TD3 algorithm in deep reinforcement learning

This work has been done as CS591 Deep Learning course project based on "Addressing Function Approximation Error in Actor-Critic Methods" paper.

Requirements

To run this algorithm you can make an virtual conda environment

conda create -n your_env_name python=3.6

and activate it by

conda activate your_env_name

You need to install some modules includes:

conda install pytorch torchvision -c pytorch

conda install -c conda-forge matplotlib

pip install pybullet

conda install -c akode gym

How to run?

To run this algorithm there are two ways:

First:

1- clone all files

2- open "TD3_training.py" and change env_name on line 39 to desired environment which is defined in PyBullet

3- run "TD3_training.py"

4- optimal policy will store in "pytorch_models"

5- to visualize the interaction of agent with environment you can open "main.py" file

6- in "main.py" line 33 change env_name to the same environment you train the algorithm on

7- you can see the video of the agent on "exp/brs/monitor" folder

Second:

1- clone all files

2- open "TD3.py" and change env_name on line 209 and 493 to desired environment which is defined in PyBullet

3- run the TD3.py file

Easy RUN!

On the other hand you can change the name of the environment in 209 and 493 lines and run "TD3_Colab.ipyn" on google colab!