StreamPI: Streaming Multimodal Temporal Modeling for Vision-Language-Action Models

Benchmark Model Rank Results
robot-manipulation-on-libero-goalStreamPI (T=5)#2Success Rate: 99.6
robot-manipulation-on-calvin-abcd-to-dStreamPI (T=5)#6Average Length: 4.547
robot-manipulation-on-libero-objectStreamPI (T=5)#20Success Rate: 99.8
robot-manipulation-on-libero-spatialStreamPI (T=5)#34Success Rate: 98.8
robot-manipulation-on-liberoStreamPI (T=5)#44Success Rate: 98.3
robot-manipulation-on-libero-longStreamPI (T=5)#50Success Rate: 95.0