Synthetic data for robotic perception and control
Real-world robotic data can be expensive to collect and hard to label, but modern machine learning techniques are often data-intensive. As a result it would be advantageous to have the ability to learn robotic behaviors from cheap and easy to label data from a physics simulator. However, models learned in simulation often perform badly on physical robots due to the 'reality gap' that separates synthetic data from real-world robotics. In this talk we will discuss a simple and surprisingly powerful technique for bridging the reality gap called domain randomization. Domain randomization involves massively randomizing non-essential aspects of the simulator so that the model is forced to learn to ignore them. We will talk about applications of this idea in robotic perception and grasping.
Josh Tobin is a Research Scientist at OpenAI and a PhD student in Computer Science at UC Berkeley working with Professor Pieter Abbeel. Josh's research focuses on applying deep learning to problems in robotic perception and control, with a particular concentration on deep reinforcement learning, domain adaptation, and generative models. Prior to Berkeley and OpenAI, Josh was a consultant at McKinsey & Co. in New York. Josh has a BA in Mathematics from Columbia University.