XNews

GPT-6 Astra Trains Robotic Hand Dexterity in Simulation

GPT-6 Astra Trains Robotic Hand Dexterity in Simulation

OpenAI's GPT-6 Astra is training a robotic hand in simulated environments to perform tasks like pen spinning, Rubik's cube solving, and hammer use. The Rubik's cube task used 4,096 parallel simulations and required about 4.2 GPU hours. The goal is to transfer these skills to the real world.

Source posts