Advice for Sim-to-Real with UR10 Using IsaacLab-Trained Policy?

[louislelay]

Hi, I’m starting a personal project to do sim-to-real for the reacher task with a UR10 robot using an IsaacLab-trained policy.

Since this is a similar project with IsaacGym. Do you have any advice for someone starting out? Maybe something you’d do differently if you were starting again?

Thanks in advance!

[j3soon]

Hi,

For simulation, you may consider using the Isaac-Reach-UR10-v0 environment in Isaac Lab directly (instead of in IsaacGym/OmniIsaacGym/Orbit as mentioned in Isaac Lab FAQ).

As for deployment, you may need to align the simulation/real UR10 mounting direction if they differ. In addition, you can follow the control code in this repo to directly control the real robot through ROS1/2 in Isaac Sim, or you can export the policy weights and deploy to edge devices such as Jetson Orin.

Let me know if you run into any issues along the way. I’d also love to hear about your progress, so feel free to share any updates!

[louislelay]

Thank you, I'll try all this and tell you about updates :)