Real-to-sim-to-real

A training paradigm in robot learning that converts real-world data into a simulated environment, trains a policy in simulation, and then deploys it back to a real robot. The core challenge is the sim-to-real gap.