What Is Odometry?

Odometry is the estimation of a moving platform's change in position and orientation over time by integrating measurements taken onboard, without reference to an external positioning infrastructure. It is a form of dead reckoning: each new pose is computed as an increment applied to the previous one, so the estimate is relative rather than absolute. Odometry belongs to the broader problems of motion detection and position measurement, and in mobile robotics it supplies the motion model that localization and mapping algorithms refine.

The defining property of any odometry system is that error accumulates. Because increments are summed, small biases in each measurement compound into unbounded drift over a long traverse, and the practical question is never whether drift exists but how fast it grows and what independent measurement will bound it. Systems are therefore rated by drift as a percentage of distance traveled rather than by absolute accuracy.

Wheel and Inertial Odometry

The classical implementation counts encoder pulses on driven or free-rolling wheels and converts them to a translation and heading change through a kinematic model of the vehicle. This is cheap and runs at high rate, but it fails whenever the no-slip assumption fails, and on loose regolith or wet pavement the wheels can turn while the vehicle barely advances. Inertial odometry substitutes accelerometers and gyroscopes, double-integrating specific force to obtain displacement. Inertial units do not care about ground contact, but their drift is worse because bias errors are integrated twice, and low-cost microelectromechanical sensors drift within seconds. In practice the two are fused, with the gyroscope stabilizing heading between encoder updates and the encoders bounding the velocity estimate.

Visual and Visual-Inertial Odometry

Visual odometry estimates motion from a sequence of camera images by detecting features, matching them across frames, and solving for the rigid transformation that best explains the observed displacement. Stereo rigs recover metric scale directly from the baseline, while monocular systems recover motion only up to an unknown scale factor unless an inertial unit or a known object size supplies it. The technique reached operational maturity on Mars: the visual odometry system flown on the Mars Exploration Rovers compared stereo pairs onboard to correct the six degree of freedom pose whenever wheel counts were unreliable, and later work extended a stereo vision approach to precise pose estimation for the Perseverance rover. Tightly coupled visual-inertial odometry, in which image features and inertial preintegration are optimized jointly, is now the standard formulation for aerial vehicles and handheld devices, and studies of monocular visual odometry for planetary flight address the low-texture and harsh-lighting conditions that break feature matching.

Lidar, Radar, and Sensor Choice

Range sensors offer a third route. Lidar odometry registers successive point clouds using variants of iterative closest point or feature-based scan matching, which works in darkness and does not depend on surface texture, though it degrades in geometrically self-similar corridors and tunnels. Radar odometry trades angular resolution for immunity to dust, fog, and rain, making it attractive for automotive and mining use. Choosing among these modalities is largely a matter of which failure mode is tolerable in the intended environment, and an experimental comparison of visual and single-receiver satellite odometry measured a lower drift rate for the satellite-based method than for a stereo visual pipeline over a 1.8 km ground traverse. Whatever the modality, absolute correction from satellite navigation, surveyed landmarks, or loop closure in a map remains necessary to keep long-run error bounded.

Applications

Odometry has applications in a wide range of fields, including:

• Planetary rover navigation, where satellite positioning is unavailable
• Autonomous vehicles and driver assistance, supplying motion priors for perception and mapping
• Warehouse and service robots operating indoors without satellite reception
• Simultaneous localization and mapping, which uses odometry as its motion model
• Augmented reality headsets and handheld devices for camera tracking
• Underwater and subterranean survey, including pipeline and mine inspection