I really want to play with Godot RL Agents but I want to do vision based learning and that’s a bit down their road map. If anyone wants to contribute to add that feature I’d love you!

https://github.com/edbeeching/godot_rl_agents