Growing Robot Navigation Based on Deep Reinforcement Learning

摘要

The recent progress in materials and structures has kick-started the development of soft eversion robot with the ability to grow in size. However, despite its promising capability to navigate challenging terrains, this type of robot still lacks a navigation strategy due to the robot's complexity courtesy of its increasing degrees of freedom as it grows. In this paper, we develop a growing robot navigation strategy based on deep reinforcement learning. The reinforcement learning was specifically designed to work with growing robot even as its degrees of freedom increase. The algorithm was shown to work in navigating growing robot in a planar environment towards a random target. The results show that the reinforcement learning is a promising candidate to be used for growing robot navigation.