Skip to content

BEHAVIOR-1K capability coverage

What the suite actually exercises, by display tag. Thin rows are the interesting ones: a tag on one or two tasks is either genuinely rare or a sign we have not finished tagging.

100 of 100 tasks tagged.

Capability

100 task(s) carry a tag from this group.

Tag Tasks Coverage
Perception & Understanding 64 ▓▓▓▓▓▓▓▓▓▓▓▓▓
Planning & Reasoning 97 ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
Control & Coordination 100 ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
Feedback & Adaptation 54 ▓▓▓▓▓▓▓▓▓▓▓

Task Domain

100 task(s) carry a tag from this group.

Tag Tasks Coverage
Manipulation 0
Locomotion & Stability 0
Navigation & Exploration 37 ▓▓▓▓▓▓▓
Mobile / Whole-body Manipulation 100 ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
Interaction & Collaboration 0

Skill primitives

The benchmark's own annotation vocabulary, as recorded in skills:. Descriptive rather than normative — see the capability reference for the distinction.

Skill Tasks
attach 0
chop 0
close door 1
close drawer 0
close lid 1
hand over 0
hang 0
hold 0
ignite 0
insert 0
move to 1
open door 1
open drawer 0
open lid 1
pick up from 1
place in 1
place in next to 0
place on 0
place on next to 0
place under 0
pour 0
press 0
push to 0
release 0
spray 0
sweep surface 0
tip over 0
turn off switch 0
turn on switch 1
turn to 0
wipe hard 0