InternVLA-A1.5: Unifying Understanding, Latent Foresight, and Action for Compositional Generalization

Benchmark Model Rank Results
robot-manipulation-on-simplerenv-widow-xInternVLA-A1.5#4Average: 0.808
robot-manipulation-on-robotwin-2-0-easy-50-tasksInternVLA-A1.5-RoboTwin (Clean)#7Success Rate: 93.3
robot-manipulation-on-robotwin-2-0-hard-50-tasksInternVLA-A1.5-RoboTwin (Randomized)#7Success Rate: 93.0
robot-manipulation-on-libero-longInternVLA-A1.5-Libero#8Success Rate: 98.4
robot-manipulation-on-libero-plus-zero-shot-transferInternVLA-A1.5-Libero#12Success Rate: 84.8
robot-manipulation-on-liberoInternVLA-A1.5-Libero#13Success Rate: 98.9
robot-manipulation-on-libero-objectInternVLA-A1.5-Libero#17Success Rate: 99.8
robot-manipulation-on-libero-goalInternVLA-A1.5-Libero#22Success Rate: 98.6
robot-manipulation-on-libero-spatialInternVLA-A1.5-Libero#44Success Rate: 98.6