“Patch Policy surpasses the performance of a fine-tuned OpenVLA-OFT model by 18%.”