Instruct2Act: Mapping Multi-modality Instructions to Robotic Actions with Large Language Model