Reproduce “Sim-to-Real gap in RL: Use Case with TIAGo and Isaac Sim/Gym” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jaume Albardaner, Alberto San Miguel, Néstor García, Magí Dalmau-Moreno

This research evaluates the practical sim-to-real gap for robotic manipulation using a TIAGo robot, comparing NVIDIA's Isaac Gym and Isaac Sim simulators. Researchers demonstrated successful transfer of RL policies to the real robot for an arm movement task, while identifying specific discrepancies in control responses and dynamic behaviors between simulators and hardware.

Key insights

  • Simulator fidelity varies; Isaac Gym generally replicated the real TIAGo's multi-joint dynamic responses more closely than Isaac Sim in controlled tests.
  • Specific simulator dynamics and control implementations directly influenced the learned RL policies, leading to distinct movement patterns even for the same task and training parameters.
  • Isaac Sim exhibited difficulty in consistently respecting joint velocity constraints, contributing to larger deviations from the real robot's behavior.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.