Robust Indoor Localization for Autonomous Mobile Robots: A Review of Bayesian Sensor Fusion and SLAM Algorithms

Indoor autonomous mobile robots (AMRs) must estimate pose without GNSS while facing occlusions, perceptual aliasing, wheel slip, and dynamic obstacles. These conditions make single-sensor localization brittle and motivate probabilistic fusion of motion cues with environment-referenced and infrastructure observations. This algorithm-focused review synthesizes robust indoor localization for AMRs by linking sensing models to estimation theory and engineering trade-offs. This paper is organized around measurement models of common sensors, Bayesian filtering (Kalman and particle) and smoothing via factor-graph optimization, and SLAM pipelines that combine front-end association with back-end optimization and loop closure. Learning-augmented components are discussed as targeted add-ons that improve perception, uncertainty modeling, and outlier rejection. Finally, practical evaluation guidelines are summarized, including recovery behavior and computational metrics that matter in deployment. Open directions include life-long mapping in changing buildings and reliability-aware autonomy regulation. The emphasis is on selection criteria that transfer to real deployments.