Has GPT-6-Astra Achieved Embodied Intelligence? A 12-Benchmark Evaluation
sensenova · hf · 2026-10-19
A joint team (SenseNova, ACROB, Ropedia, NTU Singapore, Univ. of Tokyo) evaluates GPT-6-Astra's embodied capabilities across 12 benchmarks spanning tabletop manipulation, humanoids, mobile platforms and dexterous hands, achieving unprecedented results under a unified protocol allowing in-context learning from demonstrations.
The paper identifies key gaps: demonstration cost and execution time, physical-state verification and recovery from irreversible failures, and long-horizon progress preservation. It finds that close-loop coordination between a high-level "System 2" reasoner and a low-level "System 1" controller is costly but effective, and real-world deployment experiments reveal substantial practical gaps.
Related event: GPT-6-Astra Embodied Evaluation: Strong Benchmarks, Real-World Gaps(4 posts)→
More from Embodied
- Robot brain inference cost fell 84% in two months, from $2.36 to $0.38 per task — victor_explore · 2026-10-19
- NORINCO unveils Fuxi, an anti-riot humanoid robot that mirrors its operator's movements in real time — CyberRobooo · 2026-10-19
- HexaShell: VLM Coding Agents Drive Dual-Arm Robots, SOTA on RoboDojo — HexaFuture · 2026-10-19
- Amazon FAR's StopWatch: Explicit Goal Grounding Brings Object-Goal Navigation Near-Expert Success — Amazon-FAR · 2026-10-19
- Robot can now recover its balance after a 0.1 N push — _Stocko_ · 2026-10-19
- Chinese blogger builds working electric hoverboard with 10 motors and real-time AI voice coaching — churchkey · 2026-10-19