problem. It's mostly fine in this context, for example, but could be highly problematic in others.

(Emphasis mine) -- I don't agree with that, most AI tools clearly profess not to be sources, but merely a guide to an actual source. Ask any of the AI websites and they will variously describe themselves as companions or guides to actual sources.

People falsely believing them to be a trusted source is another [major] issue entirely.

EDIT: I actually asked MS Copilot (what Tom used), and it said: "I'm not a primary source myself — I'm a companion who connects you to reliable sources and organizes the information".

[Max Rockatansky]

The issue with AI is people taking what it says at face value, not the AI itself.

The main problem (applying this to your earlier point as well) is that it is professing to be a trusted source but isn't necessarily actually a trusted source, and that's a slippery slope that I would consider to be an AI problem as much as a user problem. It's mostly fine in this context, for example, but could be highly problematic in others.

(Emphasis mine) -- I don't agree with that, most AI tools clearly profess not to be sources, but merely a guide to an actual source. Ask any of the AI websites and they will variously describe themselves as companions or guides to actual sources.

People falsely believing them to be a trusted source is another [major] issue entirely.

EDIT: I actually asked MS Copilot (what Tom used), and it said: "I'm not a primary source myself — I'm a companion who connects you to reliable sources and organizes the information".

Amusingly we have one particular user elsewhere on the forum tonight who clearly believes everything AI spits out to be credible.  The assessment they just copy/pasted about the CAHSR had some fun inaccuracies (like implying the city of Shafter to be urbanized).

Me personally, I rather Scott and the gang just close the AI Pandora's box.  We have a pretty clear sample pool of AI misuse by others outside of this thread. 

[kalvado]

The issue with AI is people taking what it says at face value, not the AI itself.

The main problem (applying this to your earlier point as well) is that it is professing to be a trusted source but isn't necessarily actually a trusted source, and that's a slippery slope that I would consider to be an AI problem as much as a user problem. It's mostly fine in this context, for example, but could be highly problematic in others.

(Emphasis mine) -- I don't agree with that, most AI tools clearly profess not to be sources, but merely a guide to an actual source. Ask any of the AI websites and they will variously describe themselves as companions or guides to actual sources.

People falsely believing them to be a trusted source is another [major] issue entirely.

EDIT: I actually asked MS Copilot (what Tom used), and it said: "I'm not a primary source myself — I'm a companion who connects you to reliable sources and organizes the information".

Some guy from a major tech company:
.. so this is the path to use AI for our factory control - from accepting orders, dealing with supplies, managing equipment maintenance, product shipping..
Same guy 3 minutes later:
... Our HR runs on AI right now. I had to deal with them last week regarding my time off... Ugh... I could make it work after all...

[Tom958]

There are two things I don't like about roundabouts. First of all, there's no "take turns" rule, so it's possible for continuous heavy traffic coming in on one entrance to make it almost anyone else to get in at the next entrance, if no one gets off at the corresponding exit. I once sat at a roundabout near Sedona for about five minutes before some lazy driver accidentally left a big enough gap for me to dart into.

That's an innate limitation of roundabouts: they require a decent balance between traffic volumes on the different legs. If said balance doesn't exist, some other solution should be used, unless overall volumes are low enough to avoid the problem you describe. This one near Macon, GA looks like a prime example since there's almost no traffic on the bottom leg and therefore little to restrain heavy volumes of traffic entering from the south, which blocks traffic trying to enter from the northeast. my predictions that it'd fail and require remedial work hasn't come to pass yet, but give it time.

[Tom958]

The thing with AI is the output is as good as the input. By asking highly specific questions and calling out errors, you can essentially control what you get out of it.

I am definitely not condoning mass sharing of AI generated content, but I actually found this particular chat to be better than most I've seen and I could tell the inputs were quality. Looking past the usual fluff/transitions and extra spacing that could have cut the scrolling by two thirds, it was decently informative.

Thanks for that!

