VAMOS: A Hierarchical Vision-Language-Action Model for Capability-Modulated and Steerable Navigation