Skip to content

Latest commit

 

History

History
30 lines (13 loc) · 919 Bytes

README.md

File metadata and controls

30 lines (13 loc) · 919 Bytes

Deep Reinforcement Learning : A3C | PPO | Curiosity applied to Super Mario Bros.

This is the final project for the Reinforcement Learning course at the MVA Masters 2018/2019.

The project was done by Amine Sadeq & Otmane Sakhi, You can check the final project paper : ["Exploring Deep Reinforcement Learning with Super Mario Bros"] in this repository.

It explores A3C and PPO algorithms and combine them with an intrinsic reward based on curiosity.

A3C curiosity vs no curiosity in Dense reward

dense_cur

A3C curiosity vs no curiosity in Sparse reward

sparse_cur

A3C curiosity vs PPO curiosity in Dense reward

dense_ppo

A3C curiosity vs PPO curiosity in Sparse reward

sparse_ppo