Benchmarks
Simulation benchmarks
The simulation suites we run a policy against, grouped by the arm layout it is scored on.
Single-arm
one arm, one gripperOne arm and a gripper, on a fixed base or on a mobile one — the layout most manipulation policies are trained and scored on.
LIBERO-Spatial
libero-spatial
10 tasks · Franka Panda
LIBERO-Object
libero-object
10 tasks · Franka Panda
LIBERO-Goal
libero-goal
10 tasks · Franka Panda
LIBERO-10
libero-10
10 long tasks · Franka Panda
RoboCasa365
robocasa
PandaOmron · 18 kitchen tasks
Bi-manual
two arms in coordinationTwo arms that have to be coordinated — dual-arm rigs and humanoid upper bodies, where the timing between the hands is part of the task.