RoboDojo execution
GPT-6 Astra · Stack the three blocks with different textures · Layout 0, seed 0
Official result: failure. 67 robot decisions, 70 model calls. Evaluation valid; evidence sealed.
Execution video
Original recorded cameras, synchronized to simulation time. 22.1 seconds; model waiting time is omitted.
Full process
6 minutes 43 seconds. Recorded observations, model rationale excerpts, actions and outcomes. Card durations are edited for readability; no rationale is invented.
Run identity
robodojo-local-local-gpt6-two-tools-3468098576e4--20260914T142831Z
Source: 7555857. Two-tool policy; SAM3 was verified separately and was not used in this attempt.