Humanoids Daily

Daily news on humanoid robotics and physical AI

RoboCurve

2 articles · Last covered September 20, 2026

Published on
September 20, 2026
Reading time
3 min read
GPT-6 Astra Rarely Refused Unsafe Robot Commands in New Test
RoboHarm adds a safety test to Astra’s robotics results: 97 attempts and 60 completions across 100 unsafe-instruction trials, with only two safety refusals.
OpenAI Benchmark Robotics-Research
Published on
September 5, 2026
Reading time
6 min read
OpenAI’s GPT-6 Astra Hits 95% on Physical Manipulation Benchmark, But Precision Tasks Expose Familiar Bottlenecks
RoboCurve benchmarked OpenAI’s new GPT-6 Astra on real robot arms, revealing dramatic efficiency and spatial reasoning gains on gross manipulation alongside a persistent 10% plateau on millimeter-precision tasks.
OpenAI US Benchmark