VLA-Δ: Learning Physical State Transition for Vision-Language-Action Models ▲ 53 💬 1 2026-10-17 arXiv HF 原文 GitHub ★14 摘要 （无摘要）