ResearchBenchmarksVision-Language-Action Control on RoboTwin place_empty_cupVision-Language-Action Control on RoboTwin place_empty_cupFollow1Success RateHY-VLA0.90640.93070.9550.9793Jul 2, 2026Evaluation resultsMethodMethodLinksSuccess RateStep Time (ms)Inference Time (ms)VRAM Usage (MiB)HY-VLAModel Backbone=Hunyuan...Model Backbone=Hunyuan-VL, Action Chunk=202026.071735.91,340.36,850pi0.5Model Backbone=PaliGem...Model Backbone=PaliGemma, Action Chunk=502026.070.9156.85266.66,546