Models

Benchmark

GO-1 normalized task-completion reports

Author-reported GO-1 Figure5 research experiments with task-specific adaptation, not exact pretrained-weight measurements or independent reproduction. The metric is normalized task-completion score with fractional credit in[0,1], NOT binary success rate or percent. Thirty rollouts per task:10seen and20with variations/distractions, corresponding to10per scenario. Training description:learningrate2e-5,batch768,30Ksteps. Unknown run dates/seeds/intervals remain unknown. The planner-free ablation is not verified as the released Air checkpoint. Figure6 uses RDT for dataset comparisons and is excluded.

GO-1 / planner-free paper ablation

Protocol go1-planner-free-paper-ablation

GO-1 / task-adapted

Protocol go1-real-world-task-adapted

Submit a source or correction