Dismantling the Illusion of Vision-Language-Action Models Competence via Explicit Distributional Shifts

ICML2026 / Benchmark and Dataset