Note, you may have to adjust --train_for_env_steps to a suitably high number as the experiment will resume at the number of steps it concluded at.
This bot is a killer
I have now trained over 10 Doom models. This one is definitely the most powerful/dangerous one so far. It unlike the others has learned a variety of techniques which make it hard to kill.
It seems to prefer the shotgun and sometimes minigun - when it has either it will often fell 3 or 4 bots in quick succession. When in a tight spot it will swivel around quickly and shoot.
I suspect it would be quite challenging to beat in a game.