SLAM (Simultaneous Localization and Mapping) has a wide application prospect in navigation of autonomous endoscopic minimally invasive surgery. However, traditional methods like ORB-SLAM2 often struggle in dynamic environments and exhibit significant modeling inaccuracies, particularly in bronchial scenes where respiratory motion induces continuous deformation of the airway structure. To address this challenge, we propose a novel monocular SLAM-based framework for dynamic bronchial environment reconstruction tailored to robotic lung invasive surgery. Firstly, a pseudo-static processing method that builds map sequences at identical respiratory phases across cycles was built. Furthermore, we present a voxel model optimization technique using curvature-consistent graph interpolation to refine the bronchial lumen surface, eliminating pores and redundancies in the voxel map. Experiments conducted on real patient demonstrate that our method has good accuracy and robustness. Compared to existing state-of-the-art methods, our framework achieves superior reconstruction completeness, showing strong potential for clinical use in robotic bronchoscopy.

In this work, a multimodal point cloud registration method using CT and video frames is proposed to optimize the modeling of the bronchial cavity environment. Preoperative CT data improve the quality of point clouds acquired from intraoperative video frames. Initially, preoperative CT scans are used to obtain bronchial point clouds and airway centerlines, while intraoperative bronchial point clouds and endoscope trajectories are captured in real-time using SLAM. Given that intraoperative frame-by-frame mapping cannot be directly globally registered, multi-modal point clouds undergo local segmentation. Subsequently, the preoperative bronchial airway centerlines guide iterative scaling and adjustment of the preoperative CT point clouds, achieving precise registration between the CT and the video frame point clouds. Experimental results demonstrate a rapid and accurate enhancement in the quality of the intraoperative bronchial point cloud, providing more precise maps of the cavity environment for surgical robots. The method is validated and evaluated using CT and video frame data collected from ex vivo pig lungs, achieving intraoperative mapping accuracy of 0.5 millimeter, respectively. These results surpass those of methods relying solely on SLAM for intraoperative mapping.

Our amphibious snake-like robot conquers the terrestrial and aquatic environments with different kinds of gait. Future work has been planned to apply the robot in inspection tasks in complicated wild environments.