What is rl_games?


I used to use rl_games in IsaacGym, but I thought it was part of IsaacGym. What are rl_games? Why is IsaacGym using rl_games?

I don’t know much yet but I think rl_games is where the implementations of the deep RL algorithms are from. I’m currently looking how to exchange those policies with some of my own.

The rl game is not part of isaac gym.

If you have used Isaac Gym versions 1.0 and 2.0, you can find there are two reinforcement learning algorithm libraries, one coded by Nvidia, and the other is the rl game.

But in version 3.0, only the rl game is left, and the other one is discarded. But this shows that the algorithm library can be replaced by yourself.

I think the reason they chose rl game is that it has good performance and is lightweight.
It is appropriate to provide it in the preview version of Issac Gym, users don’t have to download and install and configure too many things.

Here is the official explanation, you can refer to it.

So that’s it.

There was a description that I was interested in.

For maximum training performance a preliminary installation of Pytorch 1.9+ with CUDA 11.1 is highly recommended:

I thought 1.8.2 was the best, but it might be a mistake.