Maybe I'm crazy... but it seems like it got these things backwards. The US usually has tighter entries

Can you clarify what you mean by this? I wouldn't describe most US roundabout entries as "tight", and multilane ones in particular usually seem quite wide.

since I agree with Jake: I was referring to two things: what I call the 'pinwheel" pattern seen in American roundabouts, where the projected centerlines of the entering roadways pass well to the left of the center of the circle so that entering speeds are less than exiting speeds, and my impression from cursory viewings on Google satellite that there was no profound difference between American and Australian (and New Zealish) multilane roundabouts.

[Dirt Roads]

There are two things I don't like about roundabouts. First of all, there's no "take turns" rule, so it's possible for continuous heavy traffic coming in on one entrance to make it almost anyone else to get in at the next entrance, if no one gets off at the corresponding exit. I once sat at a roundabout near Sedona for about five minutes before some lazy driver accidentally left a big enough gap for me to dart into.

That's an innate limitation of roundabouts: they require a decent balance between traffic volumes on the different legs. If said balance doesn't exist, some other solution should be used, unless overall volumes are low enough to avoid the problem you describe. <snipped>

Sadly, the fact that roundabouts and traffic circles provide higher throughput for heavy left turn traffic (or right turns in the U.K.) than traffic signal installations, this becomes a cost savings solution that results in low traffic volumes getting stuck with Level F conditions.  But there are plenty of traffic signal examples that have the same issue.

[webny99]

(Emphasis mine) -- I don't agree with that, most AI tools clearly profess not to be sources, but merely a guide to an actual source. Ask any of the AI websites and they will variously describe themselves as companions or guides to actual sources.

People falsely believing them to be a trusted source is another [major] issue entirely.

We are not necessarily disagreeing. I just think that there is a slippery slope of trusting AI when you shouldn't that can definitely be exacerbated by user error, but is also somewhat inherent to the way it presents content as reliable and factual even when it's not.

[Rothman]

There are two things I don't like about roundabouts. First of all, there's no "take turns" rule, so it's possible for continuous heavy traffic coming in on one entrance to make it almost anyone else to get in at the next entrance, if no one gets off at the corresponding exit. I once sat at a roundabout near Sedona for about five minutes before some lazy driver accidentally left a big enough gap for me to dart into.

That's an innate limitation of roundabouts: they require a decent balance between traffic volumes on the different legs. If said balance doesn't exist, some other solution should be used, unless overall volumes are low enough to avoid the problem you describe. <snipped>

Sadly, the fact that roundabouts and traffic circles provide higher throughput for heavy left turn traffic (or right turns in the U.K.) than traffic signal installations, this becomes a cost savings solution that results in low traffic volumes getting stuck with Level F conditions.  But there are plenty of traffic signal examples that have the same issue.

Meh.  Not necessarily true.

[kalvado]

There are two things I don't like about roundabouts. First of all, there's no "take turns" rule, so it's possible for continuous heavy traffic coming in on one entrance to make it almost anyone else to get in at the next entrance, if no one gets off at the corresponding exit. I once sat at a roundabout near Sedona for about five minutes before some lazy driver accidentally left a big enough gap for me to dart into.

That's an innate limitation of roundabouts: they require a decent balance between traffic volumes on the different legs. If said balance doesn't exist, some other solution should be used, unless overall volumes are low enough to avoid the problem you describe. <snipped>

Sadly, the fact that roundabouts and traffic circles provide higher throughput for heavy left turn traffic (or right turns in the U.K.) than traffic signal installations, this becomes a cost savings solution that results in low traffic volumes getting stuck with Level F conditions.  But there are plenty of traffic signal examples that have the same issue.

Except I don't think that's the case. Roundabout throughput significantly drops with the fraction of 270 deg turns - until other legs are weak.
This, actually, is a direct consequence of "fewer conflict points" bullshit brought up as an advantage of roundabout.

[kphoger]

When I first drove in the U.K. more than 30 years ago, I was taught that the correct way to naviagte any traffic circle or roundabout is: (1) take the most direct path between the entrance and the exit; and (2) use turn signals the same exact way that you would in any other traffic pattern.  I still follow that simple rule here in the States, and not only do I not have any problem with "weaving naturally" but most everybody else seems to know where I am going.  But I can't say that I have ever seen anybody here in North Carolina do the same thing.

