Authors say GPT-6-Astra has competence, not practical intelligence
Original title: Has GPT-6-Astra Achieved Embodied Intelligence?
Authors report strong benchmark results with in-context learning, but real robots still face costly demos and time limits.
A report probes standalone GPT-6-Astra on manipulation, humanoid, mobile, and dexterous-hand tasks, allowing in-context learning from demonstrations and more execution steps. It also studies a costly but effective loop between the model as a high-level reasoner and a low-level controller such as π0.5. Real-robot tests show demonstration collection is expensive and longer runs are hard to trade for success. The authors say it has largely crossed embodied competence but not practical embodied intelligence.
Unconfirmed: The claim that GPT-6-Astra crossed an embodied-competence threshold is the authors' interpretation.
- The standalone evaluation covers 12 representative benchmarks.
- Real-world experiments used 8 robot configurations.
- Open problems include ICL generality, cost, verification, recovery, and long-horizon progress.
- A prototype benchmark withholds in-domain demos and scores agents on a running world clock.
From the original article
1 min read at huggingface.co
GPT-6-Astra demonstrates a level of competence in robotic manipulation and navigation that challenges the conventional paradigm of building specialized models for individual tasks and embodiments. In this report, we ask not only how capable GPT-6-Astra is, but what its capabilities imply for the field. First, we probe the limits of standalone GPT-6-Astra.
By SenseNova, Joint Research Team, Zhongang Cai · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.