Comment by __MatrixMan__ | Hacker News Reader

Comment by __MatrixMan__ | Hacker News Reader

I spent two years building a module for use with intersection signal controller firmware (makes light turn red, green, yellow). The goal was to make better use of all of this extra data that we had. Sensors have improved, and all of this V2I stuff should really complete the picture. I poured my soul into that project. I really thought a lot about extensibility and ways to make it easy to accommodate new data sources. Here we had a chance to not shoot ourselves in the foot like we did last time these algorithms were visited, back in the 80's.

And then we discovered a memory leak in some vendor's V2I component which was supposed to slurp vehicle position data from DSRC and feed it to my controller. We couldn't get it to run stably in the cabinet, so the whole thing just kind of... stopped. Then they put me on some pointless project, so I quit. This was 2017.

Last year, I ran across somebody on youtube demoing my project. I got in touch, just to say hi, and he asked me for tech support--he didn't know how to configure it. I said:

> There's a readme on the SD card, plus an SDK that should get you started writing modules for various detector types.

He said:

> What SD card? I see what looks like a memory chip, but it's soldered onto the board.

They had told me I was building a research platform, and then when it finally ends up in the hands of a researcher, it turns out they've changed the hardware to make it useless for research. I was so angry. They never wanted to change anything, they just wanted some intellectual property that they could use as a bargaining chip in some way that had no impact on the intersection whatsoever.

This "V2X" business appears to be the updated branding, and it's a good idea at its core, but from what I know about the people implementing it, I'm not going to get very excited about how it'll turn out.

Is this why I have to wait at intersections with no incoming traffic for 2 minutes only to have a platoon of cars show up so that they can all stop and I can go

Cross like a New Yorker.

It's hard to say without looking at the controller config. Likely the main corridor is running mostly "coordinated" meaning that it's ignoring sensor input and just doing whatever timing is programmed for that time of day. Then there's a little gap in the coordination that only triggers if the sensor sees a vehicle.

Older controllers use line voltage for their clocks, so they drift much more quickly than you'd expect from, for instance, a wristwatch.

So I'd guess that you already had a rather small window in the coordination during which you can get service based on your sensor, and then the clock drift has put that window in the exact worse spot.

But yeah, changing the paradigm from presence-based to trajectory-based was the whole point of the project. If we could make these things trajectory-aware then there would be no need to manually coordinate them and then have that manual coordination fall out of sync (even with an accurate time source, you've got weather and construction and all kinds of other things that throw it off).

We seem to use fairly modern systems out here. They do appear to be programmed in a ripple type green wave progression depending on congestion. You'll hit one or two reds for every five or six intersections.

The intersection I have in mind has a left turn on it. The left turn is not timed and will only cue under two conditions: at the opening phase to main traffic, and if there is no oncoming traffic for the initial ten (or so) seconds. Otherwise it waits until the entire phase is complete (with a 120 second timeout) before it resets. At night time it will cue almost immediately.

One question that I have though, and you'll have to pardon my ignorance. Would it be efficient for intersections to always be red until a vehicle approaches? Then only allow X number of vehicles before turning red again. Basically turning the traffic signal into an "advanced" stop sign, with the expectation that you stop by default. Unless the system detects there is no cross traffic. Then you can have much more rapid phases and it protects against folks racing to the next green (red when they arrive) light.

As a software engineer who worked in support of civil engineers 7 years ago, I'm not exactly an expert myself, just picked up bits of domain knowledge on the job. And yeah, what you're describing re: left turns makes sense. I know there's a hybrid mode where you only serve the turns if there is demand, but otherwise you're running coordinated--but I am not sure about the nuances re: how the interplay of those modes actually works.

I expect that there are certain conditions which would indeed benefit from what you're describing. Especially if the signals could inform one another that a platoon is heading that way. On a straight shot under light conditions you might end up having to stop only once, after which you're now on a path that was coordinated precisely for you.

I've always wanted to run a traffic simulation to see how well that type of configuration would work. Because if you look at your car's trip computer the average speed from stop and go traffic tends to hover around half the prevailing speed.

Traffic signals in holland seem to operate similar to how I described. I'm just not sure how that would translate to somewhere like the US where you have much higher traffic volumes and faster (crazier) drivers.

This. I can't count the number of times I've been at an intersection with lights, waiting to turn but being forced to wait because the arrow is red - even though there is NO competing traffic from any direction. Behind me, cars start queuing also waiting to turn, and I'd put money down that the same thought occurs to them. Then, in the distance I start to see a wave of cars approaching in the oncoming lane. I look at them. I look at the lights. I look at them again. And I know what's going to happen next. Lo and behold! Just as the oncoming traffic approaches, the lights turn red for them and the arrow turns green for us.

It's pretty hard to believe its 2024 when this happens.

Reply on news.ycombinator.com