Reproduce “Has GPT-6-Astra Achieved Embodied Intelligence?” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
SenseNova, Joint Research Team, Zhongang Cai, Ziang Cao, Tailai Chen, Han Chong Chua, Linjun Dai, Hanming Deng and 54 more
GPT-6-Astra evaluates a general-purpose model on robot tasks by preparing a frozen knowledge package from demonstrations and acting with feedback. On RoboDojo, with in-context learning, it reaches 27.6% success at the standard 1× action-step budget and 81.9% at up to 10×. These benchmark results do not establish reliable or efficient real-world operation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.