Switch to double-size field

[justbuchanan]

The league is moving to a larger field size. At the last competition, it was optional, but this year it will be required for everyone to use, so we need to adjust accordingly.

The constants that define our software's knowledge of the field sizing are in common/Constants.hpp and soccer/gameplay/constants.py.

Because the field in the shop is still the old size, we need to be able to work with both. It'd be nice if there was an option in soccer's menu for switching back and forth.

A couple other tidbits:

• The new field has four cameras instead of two
• At comp this year, they let teams practice using half of the field so that two teams could practice at once. I anticipate this being the same this year, although maybe less so since teams will want more practice using the whole field. However, we should be able to use either half of the field if that's an option, which will require listening on different network ports for SSL Vision data
• We should also change our simulator for this, but keep the option for the smaller size there as well

[barulicm]

Some of this functionality was implemented at competition last year by yours truly. At the very least, I know I added the ability to use either of the half-fields as an option in the menu bar. I think I added one for using vision for the full field, but can't remember. Either way, I'll take a look back through the changes I made and fix 'er up to use all the appropriate options. Shouldn't take but an hour.

[justbuchanan]

Sounds good :+1:. I remember you working on it at comp, but couldn't remember what all was done.

[justbuchanan]

One thought I had is that this info could potentially go into our config files. When we want to run, we could launch soccer with the appropriate config which would specify vision port to listen on and field dimensions. This would especially fit in nicely once the new config system is up and running.

Hard-coding it in constants.{hpp,py} wouldn't necessarily be bad, we just don't want our code to have to check for which field we're using before checking for the dimensions. We'd probably have to have it be a property on Processor or something.

[barulicm]

TODO:

• [x] Fix simulator. It's currently segfault-ing likely due to uninitialized Field_Dimensions objects.
• [x] Make FieldView update to new dimensions
  • [x] Fix rendering issues with robot tails
  • [x] Fix issues with mouse input
• [x] Add ability for simulator to switch between field sizes

[barulicm]

Alright, @justbuchanan, I think everything that needs to change for this is done. Both the simulator and soccer can now switch between single and double sized fields. C++ and Python constants all update when the field size is changed. Let me know if you spot anything I missed.

[justbuchanan]

Cool, looks good! Thanks for taking care of this.

A few questions/comments I had after reviewing:

• What's the benefit of having the const references and the getter methods in the Field_Dimensions class? They seem to essentially do the same thing.
• I think the play/pause/rewind/ffwd buttons in soccer disappeared.
• I didn't get a chance to investigate much, but it seemed like when using 2x mode with the simulator, clicking on the field to place the ball in some cases would make it disappear. The ball still showd in the simulator, but not in soccer.

[barulicm]

• Honestly, probably don't need both. It started with just the const references, but I had to add the getters for the python connections. Could / should probably just make those getters inline and get rid of the reference members. You might also know a better way of passing the Field_Dimensions object to Python, as I'm still not terribly familiar with all the possibilities of that API.
• They do seem to have become invisible. The same appears to be happening on the master branch on my machine. Can't find anything in recent git history that would've cause this, so I'm not sure where that's coming from.
• I encountered that a few times early on, but couldn't get it to happen very consistently. I'll take a closer look and see if I can figure it out.

[barulicm]

Relating to the third issue about mouse control of the ball, it seems to be coming out of the BallTracker, specifically, line 106:

static const Rect field(Point(-Field_Dimensions::Current_Dimensions.Width / 2, 0), Point(Field_Dimensions::Current_Dimensions.Width / 2, Field_Dimensions::Current_Dimensions.Length));

Removing the static qualifier (thus causing the field var to update on every frame), seems to fix the issue. I believe this has to do with the way we filter which observations to track. If the jump between the observations is too large, the tracker does a hit test against this field variable. If the field variable is still set to the smaller field size, observations outside of the smaller size end up getting ignored.

[justbuchanan]

• Sounds good. I think the compiler actually will inline the getters anyways regardless of whether you specify it or not. I think the cleanest way code-wise for the python stuff would probably be to just wrap the Field_Dimensions class with boost python and just pass a reference to python. I think the way you have it now though is best - stuff seems to get slow if there's a lot of back and forth over the language barrier, so just copying the values over is good.
• Oops, didn't even notice that. I'll look into it when I get a chance and see if I can get those buttons back.

[justbuchanan]

I got the playback buttons fixed in soccer. Not sure how I missed it, but they've been gone for quite a while... Everything looks good to me, so I went ahead and merged this into master.

[barulicm]

Cool. One last thing that needs to be checked is any place where we check positions for gamestate (ie. is the ball in the corner for a corner kick). I haven't looked at these yet, but if they're currently based on constants, they'll have to be redone.

[justbuchanan]

I think the only place this stuff happens is in the score() functions of the plays. Since you changed stuff to pass the new constants into python, I think it should work out as is.