Skip to content

akolishchak/doom-net-pytorch

Repository files navigation

DoomNet

DoomNet is a ViZDoom agent trained by reinforcement learning. The agent is a neural network that outputs a probability of actions given only pixels from the screen buffer and set of game variables.
DoomNet is a 1st Runner-Up at Visual Doom AI Competition 2018.

DoomNet on a simple behavior tree

DoomNet track1, submission 0

Joint work with Bobby DeSimone

DoomNet's view is at left in the middle row
DoomNet track1, elimination round 2017

D3 Battle

DoomNet trained on D3-battle config

Health Gathering

DoomNet trained on health gathering config

Rocket Basic

DoomNet trained on basic rocket config

Basic

DoomNet trained on basic config