ARCGym: Benchmarking Deep Reinforcement Learning in Autonomous Robotic Colonoscopy
In the authors' words
Simulations for learning-based autonomous colonoscopic navigation focus mainly on fully actuated capsule robots, failing to capture the contact-rich navigation of long and flexible clinical colonoscopes. We present the Autonomous Robotic Colonoscopy Gym (ARCGym), an open-source reinforcement learning environment and benchmark for image-based navigation in clinically derived deformable colon anatomies. ARCGym supports multiple types of colonoscope robots, spanning capsule robots and flexible endoscopes, with this work focusing on flexible endoscopes including magnetic-driven tip actuation and clinically used proximally translational actuation. This work includes five CT-reconstructed colons representing typical clinical scenarios, a set of clinically meaningful navigation subtasks, and unified success metrics. We introduce a reward combining depth-based lumen alignment with a lumen-visibility score to improve learning under occlusions. Experiments across tasks, robots, and anatomies show that autonomous navigation remains challenging for both magnetic-driven and proximal-insertion flexible robots, with proximal-insertion actuation remaining an open problem.
Appeared: Tuesday, September 22. arXiv. Preprint with a published version.
Published version: G. Ji, M. Finocchiaro, K. Erleben, and H. Yin, "ARCGym: Benchmarking Deep Reinforcement Learning in Autonomous Robotic Colonoscopy," IEEE Robotics and Automation Letters, 2026