M30.6 CONNECT THE MECHANISM
Learn in simulation and test what transfers
Learning by trial and error would take a real robot arm 23 days of nonstop practice. In simulation it takes minutes. The hard part is making the skill survive the trip back to reality.
LESSON OVERVIEW13 min lesson
Lesson overview
Learning by trial and error would take a real robot arm 23 days of nonstop practice. In simulation it takes minutes. The hard part is making the skill survive the trip back to reality.
What you’ll explore
- Estimate the real-robot cost of reinforcement learning, design rewards that are paired with hard safety limits, and explain how domain randomization, system identification, and staged testing close the gap between simulation and reality.
GO TO THE SOURCE
Original explanations, connected to the research.
Domain Randomization for Transferring Deep Neural Networks from Simulation to the Real World (Tobin et al., 2017)Sim-to-Real Transfer of Robotic Control with Dynamics Randomization (Peng et al., 2018)Solving Rubik's Cube with a Robot Hand (OpenAI, 2019)Learning to Walk in Minutes Using Massively Parallel Deep Reinforcement Learning (Rudin et al., 2021)Data-efficient Deep Reinforcement Learning for Dexterous Manipulation (Popov et al., 2017)Suggest a correction
A precise note can make an explanation better.
Choose the scene and describe what needs attention. Download a feedback file to share through a channel you already use. This page does not send feedback or connect you with a reviewer.