Skip to content

A Reinforcement Learning project that trains a PPO model to play the First Person Shooter game of ZDoom

Notifications You must be signed in to change notification settings

GitanshKothari/VizDoom

Repository files navigation

VizDoom

The reinforcement learning model learns to play the game of ZDoom - a popular First Person Shooter (FPS) game. This project was built using OpenAI's Gym environment wrapper for the game of ZDoom called ViZDoom. Implemented multiple models such as Deep Q-Network (DQN), Proximal Policy Optimization (PPO) and Actor Critic (A2C) using Stable Baselines.

A demo of one of the game environment - Basic - is shown below:

Vizdoom_basic_demo.mp4

Here is another demo of a different game environment called Deadly Corridor, where the enemies are set at level 2 difficulty:

Vizdoom_deadlyCorridor_demo_level2.mp4

And finally, a third demo of of the model at the same Deadly Corridor environment, but the agents are at a level 5 (max) difficulty:

Vizdoom_deadlyCorridor_demo_level5.mp4

About

A Reinforcement Learning project that trains a PPO model to play the First Person Shooter game of ZDoom

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published