MemoryVLA: Perceptual-Cognitive Memory in Vision-Language-Action Models for Robotic Manipulation

Benchmark Model Rank Results
robot-manipulation-on-liberoMemoryVLA–Success Rate: 96.65
robot-manipulation-on-libero-90MemoryVLA–Success Rate: 95.6
robot-manipulation-on-libero-goalMemoryVLA–Success Rate: 96.4
robot-manipulation-on-libero-longMemoryVLA–Success Rate: 93.4
robot-manipulation-on-libero-objectMemoryVLA–Success Rate: 98.4
robot-manipulation-on-libero-spatialMemoryVLA–Success Rate: 98.4
robot-manipulation-on-simplerenv-google-robotMemoryVLA (Google Robot Visual Matching)–Success Rate: 77.7
robot-manipulation-on-simplerenv-widow-xMemoryVLA–Average: 0.719