Dead reckoning is estimating where a robot is by adding up how far it has moved from a known start — cheap, always-available, and the reason robots drift and need GPS, landmarks, or SLAM to stay honest.
Dead reckoning means figuring out where you are by tracking every step from a known starting point — "I went 3 m forward, then turned and went 2 m." It works, but small errors add up the longer you go without a fresh fix.
Close your eyes and try to walk to your kitchen counting steps. You'll get close — for a while. Keep going and you'll drift off course. That's dead reckoning: navigation by adding up your own motion, and its famous weakness.
What it is
Dead reckoning estimates a robot's current position by integrating its motion from a known starting point. A wheeled robot counts wheel rotations (odometry); a drone integrates acceleration and rotation from its IMU; a ship historically used speed and heading over time. No external reference needed — just "I started here, and I've moved this much since."
Dead reckoning integrates motion each step
Each cycle updates the estimate from the last one. That self-referential loop is exactly why errors compound — there's no outside check.
Why it drifts
Every measurement has a small error, and dead reckoning adds them up forever. A wheel slips, a tire is slightly worn, the IMU has a tiny bias — each contributes a little, and the estimated position wanders further from the truth the longer you go without a fix. IMU-only dead reckoning can drift metres in seconds; good wheel odometry drifts slower but still without bound. There is no configuration of dead reckoning that doesn't eventually drift.
How robots keep it honest
Because dead reckoning is smooth, fast, and always available, robots don't throw it away — they correct it with occasional absolute references:
Sensor fusion — a Kalman filter blends drift-free-but-noisy fixes with smooth-but-drifting dead reckoning to get the best of both.
This is exactly the ROS convention where the odom frame is smooth but drifts, and the map frame is drift-free but jumps when a correction lands (see tf).
Why it matters
Dead reckoning is the foundation every localization system builds on and then compensates for. Understanding why it drifts — and that no filter removes the need for an occasional real fix — is core to how robots know where they are.
