Native keyboard support for CarRacing-v3 for HIL research

[hanjaek]

[Idea] Native keyboard support for CarRacing-v3 for HIL research

Hi! I'm currently working on a Human-in-the-Loop reinforcement learning setup using CarRacing-v3.

It would be great if there were built-in keyboard bindings (e.g., arrow keys) for controlling the car directly using gymnasium.utils.play, similar to how some PettingZoo environments support manual control.

This would help quickly prototype or collect human actions without manually redefining controls.

Is this something that could be considered?

[pseudo-rnd-thoughts]

Yes, this should be possible. The environments would need to implement get_keys_to_actions function that returns a dictionary of dict[tuple[str | int, ...] | str | int, ActType]