Learning dexterity

Abstract: We’ve trained a human-like robot hand to manipulate physical
objects with unprecedented dexterity. Our system, called Dactyl, is
trained entirely in simulation and transfers its knowledge to reality,
adapting to real-world physics using techniques we’ve been working on
for the past year. Dactyl learns from scratch using the same
general-purpose reinforcement learning algorithm and code as OpenAI
Five. Our results show that it’s possible to train agents in
simulation and have them solve real-world tasks, without
physically-accurate modeling of the world.

bio: Wojciech Zaremba is a co-founder of OpenAI (2016-now), where he leads
the robotics team, which is working on developing general purpose
robots via new approaches to transfer learning and teaching robots
unprecedentedly complex behaviors. He received his MS from Warsaw
University / Ecole Polytechnique (2013), and his PhD from New York
University (2015) under Prof. Rob Fergus and Prof. Yann LeCun. During
his PhD, Wojciech spent one year at Facebook AI Research and one year
at Google Brain. Wojciech’s PhD and subsequent research contributions
have involved using neural network techniques to get computers to
learn sophisticated algorithms from raw data: his contributions
include the applying translation models to computer programs and
building first neural Turing machines with discrete actions. Moreover,
he worked on discovery of adversarial examples, improved training of
GANs, and development of OpenAI gym. His current work is dedicated
towards the development of the General Purpose Robots.