SLAM Handbook: From Localization and Mapping to Spatial Intelligence

Abstract

Simultaneous Localization and Mapping —better known as SLAM— refers to the fundamental problem of building spatial models of an environment while simultaneously determining the position of a robot within that environment. The term itself was first coined in 1995 by Hugh Durrant-Whyte and John Leonard, marking the formalization of a problem that sits at the intersection of robotics, geometry, controls, and probabilistic inference. SLAM is as elegant as it is formidable. At its core, it addresses the challenge of reasoning over high-dimensional, uncertain, and dynamic systems. The process demands precise spatial inference and robust probabilistic modeling to build coherent maps of the world —maps that must be constructed in real time, often under conditions of noise and ambiguity. What makes SLAM particularly compelling is its universality. In computer vision, it is mirrored in the problem of Structure from Motion; in robotics, it underpins everything from indoor autonomous navigation to planetary exploration and selfdriving cars. Since its inception, SLAM has inspired tens of thousands of research papers, drawing deeply from disciplines as diverse as physics, statistics, computer vision, geometry, controls, and machine learning. Its evolution has catalyzed the development of increasingly capable autonomous systems, able to operate at scale in complex, open-world environments. This volume brings together contributions from some of the field’s foremost experts and rising stars. The chapters represent the state of the art in SLAM today, reflecting both the depth of theoretical innovations and the breadth of practical applications. From its early formulations based on Kalman filters and Bayesian estimation, SLAM has matured into a rich tapestry of mathematical frameworks —encompassing graph-based optimization, factor graphs, nonlinear least squares, and deep learning-based techniques. Beyond introducing the mathematical foundations of SLAM, this volume provides valuable guidance to the practitioner by discussing real-world use cases ranging from vision-based and LiDAR-based SLAM systems to legged locomotion. It also covers recent developments in Spatial AI, showing how advances in deep learning, differentiable rendering, and large vision and language models point the way toward representations that provide robots with a rich spatial and semantic understanding of their environment.