Reverse-Frogger-A3PS

This repository represents the implementation of Influencing Reinforcement Learning through Natural Language Guidance.

Prerequisites:

1. Windows 10
2. Python 3.6
3. pytorch 1.2.0
4. ml-agent-0.8.1:

ml-agents-0.8.1 download link: https://github.com/Unity-Technologies/ml-agents/tree/0.8.1

After downloading, install ml-agents-0.8.1 following the section "Installing mlagents" in the link: https://github.com/Unity-Technologies/ml-agents/blob/main/docs/Installation.md

Some setup requirements for mlagents could be found: https://github.com/Unity-Technologies/ml-agents/blob/9602a8bef9f389964b1f5c1002217d02af54d191/ml-agents-envs/setup.py#L24

Preprocessing:

After installing mlagents in ml-agents-0.8.1 folder, all codes from the "ml-agents-0.8.1" folder of the repo should be put inside of it.experience_driven_agent.py represents the coding for Experience Driven agent and a3ps.py is the code for A3PS architecture. The "Advice generator" folder contains the code for Advice Driven agent.

Training:

As an example game environment, you can download the game from the link: https://drive.google.com/file/d/1eRX32aOssgBPYm2YOcXxAknP23XaHUe5/view?usp=sharing

The downloaded windows build is required to be put inside the ml-agents-0.8.1 folder.

To train the agent, run:

$ python ppo+advice_rl_frog_max.py