What does the bolded phrase mean?

[kalvado]

When I first drove in the U.K. more than 30 years ago, I was taught that the correct way to naviagte any traffic circle or roundabout is: (1) take the most direct path between the entrance and the exit; and (2) use turn signals the same exact way that you would in any other traffic pattern.  I still follow that simple rule here in the States, and not only do I not have any problem with "weaving naturally" but most everybody else seems to know where I am going.  But I can't say that I have ever seen anybody here in North Carolina do the same thing.

What does the bolded phrase mean?

No more than 3 turns.

[Rothman]

When I first drove in the U.K. more than 30 years ago, I was taught that the correct way to naviagte any traffic circle or roundabout is: (1) take the most direct path between the entrance and the exit; and (2) use turn signals the same exact way that you would in any other traffic pattern.  I still follow that simple rule here in the States, and not only do I not have any problem with "weaving naturally" but most everybody else seems to know where I am going.  But I can't say that I have ever seen anybody here in North Carolina do the same thing.

What does the bolded phrase mean?

No more than 3 turns.

Well, for those with more than four spokes, I'd imagine the rule is really not to go all the way around?

[Max Rockatansky]

When I first drove in the U.K. more than 30 years ago, I was taught that the correct way to naviagte any traffic circle or roundabout is: (1) take the most direct path between the entrance and the exit; and (2) use turn signals the same exact way that you would in any other traffic pattern.  I still follow that simple rule here in the States, and not only do I not have any problem with "weaving naturally" but most everybody else seems to know where I am going.  But I can't say that I have ever seen anybody here in North Carolina do the same thing.

What does the bolded phrase mean?

No more than 3 turns.

Well, for those with more than four spokes, I'd imagine the rule is really not to go all the way around?

The most direct path through a roundabout demonstrated:

[kalvado]

When I first drove in the U.K. more than 30 years ago, I was taught that the correct way to naviagte any traffic circle or roundabout is: (1) take the most direct path between the entrance and the exit; and (2) use turn signals the same exact way that you would in any other traffic pattern.  I still follow that simple rule here in the States, and not only do I not have any problem with "weaving naturally" but most everybody else seems to know where I am going.  But I can't say that I have ever seen anybody here in North Carolina do the same thing.

What does the bolded phrase mean?

No more than 3 turns.

Well, for those with more than four spokes, I'd imagine the rule is really not to go all the way around?

With 4 spokes, it's "Do not pass more than 15 chances to exit"

[kphoger]

I think it should mean that, if you intend to do a 270° turn, then you go against traffic in the circle and just do a 90° turn the other way.

[kphoger]

(1) take the most direct path between the entrance and the exit

OK, so, I think I've got this mapped out correctly.  How did I do?

[kalvado]

(1) take the most direct path between the entrance and the exit

OK, so, I think I've got this mapped out correctly.  How did I do?

Like AI!

[Dirt Roads]

(1) take the most direct path between the entrance and the exit

OK, so, I think I've got this mapped out correctly.  How did I do?

Cute!  I'm not sure that the U.K. rules were prepared for that.  But to their defence (spelling Anglicized for emphasis), I doubt that they would consider changing lanes twice for the straight-ahead movement to be "the most direct".  On the other hand, it would be hard for me not to do it exactly like you drew it.

[kalvado]

(1) take the most direct path between the entrance and the exit

OK, so, I think I've got this mapped out correctly.  How did I do?

Cute!  I'm not sure that the U.K. rules were prepared for that.  But to their defence (spelling Anglicized for emphasis), I doubt that they would consider changing lanes twice for the straight-ahead movement to be "the most direct".  On the other hand, it would be hard for me not to do it exactly like you drew it.

Lane(s) - a common pavement artifact, normally removed by repeated snow plow runs.

