Hi,
I am aware that ORBIT/IsaacLabs provides photo-realistic and tiled rendering and I have also seen the throughput comparison between ORBIT and IsaacGymEnvs at the website (https://isaac-orbit.github.io/) for tasks such as training for of Allegro Cube, Shadow Hand etc.
I am interested in performing vision-in-the-loop RL. Are there any comparisons available between IsaacGym and IsaacLabs regarding camera throughput?
Earlier, I had noticed that on an RTX 4090, with a single RGB camera, using tensor api at 256x256 resolution, it is hard to spin up more than 4096 environments and it took many hours to train. Is this likely to be the case in IsaacLabs as well?
Thank you for your insights! Any directions or examples are most welcome.
Thanks