ABout AGCs in boxes. just a few points, I agree they are much better than bowties, and can be hidden from view by editing them while zoomed in high. a few pointers:
You don't need any elevation changes, everything can be done at ground level, if necessary to have a segment crossover, use a junction with red arows.
You don't need to get a double turn left instruction, if you work the angles right. you can set it up with a turn left entering the segment, and no instruction leaving it, etc..


I've been doing a couple of tests, and with what Waze said, I'm learning, that the turn angle is not as important as the roads being parallel. I have a box I edited with 90° turns all around, and got U-turns, the next day I moved the dogleg from just before the turn, to just past the previous turn, so from the previous junction to the turn the segments were parallel, and exactly 14 meters apart. since then no uturns in that direction, and I'm still able to get u-turns in the other direction.

TL:DR
We don't yet have the full list of criteria for preventing U-turns, but they are more than just the length and turn angle. However it does seem posible with careful editing to accomplish this without any extra nodes or segments, (and NO bowties).
If it seems too difficult to achieve, use the hidden AGCs instead of bowties.
don't use fake elevations to overlap segments, Livemap (and hopefully soon the client) displays elevation now.

shawndoc wrote:Are AGC allowed for U-Turns? It always used to be a big no-no to use them for that and the type of thing that would get you a talking to from a CM.

that is dependant on your RC. Its not in the wiki yet, so ask first.

taco909 wrote:

PesachZ wrote:ABout AGCs in boxes. just a few points, I agree they are much better than bowties, and can be hidden from view by editing them while zoomed in high. a few pointers:
You don't need any elevation changes, everything can be done at ground level, if necessary to have a segment crossover, use a junction with red arows.

Seems that multiple junctions with red arrows increases the complexity, and in some cases might run the risk of <5m segments, especially here in SoCal where our justification for splitting is the multitude of driveways rather than 15m+ medians. We have a lot of 13m boxes.
Also, I recall reading that there is a very small "junction penalty" imposed... would extra junctions on an AGC not promote the right-turn-u-turn avoidance bug for left turns, or is this penalty insignificant in this application?

Junction penalties only apply to new junctions. Once they have traffic speed data, that is used instead of the penalties.

taco909 wrote:

You don't need to get a double turn left instruction, if you work the angles right. you can set it up with a turn left entering the segment, and no instruction leaving it, etc..

I was referring to the double left instruction to make a u-turn from anything other than a bowtie, whether box or AGC.
I agree, a normal left turn should only get a single instruction at the beginning of the AGC.

It can be built with one left turn even in a box using AGCs, but that's another discussion.

taco909 wrote:

I've been doing a couple of tests, and with what Waze said, I'm learning, that the turn angle is not as important as the roads being parallel. I have a box I edited with 90° turns all around, and got U-turns, the next day I moved the dogleg from just before the turn, to just past the previous turn, so from the previous junction to the turn the segments were parallel, and exactly 14 meters apart. since then no uturns in that direction, and I'm still able to get u-turns in the other direction.

So effectively a double-dogleg, the first at the junction maintaining 50+ to avoid a "stay" instruction, and the 2nd minimum snap-distance away to pull the roads parallel?

I just use one dogleg, like here. the u-turn when travelling NW is blocked, but going SE allows it.

taco909 wrote:

TL:DR
We don't yet have the full list of criteria for preventing U-turns, but they are more than just the length and turn angle. However it does seem posible with careful editing to accomplish this without any extra nodes or segments, (and NO bowties).
If it seems too difficult to achieve, use the hidden AGCs instead of bowties.
don't use fake elevations to overlap segments, Livemap (and hopefully soon the client) displays elevation now.

Looking forward to the long-term results on this. It will be nice if we can document what is happening with consistency.

me too

taco909 wrote:Ahh... single dogleg with the angles right on to prevent "Stay" instructions.

I took a brief look at the configuration but not that closely.

The angle don't prevent stay, they prevent turn instructions. You only get a stay, if there's more than one turn that's less than 45°

Sent using Tapatalk for Android 4.4.2

They are working on it (possibly actively) there are rumors of a U-turn verbal prompt in the client code which just hadn't been implemented yet. Perhaps we could say "very soon". Waze has mentioned several times they are working towards this. Also the very soon release of junction boxes, will hopefully eliminate all bowties, and 14m median guesswork, while allowing real world looks, instructions, and restrictions

Sent using Tapatalk for Android 4.4.2

Linking to another thread reporting a U-turn prevention failure.

Sent using Tapatalk for Android 4.4.2

TrafficGuy01 wrote:UR reported Waze routed through a No U-Turn and the segment in questions is set at 13 meters

https://www.waze.com/editor/?env=usa&lo ... s=70700821

Wiki Guidelines indicate

U-turns can be avoided where two split roads intersect (in a # configuration, aka a Box junction) or a split road intersects an unsplit road (in an H configuration junction)

The length of the median segment (the "H" cross-bar) is 15 m (50 feet) or less.

The two parallel segments of the split road being used for the U-turn are one-way segments.

The two parallel segments of the split road being used for the U-turn are approximately parallel to each other.
*********************************************************************************
For safety to avoid possible errors caused by rounding and display discrepancies, and to ensure proper U-turn routing prevention, always avoid the gray area. Do not set median segments with an exact length of 15 m (50 feet), leave a 1 meter buffer. If you want to allow U-turns, set the median segment length at least 1 meter greater than 15 m (50 feet). Likewise to avoid U-turns, set the median segment length at least 1 meter less than 15 m (50 feet).
**********************************************************************************
So this does not seem to be working

PesachZ wrote:Linking to another thread reporting a U-turn prevention failure.

