Lilian Weng • 5/5/2019

Domain Randomization for Sim2Real Transfer

This technical article discusses the challenge of transferring models trained in simulation to real-world robots, known as the sim2real gap. It compares approaches like system identification and domain adaptation, then focuses on domain randomization (DR). DR involves training models across a wide variety of randomized simulated environments to create a robust policy that can generalize to the physical world, often with little to no real-world data.

0 comments

#simulation #Robotics #Reinforcement Learning