Skip to main content
A Field Book for the Robotics Interview · 2026 Edition

SLAM
from zero to hero

How robots figure out where they are and what the world looks like, explained from scratch, one honest chapter at a time. No prior SLAM knowledge assumed. Every paper named and linked, and every chapter ends with a setup you can run yourself.

Before You StartHow to Read This Book

Read Chapter 0 first, even if it feels basic. Everything else in the book is built on the six ideas in that chapter: odometry, drift, loop closure, the map, optimization, and coordinate frames. Once those click, the fancy stuff (factor graphs, Gaussian splatting, differentiable bundle adjustment) turns out to be variations on the same six ideas wearing different costumes.

After that the book widens out. Chapters 1 and 2 give you the math and the hardware that every later chapter leans on. Chapters 3 through 7 cover the systems themselves: lidar, cameras, the budget RGB-D rig, the ORB-SLAM lineage in depth, and learned methods. Chapter 8 teaches you to measure any of it honestly, and Chapter 9 puts the map to work driving a robot.

Each chapter follows the same rhythm: intuition first, then the real machinery, then the systems people actually run, then a worked setup you can reproduce on your own hardware, a linked reading list, and finally interview questions with answers hidden behind click-to-reveal panels. Try answering out loud before you open them.

SLAM is one problem asked two ways at once: "Where am I?" (localization) and "What does the world look like?" (mapping). The cruel joke is that each answer requires the other. That circularity is the entire field.

ContentsThe Ten Chapters

  1. 00FoundationsWhat SLAM even is. Odometry, drift, loop closure, maps, frames, and the six ideas the rest of the book keeps reusing.Start here
  2. 01The Estimation ToolboxBayes filters, Kalman and particle filters, least squares, factor graphs, robust kernels, and why rotations need special care.The math
  3. 02Sensors, Calibration & TimeWhat each sensor really measures, its noise model, and the calibration and clock-sync work that makes or breaks a rig.Hardware
  4. 03LiDAR SLAM & the Mapping PipelineScan matching, deskew, LOAM to FAST-LIO2, and OctoMap versus the distance fields that replaced it.Geometry
  5. 04Visual SLAM: The State of the ArtSix eras from ORB features to foundation models that predict geometry in one forward pass.SOTA
  6. 05RTAB-Map + RealSense D435iThe cheapest credible indoor SLAM rig, configured end to end, and every way it bites you in the field.Hands-on
  7. 06ORB-SLAM: Every Version, Explained and RunThe four threads every release shares, what changed from v1 to v2 to v3, and two builds you run yourself.Deep Dive
  8. 07Learning-Based SLAMWhere machine learning actually plugs in, and where it still loses to forty-year-old geometry.ML × SLAM
  9. 08Evaluation, Datasets & BenchmarksATE, RPE, the datasets everyone reports on, the evo toolchain, and how to benchmark your own robot honestly.Rigor
  10. 09Planning, Navigation & LocalizationWhat the robot does with the map: costmaps, A* to MPPI, and the frame contract that holds it all together.Autonomy

PromiseWhat "Hero" Means Here

By the last page you should be able to: draw the SLAM pipeline from memory on a whiteboard; explain why a map jumps during loop closure and why the wheels don't jump with it; argue when to use a camera versus a lidar; calibrate and time-sync a sensor rig without folklore; score a trajectory with numbers instead of screenshots; sort any "AI + SLAM" question into three clean buckets; and name the 2023–2026 systems that replaced the classics, with the paper links to prove it.

Compiled July 2026 · Built for study, not for citation · All external links go to the original papers, docs, or repos.