I am super curious about using and fine-tuning smaller vision-language-action style models! There are also some interesting RL projects out there focused only on PvP: https://github.com/Naton1/osrs-pvp-reinforcement-learning