Conclusion

This session provides time for remaining questions, continued experimentation, and a conclusion for this learning path.

Learning Path Summary

What You Accomplished

• Learned why simulation matters and what the sim-to-real gap is

• Built and standardized the physical lightbox workspace to match the sim task

• Got hands-on time with the SO-101 robot and LeRobot tools

• Applied Strategy 1: Domain randomization with teleoperation

• Explored NVIDIA GR00T, vision-language-action models

• Evaluated policies in simulation and on the real robot (sim-to-real gap)

• Applied Strategy 2: Co-training with real data, deployed to robot

• Applied Strategy 3: Cosmos synthetic data augmentation

• Explored Strategy 4: SAGE + GapONet (actuator gap estimation)

The Four Strategies We Covered

Strategy	Approach	Key Benefit
1. Domain Randomization	Vary simulation parameters	Robust to physics variations
2. Co-training	Mix sim and real data	Better real-world distribution
3. Cosmos Augmentation	Synthetic visual diversity	Robust to visual variations
4. SAGE + GapONet	Measure and model the gap	Targeted actuation fixes

Key Lessons

1. The gap is real — simulation success doesn’t guarantee real-world success

2. Multiple strategies combine — no single approach solves everything

3. Measurement enables improvement — SAGE shows you where to focus

4. Iteration is essential — systematic improvement beats one-shot attempts

5. Documentation matters — recorded observations guide decisions

Resources

Courses

• Getting Started with Isaac Lab - Transferring Robot Learning Policies from Simulation to Reality

Documentation

• LeRobot Documentation

• Isaac Sim Documentation

• Isaac Lab Documentation

• GR00T Developer Guide

• SAGE Repository

Community

• Hugging Face Discord

• NVIDIA Developer Forums

• LeRobot GitHub

Papers

• The Reality Gap in Robotics: Challenges, Solutions, and Best Practices

Conclusion

Congratulations on finishing this course “Train an SO-101 Robot From Sim-to-Real With NVIDIA Isaac.”

We hope this will enable and inspire you to keep learning and practicing your skills in Physical AI!

Feedback

Taking a few minutes to fill out our survey gives us valuable feedback to improve the course for future participants.

If you have any feedback, suggestions, or ran into issues, please visit this survey.