ArmnetBench v0.1

ArmnetBench v0.1 The ArmnetBench v0.1 core benchmark contains 50 human-teleoperated reference trajectories per task plus 2,518 evaluation trajectories from 7 policies trained or fine-tuned on those reference datasets, across 8 single-arm and 4 bimanual tasks performed on the low-cost SO-101 robot arm. It was collected using the Armnet arm farm. The full release contains 3,718 labelled episodes: the 3,118 core benchmark episodes plus 600 episodes from extra runs. The extra runs… See the full description on the dataset page: v01 robometer.

Source: Hugging Face Hub (armnet/armnetbench_v01_robometer). Metadata imported from the dataset’s Hub tags.