Skip to content

Reinforcement Learning with Euclidean Data Augmentation for State-Based Continuous Control

Notifications You must be signed in to change notification settings

JinzhuLuo/EuclideanDA

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Reinforcement Learning with Euclidean Data Augmentation for State-Based Continuous Control

Setup

Please update the 3d files to your dm_control file.

  • To train agents
python train_qua.py pixel_obs=false action_repeat=1 task=quadruped_run agent=ddpg_rotate aug_ratio=4  seed=1agent=hpg 

train_xx.py: the run file for each task task=xxx : the task: 1.quadruped_run 2.reacher_hard 3.cheetah_run 4.cheetah3d_run 5.hopper_hop 6.hopper3d_hop 7.Humanoid_stand 8.humanoid_run 9.walker_run 10.walker3d_run agent=xxx: the method you will use: 1.ddpg_our: the original DDPG 2.ddpg_rotate: the DDPG + aug 3.ddpg_rad: DDPG + RAS 4.ddpg_guass: DDPG + GN

aug_ratio: The rotate ratio in the batch 0:no rotate. 1:100% rotate. 2:50% rotate 3:50% rotate 4:25% rotate

The results will be saved at ./exp

The code is adapted from "Continuous MDP Homomorphisms and Homomorphic Policy Gradient" by Sahand Rezaei-Shoshtari, Rosie Zhao, Prakash Panangaden, David Meger, and Doina Precup, presented at the Advances in Neural Information Processing Systems (NeurIPS) conference in 2022. We gratefully acknowledge their significant contributions to this field.

About

Reinforcement Learning with Euclidean Data Augmentation for State-Based Continuous Control

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages