Manipulate-Anything: Automating Real-World Robots using Vision-Language Models