TurboVLA: Real-Time Vision-Language-Action Model at 32 Hz on an RTX 4090 with <1 GB VRAM

Benchmark Model Rank Results
robot-manipulation-on-libero-spatialTurboVLA (DINOv3 ViT-B)#16Success Rate: 99.2
robot-manipulation-on-libero-objectTurboVLA (DINOv3 ViT-B)#18Success Rate: 99.8
robot-manipulation-on-robotwin-2-0-hard-50-tasksTurboVLA (DINOv3 ViT-L)#24Success Rate: 88.10
robot-manipulation-on-robotwin-2-0-easy-50-tasksTurboVLA (DINOv3 ViT-L)#27Success Rate: 88.02
robot-manipulation-on-libero-goalTurboVLA (DINOv3 ViT-B)#48Success Rate: 97.4
robot-manipulation-on-libero-longTurboVLA (DINOv3 ViT-B)#60Success Rate: 94.2
robot-manipulation-on-liberoTurboVLA (DINOv3 ViT-B)#61Success Rate: 97.7