Sent using Tapatalk for Android 4.4.2

TrafficGuy01 wrote:UR reported Waze routed through a No U-Turn and the segment in questions is set at 13 meters

https://www.waze.com/editor/?env=usa&lo ... s=70700821

Wiki Guidelines indicate

U-turns can be avoided where two split roads intersect (in a # configuration, aka a Box junction) or a split road intersects an unsplit road (in an H configuration junction)

The length of the median segment (the "H" cross-bar) is 15 m (50 feet) or less.

The two parallel segments of the split road being used for the U-turn are one-way segments.

The two parallel segments of the split road being used for the U-turn are approximately parallel to each other.
*********************************************************************************
For safety to avoid possible errors caused by rounding and display discrepancies, and to ensure proper U-turn routing prevention, always avoid the gray area. Do not set median segments with an exact length of 15 m (50 feet), leave a 1 meter buffer. If you want to allow U-turns, set the median segment length at least 1 meter greater than 15 m (50 feet). Likewise to avoid U-turns, set the median segment length at least 1 meter less than 15 m (50 feet).
**********************************************************************************
So this does not seem to be working

After investigating this intersection, and the segments, it seems the issue may be that it didn't meet one of the criteria listed in the wiki.

The two parallel segments of the split road being used for the U-turn are approximately parallel to each other.

The southern segment was connected at a widening angle so where it connected to the median segment it was 13 meters apart from the northern segment. However where it junctioned with the ramp the northern and southern segments were 19 meters apart.

I've adjusted the segments to be parallel for the length of that first segment, and let's see if that fixes it after a tile build.

I've received and tested new information from Waze which radically change what we thought about how U-turns are prevented.
I'll be working to elegantly add this to the wiki. But in all preliminary tests it has been working [emoji1]

Waze staff wrote: 1. A median shorter than 15M.
2. Parallel roads. The angle is determined regardless of the median. It's the angle of the start of each segment in each road relative to the north and the driving direction. In the case of parallel roads with opposite driving directions, one will have an angle of 0 degrees and one of a 180 degrees (give or take 5 degrees). This is regardless of what happens to the segment after the start point in that intersection​.

To explain this a bit clearer I like to look at it as 3 rules.
1) The median segment is less than 15m.
2) The first and last segment of the uturn are one-way segments, with opposite driving directions.
3) The portion of those two segments closest to the median segment must be exactly parallel ±5°.

How is parallel measured? Look at the junction between the first segment and the median as the center of a compass, and the draw a straight line from the last portion of the approaching road through the junction to the edge of the compass. The angle it points to is the angle of that segment (we'll call it segment A).
Now position the junction between the median and the second segment at the center of the compass. Draw a straight line from the junction over the first portion of that last segment and continue to edge of the compass. That angle it's pointing at is the angle of the last segment (segment C). Now if those two angles are 180° apart (±5°), the segments are considered parallel.
The math, start with the higher of the two angles mentioned, subtract 180, then subtract the other (lower) angle. If the number you're left with is between -5 to 5, the segments are considered parallel.

Sent using Tapatalk for Android 4.4.2

taco909 wrote:Good to hear.

So basically, everything we've been doing to meet the last information (make the roads as parallel as possible through the junctions) has resulted in us creating proper geometry for this to work.

Nice that the angles of the cross street do not matter, so two 90s or a 45 and a 135 will still function.

Correct we had basically stumbled across certain criteria, which when they were met, by default meant the real criteria were also being met. Witynthe complete information I have now successfully prevented U-turns on curved roads, rounded medians, diagonal medians, and using micro doglegs.

I want to publicly thank the Waze staff for working closely with us to get to the bottom of this once-and-for-all. [emoji106]

Sent using Tapatalk for Android 4.4.2

voludu2 wrote:For the wiki:

1) The median segment is less than 15m.
2) The first and last segment of the uturn are one-way segments, with opposite driving directions.
3) The portion of those two segments closest to the median segment must be exactly parallel ±5°.

A convenient way to check for parallel is relative to the short segment. There should be no geometry nodes on this segment. then we need to check two angles:
X -- the angle between the incoming one-way segment and the short segment
Y -- the angle between the outgoing one-way segment and the short segment.
If the sum of these two angles is within 5 degrees of 180, then rule 3 is satisfied.

In the example below, angle X is 102 degrees and angle Y is 75 degrees. The sum is 177, which is within 3 degrees of 180. Therefore rule 3 is satisfied.

LilyPondAve_no_U.JPG

I was thinking something along those lines myself, but you wrote it already so nicely. Two items I want to incorporate to emphasize are;
1) when we say "opposite direction" we are not referring to one segment being A->B, and the other being B->A. It just matters that the physical direction of travel is opposite.
2) if the actual median segment is purposely not straight, we can measure the angles by creating a temporary new straight median segment connected at both junctions, check (adjust) the angles as necessary, then delete the new temporary segment. Since urge actual geometry of the median is irrelevant we could have round, or angulated medians as long as they are still less than 15m.

Sent using Tapatalk for Android 4.4.2