User:Tonestertm/Box intersections in depth

From Wazeopedia

< User:Tonestertm

Revision as of 11:25, 23 January 2015 by Tonestertm (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Place holder

Ok, going WAY back to the thing about I-10, Atlantic, and Hellman... If you see something with my name locked at 6, PLEASE direct any questions to me first, and wait for an answer before bringing in any other L6 opinions which might alter things before I've had a chance to weigh in.

These are pretty much exclusively places where things have been set in a way that other editors, particularly visiting ones, even at champ level, might not be able to understand the reasoning for without a bunch of study or explanation.

Pretty much anywhere you see a 6 lock, that means that there's something there the particular editor wants to hear about before there are changes. It's not exclusive to me. Honestly, the same goes for 5 locks in places where the roads would otherwise be lower.

We don't have anything 5 locked in LA by basic minimum rules such as the way NYC does. If there are components of an intersection locked above 3, it usually means someone has spent time tweaking, and that there may be something not immediately apparent about why it was set up that way.

Specifically, regarding Atlantic and Hellman, I apparently made a mistake there! OUCH. I know that at some point I very specifically restricted the turn from the special one-way SB Atlantic segment to EB Hellman, but I somehow flipped it to allowed before I walked away. Maybe thought I had the other one selected. The reason for those segments is so that the lefthand branch of that ramp can get a "Turn Right on Atlantic Blvd" prompt, but not then be allowed to turn onto eastbound Hellman.

Regular SB Atlantic traffic should, however, be allowed to turn left there. The reason for the one-way segment north of the ramp junction is just to insure the Turn Right prompt from the ramp. Without it, Waze would see the turn as best continuation.

It could be just a stub, but I made it connect to the junction above just for a cleaner look, and to help keep Validator from screaming about it and attracting unwanted attention. The negative elevation helps keep it from making the regular Atlantic look funny in WME, live map, and the client.

Ian Langford oh, ok See, it was still making the regular atlantic look funny in WME despite that Which is why I saw it in the first place

Otto Plunkett Was it the elevation highlight, or just the general funniness of it?

Ian Langford no, I saw one-way arrows

I only had lock 5 highlights on for some reason

Otto Plunkett Two way could work there, but in discussion with Pesach, I have been kinda convinced that one-way is safer to prevent some of the crazy oddities of Waze routing which shouldn't ever happen, but do.

Anything else on Atlantic/Hellman that bugged you? Or the ramps?

Ian Langford eh, I think there's a ramp there that needs naming, but it isn't critical

Otto Plunkett I'm pretty sure the south side of the freeway is good, but I DO miss things sometimes. I make no claim to perfection.

Ian Langford it was north side initial offramp if I remember correctly

Otto Plunkett Ok, regarding bow-ties. Waze regularly and happily routes U-turns at bow ties. There is no penalty. It's merely a sharp left turn. There has been some speculation that turns in excess of 150 degrees were penalized somehow, but I have not found this to be true in testing, and I think that we got word from Waze that it was not the case. If you hear that from someone who should know, like Marc or Pesach, then I really want to hear the details. It's of great interest to me. North side, I don't remember if I actually went over the ramps the way I like to.

Where roads are narrowly divided, and a physical U-turn does not consist of a double-left, the prompt for a bow-tie U can actually be less confusing than that of a box intersection.

Which brings me to box intersections, which are, until Waze fully implements the fabled Junction Box, the scourge of good routing in my opinion. The are BAD BAD BAD in the way they deal with speed/turn data. I shall explain.

Let me preface this by saying that the addition of left turn AGCs in one form or another can partially or completely negate the badness. As much as AGCs annoy me.

Ian Langford Oh, is it left turn slowdown? Paul mentioned that

Otto Plunkett What Paul said, but I want to explain it differently, because there may be more to it than what you took away. Forgive me if any of this is telling you what you already know. I'm saying this in the main MapRaid hangout for a reason. Others can read it here.

Ian Langford Ok. Best to get a good understanding

Otto Plunkett Ok, Waze has two kinds of speed data stored for segments. Part of it relates to the segment itself, and is portable, meaning that you could take that segment, bend it, shape it, change its road type, even relocate it to another road and rename it, and that data would, for a time, until overwritten by new data, go along for the ride.

This portable speed data has to do with the time it takes to get from one end to the other. Segment length is taken into account, and there is some correction done if the segment is stretched, merged with another, or cut in pieces. The portable data is reportedly very precise as to time of day and day of week.

Then there is the non-portable data, meaning speed/turn data, which relates to the different elapsed times between entering the segment and leaving it TO another specific segment, be that left, straight, or right.

This data is junction dependent. When you pull a segment away from a junction, it is gone. There is no how-long-to-turn-left-from-this-segment data when it's not connected to another segment to which it can turn left. And it's not a matter really of left versus right, but of this segment ID to that segment ID

If the turn angle from one to another was set as a straight continuation one day, and a left turn the next, by an editor, Waze would still see it as the elapsed time from entering the one to entering the other. (was changed)

Ian Langford So If you pull away, reconnect, save, is it lost? Because that was confusing me earlier

Otto Plunkett I believe so, though WME is a little weird about what it keeps/discards between saves. Always better, if you want to preserve segment to segment data, to move the junction, rather than even momentarily separating segments. If you save while disconnected, then it is definitely lost.

