/SAC

Implementation continuous SAC using pytorch in gymnasium

Primary LanguagePython

Soft-Actor-Critic

Papers

Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor : https://arxiv.org/abs/1801.01290

Algorithms

Alt text

results

Alt text

Hopper-v4 Walker2d-v4 HalfCheetah-v4
hopper video walker video halfcheetah video
Ant-v4 Humanoid-v4 Swimmer-v4
ant video humanoid video swimmer video

Dependencies

  • python == 3.10.0
  • numpy == 1.23.5
  • torch == 1.13.1
  • gymnasium == 0.26.3
  • mujoco-py == 2.2.0

How to Run

arguments

if you want see args,

python3 train.py -h

training

python3 train.py

evaluation

python3 eval.py

Acknowledgement

All credits goes to @pranz24 for his brilliant Pytorch implementation of SAC.