Permanence on RoboMME Permanence
98.78Video Umsk Success RateMME-VLA (GroundSG)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MME-VLA (GroundSG)Memory Representation Type=Symbolic Memory, Integration Mechanism=Oracle2026.03 | 98.78 | 95 | 99.22 | 80.22 | |
| Human Performance2026.03 | 90 | 92 | 92 | 90 | |
| MME-VLA (GroundSG)Memory Representation Type=Symbolic Memory, Integration Mechanism=QwenVL2026.03 | 88.67 | 24 | 30.67 | 14 | |
| MemER2026.03 | 81.33 | 72 | 38 | 21.33 | |
| MME-VLA (GroundSG)Memory Representation Type=Symbolic Memory, Integration Mechanism=Gemini2026.03 | 36 | 14 | 13 | 0 | |
| MME-VLA (SimpleSG)Memory Representation Type=Symbolic Memory, Integration Mechanism=QwenVL2026.03 | 34.22 | 19.33 | 15.33 | 9.56 | |
| MME-VLA (TokenDrop)Memory Representation Type=Perceptual Memory, Integration Mechanism=Context2026.03 | 33.78 | 31.56 | 26.22 | 16 | |
| MME-VLA (SimpleSG)Memory Representation Type=Symbolic Memory, Integration Mechanism=Oracle2026.03 | 33.11 | 22 | 15.56 | 15.56 | |
| MME-VLA (FrameSamp)Memory Representation Type=Perceptual Memory, Integration Mechanism=Modul2026.03 | 32.67 | 25.11 | 24.44 | 18.22 | |
| MME-VLA (FrameSamp)Memory Representation Type=Perceptual Memory, Integration Mechanism=Expert2026.03 | 31.78 | 25.78 | 22.89 | 20.22 | |
| MME-VLA (TTT)Memory Representation Type=Recurrent Memory, Integration Mechanism=Expert2026.03 | 31.78 | 22.44 | 19.56 | 18 | |
| MME-VLA (RMT)Memory Representation Type=Recurrent Memory, Integration Mechanism=Context2026.03 | 31.33 | 10.89 | 17.33 | 2 | |
| MME-VLA (RMT)Memory Representation Type=Recurrent Memory, Integration Mechanism=Modul2026.03 | 31.11 | 11.78 | 17.78 | 2.44 | |
| pi_0.5past actions=true2026.03 | 30.67 | 23.67 | 20.67 | 16 | |
| MME-VLA (TTT)Memory Representation Type=Recurrent Memory, Integration Mechanism=Context2026.03 | 29.78 | 22.89 | 18.44 | 14.44 | |
| MME-VLA (SimpleSG)Memory Representation Type=Symbolic Memory, Integration Mechanism=Gemini2026.03 | 29 | 9 | 14 | 2 | |
| MME-VLA (TokenDrop)Memory Representation Type=Perceptual Memory, Integration Mechanism=Modul2026.03 | 28.22 | 29.33 | 28.44 | 21.33 | |
| MME-VLA (RMT)Memory Representation Type=Recurrent Memory, Integration Mechanism=Expert2026.03 | 28 | 17.11 | 15.78 | 2 | |
| SAM2Act+2026.03 | 27.33 | 32 | 18 | 26.67 | |
| MME-VLA (TTT)Memory Representation Type=Recurrent Memory, Integration Mechanism=Modul2026.03 | 27.22 | 22.11 | 25.22 | 14.11 | |
| MME-VLA (FrameSamp)Memory Representation Type=Perceptual Memory, Integration Mechanism=Context2026.03 | 26.89 | 30.22 | 20.89 | 15.22 | |
| MME-VLA (TokenDrop)Memory Representation Type=Perceptual Memory, Integration Mechanism=Expert2026.03 | 26.67 | 30.44 | 18.44 | 18.89 | |
| pi_0.52026.03 | 20.44 | 22.22 | 18.67 | 6.67 |