[Dirt Roads]

Sadly, the fact that roundabouts and traffic circles provide higher throughput for heavy left turn traffic (or right turns in the U.K.) than traffic signal installations, this becomes a cost savings solution that results in low traffic volumes getting stuck with Level F conditions.  But there are plenty of traffic signal examples that have the same issue.

Except I don't think that's the case. Roundabout throughput significantly drops with the fraction of 270 deg turns - until other legs are weak.

I've been chewing on a response to this all week.  I was surprised that the common thought amongst Traffic Engineers is that, in general, traffic circles (and presumably roundabouts) can be up to 50% higher capacity than a similar traffic signal installation.  I tried to put this to the test mathematically, and the theoretical numbers are even greater for traffic circles:  (A) For a basic simple two-lane crossroads intersection with equal throughput in both directions, the traffic circle has about 150% higher capacity; and (B) For the same basic simple two-lane crossroads intersection with the majority of traffic making said 270-degree turn (90-degree turn in the opposite direction), the traffic circle has about a 97.5% higher capacity.

All that being said, the theoretical throughput of a traffic circle doesn't change much in either alternative.  Yet as <kalvado> argues, the observed throughput of a traffic circle (or roundabout) actually decreases under certain traffic conditions.  For the record, I assumed that the theoretical capacity for the basic four-leg traffic circle is decreased by 25% hesitancy rate (drivers unable/unwilling to hit their "slot" on the loop; which in turn increases the available "slots" for the next "on-ramp" by half of that.  We know that, particularly here in the States, that the observed throughput of traffic circles is way less than the theoretical, *whereas* on the other hand, the observed throughput of traffic signals tends to be more than the theoretical. 

I was surprised that the prevailing view of traffic engineers has remained unchanged; when I first encountered this type of analysis some 30 years ago or so, the same 50% rule was being espoused by proponents of traffic circles.  I am curious if anyone in this audience has seen any different comparisons.

Except I don't think that's the case. Roundabout throughput significantly drops with the fraction of 270 deg turns - until other legs are weak.

And that is the point that the traffic engineers are really trying to make.  When the "other legs" have weak traffic demand, then the traffic circle (or roundabout) adds a significant benefit to the primary routing, if for no other reason, by increasing the average speed of each vehicle (quite a bit, as opposed to remaining stopped during the red phase of a traffic signal).

The counter-point here is that if the "primary route" has too much demand and the "other legs" are not inherently weak, the "on-ramps" get congested because all of the capacity is consumed by the "primary route" traffic.


For those who are not paying close attention, I'm not trying to argue with <kalvado> here.  Were are on the same side of the same coin, just discussing the issue from different perspectives.  But I should note that many DOTs are designing traffic circles to add a traffic calming feature that inherently decreases throughput, whereas the design of a roundabout typically increases traffic speeds through the intersection.

[kalvado]

Sadly, the fact that roundabouts and traffic circles provide higher throughput for heavy left turn traffic (or right turns in the U.K.) than traffic signal installations, this becomes a cost savings solution that results in low traffic volumes getting stuck with Level F conditions.  But there are plenty of traffic signal examples that have the same issue.

Except I don't think that's the case. Roundabout throughput significantly drops with the fraction of 270 deg turns - until other legs are weak.

I've been chewing on a response to this all week.  I was surprised that the common thought amongst Traffic Engineers is that, in general, traffic circles (and presumably roundabouts) can be up to 50% higher capacity than a similar traffic signal installation.  I tried to put this to the test mathematically, and the theoretical numbers are even greater for traffic circles:  (A) For a basic simple two-lane crossroads intersection with equal throughput in both directions, the traffic circle has about 150% higher capacity; and (B) For the same basic simple two-lane crossroads intersection with the majority of traffic making said 270-degree turn (90-degree turn in the opposite direction), the traffic circle has about a 97.5% higher capacity.

