This chapter examines the methodologies for defining task models based on contact transitions in the polyhedral world, as well as the resulting LfO implementation based on these definitions, as presented by Ikeuchi and Suehiro [9]. Similar to the two-block world discussed in the previous chapter, the system is composed of four key processes: image sequence segmentation via a stop-and-go approach, object recognition, task recognition, and task execution. This chapter primarily focuses on the definition and recognition methodologies of the task models.