Visual referee update: Free kick signal detection

[ThagonDuarte]

Why? What?

The rule update for the 2025 season introduces a change to the set play of free kicks. Instead of sending the kicking team in the game controller message during all free kicks, now, during CornerKick, GoalKick, KickIn, and PushingFreeKick no kicking team will be sent in the game controller message. Instead, the head referee will give a visual signal to indicate the direction of play for the kicking team. This means a signal in the direction of the goal of the HULKs indicates that the Opponent is the kicking team. This signal can be seen below:

This expands our current visual referee implementation to now also detect this free kick visual referee signal.
We only try to detect the visual referee signal in KickIn and PushingFreeKick, since the kicking_team during CornerKick and GoalKick can be deduced from the current ball position and the teams GlobalFieldSide, i.e. whether their are home or away. During KickIn and PushingFreeKick, when Home the DefenderRight and MidfielderRight should look at the referee and do the detection. When Away, this is done by DefenderLeft and MidfielderLeft and the first two non-penalized Searchers that are not the keeper.

This PR only implements the detection of the signal. It does not yet apply the detection to the kicking_team in the FilteredGameControllerState, since the game controller currently still always sends the kicking team, as it has not been updated yet (see RoboCup-SPL/GameController3) (EDIT: It has been updated).

Fixes #1476

ToDo / Known Issues

• Implement behavior for looking at the referee for pose detection during set plays.
• Implement detection logic and filtering.
• Implement bevyhavior simulator scenarios for
  • Check behavior for visual referee during free kicks works correctly
• Make NAOs look further up to better see referee during detection

Ideas for Next Iterations (Not This PR)

• Handle not being sent a kicking team. This is tracked in Visual referee update: Kicking team filtering #1647

How to Test

• In play, trigger the four free kicks CornerKick, GoalKick, KickIn, and PushingFreeKick.
  • Check that during KickIn and PushingFreeKick only the correct defender and midfielder or searcher look at the referee and do the visual signal detection.
  • Check that the visual referee free kick pose signal is detected, filtered, and interpreted correctly.

[oleflb]

review not yet complete

[oleflb]

Why should Team::Opponent be default? I don't think it is a good idea to add this default impl

[ThagonDuarte]

True. I only did this, because our implementation of PathDeserialize for Option<T> requires Default and I changed the kicking_team to a Option<Team>. I don't think there is another way to specialize the implementation of PathDeserialize specifically for Option<Team>... Dunno what else to do :/

[oleflb]

hmm yes you are right about that. You can still do a manual implementation of PathDeserialize when Team does not implement Default. @schmidma I am a bit confused that PathDeserialize for Option<T> requires T: Default. Looking at the code It looks like a totally different value is deserialized in the None case

[ThagonDuarte]

I did not manage to manually implement PathDeserialize for Option<Team>, since adding spl_network_messages, which defines Team, as a dependency to the path_serde crate, creates a cyclic dependency 🫠.
I now simply removed the requirement for the Default derive for Option<T> by just not deserializing in the None case. This should be fine, since the value is a None anyway, right?

[knoellle]

Depends on what you mean by "fine". It changes the behavior.
If the previous value was none, we'd construct a default one and recurse with the deserialization.
That was a design decision we made at some point, though I'm not sure we depend on this behavior anywhere currently.

[oleflb]

what if the keeper is penalized? If 1 is penalized and rest of the team is not, 3 & 4 would be selected instead of 2 & 3

[ThagonDuarte]

This is intentional. The lowest non-penalized player number, in this case 2, is chosen as replacement keeper, which should not look for the referee, imo.