State Estimation: How a Robot Tracks Where It Is

A robot never directly "knows" its own position or velocity — it estimates them continuously from imperfect sensors.

State estimation is the ongoing process of inferring a robot's true position, velocity, and orientation from noisy, incomplete, indirect sensor measurements, since no sensor directly hands over a perfectly accurate answer.

Why this is genuinely hard

Every sensor has noise and bias, and no single sensor measures everything you need — an IMU measures acceleration and rotation rate, not position directly; leg odometry can be thrown off by foot slip; vision can lose tracking in a low-texture environment. State estimation has to reconcile all of these imperfect, sometimes-contradictory sources into one coherent best estimate.

Why everything downstream depends on it

Balance control, footstep planning, navigation, and manipulation all assume the robot has an accurate sense of its own state as a starting point. A state estimation error doesn't just cause one isolated mistake — it propagates into every control decision built on top of that bad estimate.