Hi Robot: Open-Ended Instruction Following with Hierarchical Vision-Language-Action Models