Robot Learning For Deformable Object Manipulation Tasks

This article discusses the challenges of Deformable Object Manipulation (DOM) in robotics and the potential application of Machine Learning (ML) techniques to solve these problems. The author introduces the ReForm simulation sandbox, which includes six tasks involving Deformable Linear Objects (DLOs), and focuses on two representative tasks: shape-servoing and cable-routing. The use of Reinforcement Learning (RL) methods for shape-servoing is also explored.