go back

Robot Learning from Randomized Simulations: A Review

Fabio Muratore, Fabio Ramos, Wenhao Yu, Greg Turk, Michael Gienger, Jan Peters, "Robot Learning from Randomized Simulations: A Review", Frontiers Robotics and AI, 2022.


The rise of deep learning has caused a paradigm shift in robotics research, favoring methods that require giant amounts of data. It is prohibitively expensive to generate such data sets on a physical platform. Therefore, state-of-the art approaches learn in simulation where data generation is fast as well as inexpensive, and subsequently transfer the knowledge to the real robot sim-to-real. Despite becoming more and more realistic, all simulators are by construction based on models, hence inevitably flawed. This raises the question how simulators can be modified to facilitate learning robot policies which overcomes the mismatch between simulation and reality, often called reality gap. We provide a comprehensive review on sim-to-real research for robotics, focusing on a technique called domain randomization which comprises ways to learn from randomized simulations.

Download Bibtex file Download PDF