All that being said, the theoretical throughput of a traffic circle doesn't change much in either alternative.  Yet as <kalvado> argues, the observed throughput of a traffic circle (or roundabout) actually decreases under certain traffic conditions.  For the record, I assumed that the theoretical capacity for the basic four-leg traffic circle is decreased by 25% hesitancy rate (drivers unable/unwilling to hit their "slot" on the loop; which in turn increases the available "slots" for the next "on-ramp" by half of that.  We know that, particularly here in the States, that the observed throughput of traffic circles is way less than the theoretical, *whereas* on the other hand, the observed throughput of traffic signals tends to be more than the theoretical. 

I was surprised that the prevailing view of traffic engineers has remained unchanged; when I first encountered this type of analysis some 30 years ago or so, the same 50% rule was being espoused by proponents of traffic circles.  I am curious if anyone in this audience has seen any different comparisons.

Except I don't think that's the case. Roundabout throughput significantly drops with the fraction of 270 deg turns - until other legs are weak.

And that is the point that the traffic engineers are really trying to make.  When the "other legs" have weak traffic demand, then the traffic circle (or roundabout) adds a significant benefit to the primary routing, if for no other reason, by increasing the average speed of each vehicle (quite a bit, as opposed to remaining stopped during the red phase of a traffic signal).

The counter-point here is that if the "primary route" has too much demand and the "other legs" are not inherently weak, the "on-ramps" get congested because all of the capacity is consumed by the "primary route" traffic.


For those who are not paying close attention, I'm not trying to argue with <kalvado> here.  Were are on the same side of the same coin, just discussing the issue from different perspectives.  But I should note that many DOTs are designing traffic circles to add a traffic calming feature that inherently decreases throughput, whereas the design of a roundabout typically increases traffic speeds through the intersection.

I've seen a (poorly performing) 54k daily traffic light on my daily commute. Fhwa somehow believe roundabout can handle 40k, although I've seen research papers arguing 30-35k at most. I'll try to bring some links tomorrow.

And my disclaimer was primarily for primary route making a turn at roundabout location, with major legs being 12 and 3 o'clock.

[Scott5114]

When I first drove in the U.K. more than 30 years ago, I was taught that the correct way to naviagte any traffic circle or roundabout is: (1) take the most direct path between the entrance and the exit; and (2) use turn signals the same exact way that you would in any other traffic pattern.  I still follow that simple rule here in the States, and not only do I not have any problem with "weaving naturally" but most everybody else seems to know where I am going.  But I can't say that I have ever seen anybody here in North Carolina do the same thing.

What does the bolded phrase mean?

No more than 3 turns.

Well, for those with more than four spokes, I'd imagine the rule is really not to go all the way around?

I always thought the rule was, if it doesn't go all the way around it doesn't count, and you don't get to spin again. So said Bob Barker, anyway.

[kphoger]

I'm currently teaching my eldest son how to drive.  Yesterday, I had him go through a roundabout for the first time.  About a block or two in advance, I started explaining what to do, but he interrupted:  "I know what to do at a roundabout now."  He yielded at entry and signaled his exit.

Single lane roundabout, though.

(1) take the most direct path between the entrance and the exit

OK, so, I think I've got this mapped out correctly.  How did I do?

Like AI!

Ha.  Ugh.

I had my son navigate his first multi-lane roundabout on Saturday.  Because of his previous comment about already knowing how roundabouts work, I didn't say anything ahead of time other than that it was a multi-lane roundabout and so it works kind of like a dual turn lane.

His path through the roundabout was just like my blue line.  I'm glad there was nobody on our left at the time!  Actually, it was surprisingly fortunate nobody was next to us, because the roundabout is in a busy shopping area, a lot of people turn left at the next intersection, and we were there during a crowded holiday shopping time.

https://maps.app.goo.gl/CAvsnoVdo3ERXHnJ7

Looks like I'll have do some more detailed explanation with him.  The trouble is that he learns best by doing, not by classroom-type explanations.

[kalvado]

His path through the roundabout was just like my blue line.  I'm glad there was nobody on our left at the time!  Actually, it was surprisingly fortunate nobody was next to us, because the roundabout is in a busy shopping area, a lot of people turn left at the next intersection, and we were there during a crowded holiday shopping time.

While learning and actual stuff are pretty different, I would do blue line if there is no traffic as well.