axisrobotics/Test-Dataset
AXIS held-out 20 — a cross-embodiment few-shot adaptation benchmark 20 tasks never seen in pretraining, 20 demonstrations each, LoRA adaptation, rollout evaluation. This release is the SPECIFICATION and the INDEX, not the demonstration data. It is published first and on purpose: everything here is what you need to render the benchmark on your own embodiment, and none of it depends on our video encoding being finished. Status: the task list is CANDIDATES. The learnability gate… See the full description on the dataset page: https://huggingface.co/datasets/axisrobotics/Test-Dataset.
render inputs: render_bundles.tar.gz
render inputs: render_assets20.tar.gz
render inputs: render_code.tar.gz
baseline: pi05_base + 20-demo LoRA on all 10 clean tasks, 500 dual-sim rollouts
drop the stale index: it predates quality ranking and the survey filter, so its attempt ids conflict with render_request/attempts (the authority)
spec: match the render request -- 10 active + 10 conditional, data awaiting re-render
spec: match the render request -- 10 active + 10 conditional, data awaiting re-render
render request: all 20 tasks, rig + appearance pinned, active/conditional marked
render request: pin object materials, floor and lighting too (meshes already match)
render request: ALSO pin appearance (texture + lighting), not just the camera rig
held-out gate: built corpora + splits for the remaining 6 clean tasks
eval inputs: AXIS-Render table texture bank
eval inputs: AXIS-Render table texture bank
eval inputs: axis_render_src.tar.gz
eval inputs: assets_heldout10.tar.gz
pilot: built 224px corpus + idle-filtered splits for 4 held-out tasks
drop the superseded render request; render_request/ carries the quality-ranked list
render request: 10 held-out tasks x 20 demos at the stage1_5k rig (read + upload here)
render request: 10 held-out tasks at the stage1_5k camera rig
held-out 20: per-attempt traces, goals and checker configs (corpus build inputs)
publish stage-2 LIBERO norm stats beside the build they describe
README: restore the repo license (mit), note the pre-existing stage-1 releases
held-out 20: benchmark spec, prescriptive camera rig, eval seeds, Franka index
remove transfer log; not dataset content
remove transfer log; not dataset content
publish 5k stage-1 artifacts (weights, schedules, quality, roots, ranges, norm stats)
publish round-1 stage-2 LIBERO build (2,000 eps / 338,575 frames / 128px / v3.0)
publish stage-1 corpus (5k camera_fixed, physics columns transplanted)
initial commit