Skip to content

A Minimal Example of Isaac Gym with DQN and PPO.

Notifications You must be signed in to change notification settings

sibocw/minimal-isaac-gym

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Minimal Isaac Gym

This repository provides a minimal example of NVIDIA's Isaac Gym, to assist other researchers like me to quickly understand the code structure, to be able to design fully customised large-scale reinforcement learning experiments.

The example is based on the official implementation from the Isaac Gym's Benchmark Experiments, for which we have followed a similar implementation on Cartpole, but with a minimal number of lines of code aiming for maximal readability, and without using any third-party RL frameworks.

Note: The current implementation is based on Isaac Gym Preview Version 3, with the support for two RL algorithms: DQN and PPO (both continuous and discrete version). PPO seems to be the default RL algorithm for Isaac Gym from the recent works of Learning to walk and Object Re-orientation, since it only requires on-policy training data and therefore to make it a much simpler implementation coupled with Isaac Gym's APIs.

Both DQN and PPO are expected to converge under 1 minute.

Usage

Simply run python trainer.py --method {dqn; ppo, ppo_d}.

Disclaimer

I am also very new to Isaac Gym, and I cannot guarantee my implementation is absolutely correct. If you have found anything unusual or unclear that can be improved, PR or Issues are highly welcomed.

About

A Minimal Example of Isaac Gym with DQN and PPO.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%