Study evaluates GPT-6 Astra as a robot policy alongside PI0.5
A study evaluates Astra as an embodied policy: on ten RoboDojo tasks PI0.5 scores 24%, Astra 37% and the two together 62%, with Astra correcting only 14.4% of control steps.
Great work that measures GPT6 Astra as an Embodied Policy by Yu-Mool Shu and Lipxin Zheng. It further proves with numbers that VLA handles dexterous movement, while the General Model oversees supervision, error correction, and replanning. Interesting numbers: For the average performance across 10 tasks in RoboDojo, PI0.5 gets 24%, Astra gets 37% but together they achieves 62%. But for RoboLab, Astra itself gets 98%. As for the contribution, GPT 6 Astra is correcting only 14.4% of the executed control steps and the remaining 85.6% following the actions by PI0.5. Link: https://anonymous-report-421.github.io/public-website/?lang=en&view=1