Paper Title
Tightly-Coupled Sensor Fusion and SLAM Frameworks for Autonomous Indoor Delivery Robots: A Comprehensive Survey and System-Level Analysis

Abstract
Autonomous Mobile Robots (AMRs) are increasingly adopted for indoor logistics applications in structured environments such as hospitals, university campuses, research laboratories, and office complexes. Although substantial progress has been achieved in planar autonomous navigation, maintaining long-term localization accuracy and map consistency in dynamic indoor environments remains a significant challenge. Factors such as cumulative odometric drift, sensor noise, frequent rotational maneuvers, and human–robot coexistence impose strict constraints on reliable navigation and mapping performance. This paper presents a comprehensive survey and system-level analysis of vision-guided and sensor-fusionbased SLAM and navigation frameworks designed for indoor autonomous delivery robots operating on a single floor. The study critically reviews recent advances in tightly coupled LiDAR–Visual–Inertial SLAM systems, including LVI-SAM, FAST-LIO2, FAST-LIVO, RI-LIO, and W-VSLAM, with a particular emphasis on their robustness, observability properties, and performance in structured indoor environments with limited geometric features. Building upon the surveyed methodologies, a tightly coupled multi-sensor fusion architecture is implemented within the ROS 2 ecosystem. The proposed system integrates wheel odometry, inertial measurements, and 360° LiDAR data to provide a resilient and high-rate state estimation pipeline. An Extended Kalman Filter (EKF) is employed to fuse proprioceptive and exteroceptive sensor data, generating a drift-mitigated motion estimate that serves as a motion prior for graph-based SLAM using slam toolbox. Careful design of the TF tree ensures strict frame decoupling between odometry, base, and map frames, thereby minimizing distortion effects and preventing rotational instability during aggressive yaw motions. Experimental evaluation demonstrates that accurate yaw observability, time-synchronized sensor fusion, and a well-structured transform hierarchy are critical factors for preserving map fidelity and achieving stable localization. The proposed framework produces consistent occupancy grid maps, smooth trajectory execution, and reliable obstacle avoidance in cluttered indoor environments. Furthermore, the paper discusses system-level limitations such as sensor calibration sensitivity and dynamic obstacle interference, and outlines future research directions including semantic-aware mapping and adaptive perception pipelines. This work contributes both a consolidated review of state-of-the-art indoor SLAM techniques and a practical, scalable reference architecture for autonomous indoor logistics robots. Keywords - Autonomous Mobile Robots, Indoor Navigation, Simultaneous Localization and Mapping (SLAM), Multi-Sensor Fusion, LiDAR–Inertial Odometry, Visual–Inertial Odometry, Extended Kalman Filter (EKF), ROS 2 Navigation Stack, Occupancy Grid Mapping, Autonomous Delivery Systems