PerturBot: Breaking Shortcut Priors in Vision-Language-Action Models with Perturbative Training ▲ 7 💬 1 2026-10-07 arXiv HF 原文 GitHub ★6 摘要 （无摘要）