Deep Q Reinforcement Learning for Autonomous Navigation of Surgical Snake Robot in Confined Spaces

Abstract

Airway management is fundamental for all anesthetic as well as emergency medicine procedures to maintain airway patency, prevent aspiration and permit ventilation without leakage. While endotracheal intubation and tracheostomy are regarded as the go-to procedures in such incidents, they are reportedly correlated with numerous side effects which can sometimes even be life-threatening [1]. These complications stem from the fact that essentially a human is blindly and manually maneuvering the intubation tube. In order to mitigate the ensued risks and aftereffects of currently employed methods, this work uses a surgical snake robot [2] to autonomously navigate down the airway. The contribution of this paper is developing the navigation policy that utilizes images from a monocular camera mounted on its tip. We use Q Reinforcement Learning in Deep Convolutional Neural Networks (DCNN) [3], widely referred to as Deep Q Reinforcement Learning Neural Networks (DQNN), to produce these policies. The system can serve as an assistive device for medical personnel to perform endoscopic intubation, with minimal to no human input.