KunBB/RL-Simple-implementation-of-AWAC-algorithm

Reinforcement Learning : This project aims at implementing an advanced algorithm called AWAC, Advantage weighted Actor Critic algorithm , which is discussed in \cite{nair2020accelerating} This project will attempt to understand the advantages and explore disadvantages (if any) of this algorithm. The main goal of this algorithm is to accelerate online reinforcement learning using offline datasets, which makes it a very useful tool for using reinforcement learning more efficient. But this task is also an extremely complicated and difficult task. The approach discussed in the paper aims to navigate through this task by efficiently handling its challenges with accumulation of error while bootstrapping, stemming from data inefficiency and excessive conservative on line learning.

Jupyter Notebook