Soft Actor-Critic implementation

This repo contains a Soft Actor-Critic implementation compatible with gymnasium environments. Furthermore, it contains a gymnasium environment linked with SPQR Team framework for the creation of KeyMotionFrame kicks.

Overview

main_sb.py contains a call to SAC implementation by stable-baselines3 applied on Pendulum-v1 environment.

main_SAC.py explores my custom implementation of SAC algorithm on Pendulum-v1 environment.

main_RoboCup.py calls my custom implementation of SAC algorithm on custom gym compatible RoboCup environment (RoboCup_env.py)

The folder trained_models contains some trained models for different environments (according to their names).

The folder logs contains tensorboard files which show the logs of the trainings.

The folder src contains all the implementation of my custom SAC algorithm.

Results

NOTE: WORK IN PROGRESS