Agent runs autonomously for 1.5 days to train a dexterous-hand pen-spinning RL policy
SatelliteNetSec · x · 2026-09-18
In a fourth test of what's called GPT-6 Astra, walterzhu8 gave an agent a single prompt: implement pen spinning on a Sharpa dexterous hand using Isaac Lab for RL training, create the pen mesh, and freely search the web and download papers as needed. The agent ran autonomously for a day and a half — including training the policy itself — and delivered a trained RL policy plus a visualization video, with credit to student Chengyang Li. LiveOverflow's reshare quips that "pen testers will soon lose their job," underscoring how far long-horizon autonomous agents have come on real engineering tasks.
Related event: GPT-6 Astra Autonomously Trains Dexterous Hand to Spin a Pen in 36 Hours(3 posts)→
More from coding & agent
- Cohere Labs agent team-construction method cuts multi-agent costs by 70% — Cohere_Labs · 2026-11-02
- DeepSeek design doc: a person can only follow 3-5 agent sessions at once, attention is the real bottleneck — heyshrutimishra · 2026-11-02
- Software fundamentals are key to getting good results from AI coding agents — bendee983 · 2026-11-02
- Open-source LLM regression gate digline hits 1.0: quantifying judge noise before calling a regression — alexpran · 2026-11-02
- PR adds patient retry logic for model capacity errors in OpenAI Codex — bytebot · 2026-11-02
- Perplexity open-sources WANDR: 500 research tasks, 170,495 evidence points, URL-rechecking grader — thisguyknowsai · 2026-11-02