Counting on RoboMME Counting benchmark leaderboard · SOTA2 Research

Counting on RoboMME Counting

96Bin Fill Success RateHuman Performance

Evaluation Results

Method	Links
Human Performance2026.03		96	100	80	78
MME-VLA (SimpleSG)2026.03		85.78	99.78	100	44.67
MME-VLA (GroundSG)2026.03		85.78	100	100	49.67
MME-VLA (SimpleSG)2026.03		77.56	95.33	5.11	0.44
MME-VLA (FrameSamp)2026.03		57.33	86.22	94.67	28.89
MemER2026.03		56.67	79.33	59.33	0
MME-VLA (TokenDrop)2026.03		54.22	87.56	91.78	4.22
MME-VLA (GroundSG)2026.03		52	92.67	7.33	0
MME-VLA (TokenDrop)2026.03		48.67	85.11	94.67	3.11
MME-VLA (SimpleSG)2026.03		46	63	45	2
MME-VLA (FrameSamp)2026.03		41.22	72	73.67	13.67
SAM2Act+2026.03		40	76	25.33	0
MME-VLA (FrameSamp)2026.03		39.56	87.33	92	42
MME-VLA (RMT)2026.03		35.78	60.22	36	5.56
MME-VLA (TTT)2026.03		35.56	62.89	42.44	3.33
MME-VLA (TTT)2026.03		34.89	63.78	41.33	4
MME-VLA (TokenDrop)2026.03		34.44	83.56	86	5.33
MME-VLA (TTT)2026.03		34.22	65.11	36.67	2.11
MME-VLA (RMT)2026.03		33.33	60.78	37.78	4.67
MME-VLA (RMT)2026.03		32.44	56.89	33.56	5.78
pi_0.52026.03		30	42.89	35.56	6.67
pi_0.52026.03		26.67	58.33	26.67	4.67
MME-VLA (GroundSG)2026.03		26	18	4	3