VQ-VLA: Improving Vision-Language-Action Models via Scaling Vector-Quantized Action Tokenizers

Benchmark Model Rank Results
robot-manipulation-on-libero-90VQ-VLA (VQO+L+M)#5Success Rate: 86.61
robot-manipulation-on-libero-goalVQ-VLA (Conv Residual VQ-VAE, all-LIBERO)#106Success Rate: 75.2
robot-manipulation-on-libero-longVQ-VLA (Conv Residual VQ-VAE, all-LIBERO)#106Success Rate: 60.0