Tony M If the segment ID number changes, you know you've lost history.

Otto Plunkett That is what we call rebuilding a junction. Pull all segments away, save, reconnect. Fresh turn data must be made by Wazers, and corrupt existing data is gone.

Ian Langford Ok. How often must that be done?

Otto Plunkett Never need to rebuild a junction unless there is something absolutely crazy going on, like Waze refusing to route an allowed turn.

It's a fallback routine when nothing else has worked. Kinda like reinstalling an OS.

Ok, so what about box intersections (without AGCs)... When you are waiting to proceed at a box intersection, whether you are turning left, proceeding straight, or turning right, that waiting is done behind the first junction (with the exception of mid-intersection waiting for a left).

Ian Langford Right

Otto Plunkett So Waze can tell which traffic goes right, but it cannot tell straight traffic from left traffic. Both straight and left proceed straight from that first junction.

Most of the Wazers going through the second junction are not going to wait. So, at the first junction, people waiting to make lefts corrupt the straight data by not moving, and people going straight pollute the left turn data by moving fast.

At the second junction, where the turn transition happens or not, most Wazers are already going.

Ian Langford Right. I understand. So lefts seem fast and straights seem slow.

Otto Plunkett Exactly

Ian Langford Relative to reality

Otto Plunkett This is one of the reasons why Waze sometimes will route you through a horribly long left turn, or avoid an intersection which moves quickly going straight.

Ian Langford And a bow tie avoids this.

Otto Plunkett It does, because it allows Waze to see every possible out-route from the intersection at one junction.

Ian Langford Right. Because there is no interfering segment

Otto Plunkett If you add AGCs, then Waze uses the speed data specifically for those. AGCs that start back before an intersection can have their own speed data pollution issues.

Sometimes they are necessary for early prompts, but if there are very long lines of waiting cars, and some of them are stopped back on the main segment behind where the AGC splits, Waze can get corrupt data for the mainline there. So we try not to use the early prompt type AGCs except where the turn pockets are crazy long and need to get you over there well before the intersection.

Bow-ties look weird in the app, and can create MPs at wide intersections. They also can confuse people with their single U-turn prompts in places where the real world seems to have a double left.

There's a special kind of box intersection setup, sometimes called "cross-diagonals" which handles right/straight/left at the first junction, provides better GPS tracking of the driver's movement in comparison to the segment layout, allows complete control of allowed or disallowed lefts and U turns, and is more or less invisible in the app when higher road types are used and the intersection is not extremely large. or "box diagonals"

I used to really dislike this approach, because it looks like a tangled bunch of spaghetti in WME, but... I'll give you an example in a minute.

The box diagonals method is tricky to set up. NOT for beginners. And I'm not really comfortable with proliferating these intersections all over, as new editors tend to try to learn by example, but don't catch the nuances.

It requires doglegs and an elevation difference where the segments cross.

BUT, it uses only the four junctions already included in the box, so there aren't that many more data points for the routing server to deal with.

With the doglegs set properly, there is a prompt at the first junction, and none at the second. They double in function for opposing left turns.

So only two are needed to cover all four lefts.

Ian Langford So when SHOULD this be used?

Otto Plunkett Well, that's the fun part... We are waiting for Junction Box. How long? No one knows. Do we want to create a ton of these?

Ian Langford probably not

Otto Plunkett Do we want people out there trying to set them up and doing them wrong?

Ian Langford definitely not

Otto Plunkett I have been saving them for specific places where a bow-tie just seems to pull everything too far to the center of a huge intersection, or where I believe that straight versus left turn data is causing major issues. There is one more advantage to these. WHEN Junction Box finally comes, it's extremely easy to just delete the two cross segments. Much easier than opening a bow-tie.

Anyway, the box at Rosemead and Las Tunas isn't going to inhibit my sleep. I just wanted all this understanding to be out there. You seem to me the kind of editor who likes to fully get it. So I bothered.

Ian Langford yes I have learned a lot from this

There are few places I could use this back home, since we have so few split roads

  • divided roads

but it's good to know for MapRaids

Paul Yeatts Thank you Otto, very thorough ?? Love the glasses effect.

Otto Plunkett I know that Marc is ok with this setup, because he added the first one to the LA map. I'd check with the local authorities before constructing them all over.

Paul Yeatts And Ian, as I am privy to most of what Otto has explained, any issue with the work done i will take the credit on not informing or mis-guiding, but you do good work.

Otto Plunkett Jesse kinda got a funny look on his face when I mentioned them the first time, but he seemed to warm a little to the idea once he had checked out all the possible turns and respective prompts. Let me know if you want any tips on building these things.