GPT-6 Astra Autonomously Trains Dexterous Hand to Spin a Pen in 36 Hours

In the fourth test of GPT-6 Astra, Walter Zhu showed that a single prompt was enough for the agent to autonomously run a 36-hour reinforcement learning pipeline in Isaac Lab, building a 3D pen model and training a Sharpa dexterous hand to spin a pen.

2026-09-17 ~ 2026-09-18 · 3 related posts

• GPT-6 Astra Autonomously Trains a Pen-Spinning RL Policy in 36 Hours — JasonMa2020 · 2026-09-17
• One prompt: GPT-6 Astra trains a dexterous hand to spin a pen over 1.5 days — FinanceYF5 · 2026-09-18
• Agent runs autonomously for 1.5 days to train a dexterous-hand pen-spinning RL policy — SatelliteNetSec · 2026-09-18