Hello, I'm interested in looking at "auto-tuning traffic signals".

I would like to support simulation of actuated traffic signals. I think a lot of behavior could be captured by including features found here like (1) skipping phases that aren't called, (2) making artificial calls to "recall" preferred phases every cycle, (3) specifying parameters like minimum green interval, maximum green interval, passage time gap before terminating a phase.

If implementing actuated traffic signals with the parameters listed above sounds good, I can move ahead with it.

Without having gotten my hands dirty yet, I just have a couple other thoughts:

1. In real life, I assume that actuated left turns are only possible when left-only turn lanes are present. (Then again, I suppose the technology exists to identify when cars are waiting for a left turn.) For now I'll just ignore this and look at what turns are requested?

2. I am not planning to touch phase generation or phase sequencing in this first piece of work. The only change to phase sequence will be the opportunity to skip a phase while running if it has not been called.

3. Multi-signal coordination with actuated signals will require a bit more thought - I need to look into how they maintain a synchronized cycle.

Later, as a separate task, I could look into optimizing phase durations for pre-timed signals and optimizing maximum green intervals and preferred phase for actuated signals.

Hi! Implementing those parameters would be a huge step forward. Feel free to file a separate issue to track things, track questions, etc. #91 is also relevant. As far as pointers to relevant code, bb3da1f and some of the explanations in the other bug are a start. You have to represent the new settings in the ControlTrafficSignal struct (with some boilerplate to save it in PermanentMapEdits), then update the UI in TrafficSignalEditor to modify this, and finally interpret the values in sim's update_intersection. Feel free to reach out as you dive in; I can also send a Slack invite if you prefer real-time chat.

In real life, I assume that actuated left turns are only possible when left-only turn lanes are present. (Then again, I suppose the technology exists to identify when cars are waiting for a left turn.) For now I'll just ignore this and look at what turns are requested?

Looking at the turns requested is appropriate for now. The phases that're heuristically generated should probably opt for a separate left-turn phase when a dedicated turn lane is there, but this isn't implemented yet.

3. Multi-signal coordination...

That can definitely be a later step. I recently got the signal editor to allow modifying multiple intersections at once, but we haven't yet figured out how to express an offset between two intersections. From what I understand, the offset is usually timed to match how long it'll take agents to travel between the two intersections. You have to pick a direction for the offset; it's not symmetric from i1 -> i2 and i2 -> i1. And this timing is usually based on free-flow conditions. I think figuring out all of this better would be relevant to multi-signal coordination too.