Dita: Scaling Diffusion Transformer for Generalist Vision-Language-Action Policy

Benchmark Model Rank Results
robot-manipulation-on-simpler-envDita-300M#2Visual Matching: 0.687Visual Matching-Pick Coke Can: 0.837…
robot-manipulation-on-simplerenv-google-robotDita (Google Robot Visual Matching)#11Success Rate: 68.7
robot-manipulation-on-calvinDita (CALVIN ABC→D)#15avg. sequence length (D to D): 3.61
robot-manipulation-on-libero-spatialDita w/ wrist (LIBERO-Spatial)#64Success Rate: 97.4
robot-manipulation-on-libero-spatialDita (third-person RGB) (LIBERO-Spatial)#100Success Rate: 84.2
robot-manipulation-on-libero-objectDita (third-person RGB) (LIBERO-Object)#87Success Rate: 96.3
robot-manipulation-on-libero-objectDita w/ wrist (LIBERO-Object)#93Success Rate: 94.8
robot-manipulation-on-libero-goalDita w/ wrist (LIBERO-Goal)#89Success Rate: 93.2
robot-manipulation-on-libero-goalDita (third-person RGB) (LIBERO-Goal)#101Success Rate: 85.4
robot-manipulation-on-libero-longDita w/ wrist (LIBERO-Long)#91Success Rate: 83.6
robot-manipulation-on-libero-longDita (pretrained on Droid)#95Success Rate: 77.9
robot-manipulation-on-libero-longDita (third-person RGB) (LIBERO-Long)#104Success Rate: 63.8
robot-manipulation-on-liberoDita w/ wrist#121Success Rate: 92.3
robot-manipulation-on-liberoDita (third-person RGB)#132Success Rate: 82.4