Hi, we have been working on our project based on IsaacGym Preview4 for some time, and we are satisfied with the performance boost it provides, thanks a lot for the amazing tool!
Currently, you have released Omniverse IsaacGym which seems to be a functional equivalent of IsaacGym Preview and it is RL-ready (GitHub - NVIDIA-Omniverse/OmniIsaacGymEnvs: Reinforcement Learning Environments for Omniverse Isaac Gym). Since then I’m curious about the differences between Omniverse version compared to Preview version:
- Is there a document that states the difference between these two platforms (performance, target scenario, etc.)?
- Is Omniverse IsaacGym gonna be a replacement for the Preview version? Would Previews be maintained in the future?
Looking forward to your replies. Thank you for your time.
Regarding 1), some key differences currently seem to include:
- Quaternion Convention and Joint Ordering
- Physics Parameters and APIs
- Task and Training Configuration Files
Compare 3. Transferring Policies from Isaac Gym Preview Releases — Omniverse Robotics documentation
@julian.esser is correct regarding your first question, and we will continue to work to make the transition from the standalone version to Omniverse Isaac Sim easier in the future. We’ve now released an updated set of examples in the OmniIsaacGymEnvs open source repository that you can compare against the equivalent standalone Isaac Gym samples to make the transition process easier.
Regarding your second question, we don’t intend to update the standalone Isaac Gym release further, though it’s not impossible that we’ll do some more minor bug fixes if required. All of our efforts are now focused on making the Omniverse incarnation the best way to do advanced RL going forward.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.