A Review of Robotic Manipulation of Deformable Objects with Imitation Learning Techniques: Progress and Outlook

Abstract

Imitation learning (IL) for deformable object manipulation (DOM) has become an increasingly relevant approach in recent years, owing to its ability to replicate behaviors from expert demonstrations. Compared to rigid object manipulation, DOM is significantly more intricate, posing challenges such as complex object dynamics and high-dimensional state spaces. We review the recent literature on the topic and classify it into 5 categories based on the type of object the study aims to manipulate: Fabric-like, rope-like, malleable, amorphous flowing, and others. A comparative overview of methods that tackle each category using IL techniques is provided. Finally, current obstacles in manipulation strategies are described, and future research approaches for this area are proposed.