Autonomous Robotic Colonoscopy: A Supervised Learning Approach for Enhanced Navigation and Collision Detection

Among conventional colonoscopy techniques, improper insertions can lead to a patient discomfort, and in severe cases, intestinal perforation. To minimize such risks, endoscopists usually should put their efforts into exercising considerable skill and caution, which requires both physical dexterity and an in‐depth understanding of colonic anatomy. Thus, it is noted that the steep learning curve associated with conventional colonoscopy remains a persistent and critical issue in clinical practice. To address these open challenges, an autonomous colonoscopy system that integrates the procedural knowledge of experienced doctors enabling to enhance safety and reduce operator dependency is proposed. The proposed system combines a motorized endoscopic platform with an AI‐based navigation module. The motorized platform employing a tendon‐driven mechanism is capable of three degrees of freedom (two rotational and one translational motions) while ensuring mechanical stability throughout the operation. The navigation system, based on supervised learning, is designed to predict target steering points and estimate collision probabilities, contributing to real‐time medical safety. Conclusively, this work presents a fully integrated autonomous colonoscopy system that demonstrates a high level of autonomy, achieving a promising average cecum‐reaching time of 3 min 36 s ± 1 min 47 s, with 90% success rate.