[TechTalk] by OpenAI - Learning Robotic Dexterity

Dies ist ein vergangenes Event

65 Personen haben teilgenommen

Bild des Veranstaltungsortes


In this talk, Matthias from OpenAI will present the recently published results of their robotics team. They showed that complex and human-like dexterous manipulation can be achieved on a real humanoid hand. They trained their system only on simulated data using large-scale deep reinforcement learning. Still, the system generalizes to the real world, which they achieved through a technique called domain randomization. This talk will not only present their results but will also introduce the fundamental ideas and algorithms behind reinforcement learning as well as domain randomization.
A basic understanding of Machine Learning and simple Neural Networks is recommended.

Matthias Plappert graduated from the Karslruhe Institute of Technology with a Masters degree in Computer Science and works as a research
scientist for OpenAI, where he wrote his master thesis.

Blog: https://blog.openai.com/learning-dexterity/
Paper: https://arxiv.org/abs/1808.00177
Video: https://www.youtube.com/watch?v=jwSbzNHGflM

This event is a co-organization of Hack & Söhne and Karlsruhe.AI