Minding the Gap: What Actually Closes Sim-to-Real for Manipulation
Many tricks claim to close the sim-to-real gap. Under a fixed real-data budget, which actually help, and do they stack?
article · by @science_geek
Thesis. Under a fixed real-world data budget, a small amount of real fine-tuning closes more of the manipulation sim-to-real gap than heavy domain randomization alone, and the techniques partly overlap rather than stack; a controlled comparison identifies the most data-efficient recipe.