Benchmark
LIBERO
Robot manipulation benchmark with 130 tasks across five suites.
LIBERO-Goal
Protocol LeRobot 400-episode protocol
Success rate
π₀ · π₀.₅
98 %
Publisher evaluation · Reported; not independently reproduced
VLA-JEPA · VLA-JEPA
98 %
Publisher evaluation · Reported; not independently reproduced
MolmoAct2 · MolmoAct2
98 %
Publisher evaluation · Reported; not independently reproduced
GR00T · GR00T N1.7
97 %
Publisher evaluation · Reported; not independently reproduced
LIBERO-10
Protocol LeRobot 400-episode protocol
Success rate
π₀ · π₀.₅
96 %
Publisher evaluation · Reported; not independently reproduced
VLA-JEPA · VLA-JEPA
93 %
Publisher evaluation · Reported; not independently reproduced
MolmoAct2 · MolmoAct2
96.6 %
Publisher evaluation · Reported; not independently reproduced
GR00T · GR00T N1.7
84 %
Publisher evaluation · Reported; not independently reproduced
LIBERO-Object
Protocol LeRobot 400-episode protocol
Success rate
π₀ · π₀.₅
99 %
Publisher evaluation · Reported; not independently reproduced
VLA-JEPA · VLA-JEPA
100 %
Publisher evaluation · Reported; not independently reproduced
MolmoAct2 · MolmoAct2
100 %
Publisher evaluation · Reported; not independently reproduced
GR00T · GR00T N1.7
81 %
Publisher evaluation · Reported; not independently reproduced
LIBERO average
Protocol LeRobot 400-episode protocol
Success rate
π₀ · π₀.₅
97.5 %
Publisher evaluation · Reported; not independently reproduced
VLA-JEPA · VLA-JEPA
96.5 %
Publisher evaluation · Reported; not independently reproduced
MolmoAct2 · MolmoAct2
98.25 %
Publisher evaluation · Reported; not independently reproduced
X-VLA · X-VLA 0.9B
93 %
Publisher evaluation · Reported; not independently reproduced
LIBERO-Spatial
Protocol LeRobot 400-episode protocol
Success rate
π₀ · π₀.₅
97 %
Publisher evaluation · Reported; not independently reproduced
VLA-JEPA · VLA-JEPA
95 %
Publisher evaluation · Reported; not independently reproduced
MolmoAct2 · MolmoAct2
98.4 %
Publisher evaluation · Reported; not independently reproduced
GR00T · GR00T N1.7
91 %
Publisher evaluation · Reported; not independently reproduced