Tactile Sensing, Skill Learning, and Robotic Dexterous Manipulation

Chapter 9: Learning hierarchical control for robust in-hand manipulation

Tingguang Li    Tencent Robotics X, Shenzhen, China

Abstract

In-hand manipulation has been a long-standing challenge due to the complexity of the dexterous hand's dynamic model and the difficulty in leveraging different kinds of manipulation motions. To address these challenges, researchers have been using either dynamic modeling methods or deep reinforcement learning methods, which are limited in different ways. In this chapter, we introduce a hierarchically structured method that combines a dynamic controller for conducting primitive motions and a deep reinforcement learning network to generate the action sequence of the primitive motions. The primitive motions for the in-hand ...