Computer0 31 minutes ago
coatol5 an hour ago
We wanted the ability to talk to and coordinate with it in real time.
We used a combination of imitation learning (we labeled thousands of hours of data using english language predicates) and RL to create an action policy for doubles in Melee.
Full write up (and model weights) on our website: https://www.frisson-labs.com/reyn
Built upon our previous work creating a competitive Melee agent: https://www.frisson-labs.com/faynt