Dead reckoning

On this page 7

Dead reckoning is a navigation method that estimates a robot’s current position by integrating measured motion from a known starting pose, using speed, heading and elapsed time rather than external position references. In robotics navigation, it is the continuous, short-term pose estimate that keeps a robot moving between corrections from GNSS, landmarks or map matching.

The method is old: ship navigators used log lines, compasses and clocks long before satellites existed. Mobile robots apply the same principle with wheel encoders, inertial measurement units (IMUs) and visual or LiDAR odometry. Its defining property is that errors accumulate without bound, so dead reckoning is almost always paired with absolute localization.

How dead reckoning integrates motion

Dead reckoning works by repeatedly adding small displacement increments to the previous pose estimate. For a ground robot in the plane, each update takes the measured forward distance and heading change during a short interval and applies them to the pose (x, y, theta).

For a differential-drive robot, the left and right wheel distances from encoders give the forward motion as their average and the rotation as their difference divided by the wheel track width. The new position follows from the old position plus the distance projected along the current heading. Update rates of 50 to 200 Hz are typical, which keeps discretization error small.

Inertial dead reckoning instead integrates acceleration twice and angular rate once. This makes it independent of wheel contact, but it is far more sensitive to sensor bias, because a constant accelerometer bias produces a position error that grows with the square of time.

The table below lists the main sources of motion information used for dead reckoning in robots.

SourceQuantity measuredTypical strengthMain weakness
Wheel odometryWheel rotation via encodersCheap, accurate on firm floorsSlip, uneven terrain, wheel wear
IMUAcceleration and angular rateHigh rate, no contact neededBias drift, double integration error
Visual odometryCamera image motionWorks on any vehicle typeFails in low texture or poor light
LiDAR odometryScan-to-scan alignmentPrecise in structured spacesDegenerates in long featureless corridors
Compass or magnetometerMagnetic headingAbsolute heading referenceDisturbed by steel and motors indoors
Motion sources for dead reckoning

Why dead reckoning error grows over time

Dead reckoning error grows because every increment carries a small error that is never removed. Errors in distance add up roughly linearly, but heading errors are the most damaging: a small angular error rotates every later displacement, so lateral position error grows with distance traveled.

Errors fall into two groups. Systematic errors, such as unequal wheel diameters or an incorrect track width, bias every step in the same direction and can be reduced by calibration. Non-systematic errors, such as wheel slip on a wet floor or a bump, are random and can only be bounded by external corrections.

MethodTypical driftDominant error source
Calibrated wheel odometry, indoorsAbout 1-5% of distance traveledHeading error from slip and track width
Consumer-grade MEMS IMU aloneMeters to tens of meters within a minuteAccelerometer and gyroscope bias
Tactical-grade IMUMeters over several minutesResidual gyro bias
Visual-inertial odometryOften below 1% of distanceScale and yaw drift in poor scenes
Typical drift characteristics by method

These figures depend strongly on sensor quality, terrain, speed and calibration, so they serve as orders of magnitude rather than guarantees.

Dead reckoning compared with absolute localization

Dead reckoning is relative localization: it knows how far the robot moved, not where it is in a global frame. Absolute methods such as GNSS, RTK GPS or map-based localization with AMCL measure position directly but update less often and can fail intermittently.

PropertyDead reckoningAbsolute localization
ReferencePrevious poseExternal map, satellites or beacons
Update rateHigh, often 50-200 HzLower, often 1-20 Hz
Short-term smoothnessVery smoothCan jump between estimates
Long-term errorUnbounded driftBounded by measurement accuracy
AvailabilityAlways available while movingDepends on signal or map features
Relative and absolute positioning compared

The two are complementary. In ROS-based stacks such as Nav2, a smooth odometry frame provides continuous motion, while a separate map frame correction from AMCL or GNSS removes accumulated drift. A Kalman filter or factor graph typically fuses both sources.

Using dead reckoning in robot navigation

In practice, dead reckoning serves as the motion prediction step of most localization pipelines and as a fallback during outages of absolute sensors. A robot crossing a GNSS-denied tunnel or a featureless hall relies on it until a reliable correction returns.

  • Calibrate systematically: measure wheel diameters and track width with closed-path tests such as the UMBmark square path.
  • Fuse wheel and IMU data: the gyroscope usually gives a better heading than wheel differences on slippery floors.
  • Model uncertainty: propagate a covariance so that planners and filters know how much to trust the estimate.
  • Limit outage duration: define a maximum distance or time without correction before the robot slows or stops.

The decision table below shows which dead reckoning strategy suits common deployment situations.

SituationRecommended approachReason
Warehouse robot on smooth concreteWheel odometry with gyro headingLow slip, inexpensive and accurate
Outdoor robot on grass or gravelVisual-inertial odometryWheel slip makes encoders unreliable
Legged robotLeg kinematics fused with IMUNo wheels, frequent foot impacts
Underwater vehicleDoppler velocity log with IMUNo GNSS signal below the surface
Short GNSS outage in a tunnelIMU and wheel fusion until reacquisitionBridges the gap with bounded drift
Choosing a dead reckoning strategy

Common mistakes with dead reckoning

The most common mistake is trusting dead reckoning for too long without correction. Engineers also underestimate how quickly small heading errors become large position errors over long straight runs.

  • Ignoring wheel slip during acceleration, braking or turning on smooth floors.
  • Using nominal wheel dimensions instead of measured ones, which introduces systematic drift.
  • Integrating raw IMU acceleration without bias estimation, which diverges within seconds.
  • Mixing frames incorrectly, for example treating the odometry frame as a fixed global map frame.

In plain words

Imagine walking through a dark room with your eyes closed. You know where you started, and you count your steps and notice when you turn. From that, you can guess where you are now. That is dead reckoning.

Robots do the same by counting wheel turns and measuring rotation. The guess is good for a short distance, but small mistakes keep adding up. That is why robots regularly check their position against something fixed, such as a map or satellite signals, to correct the guess.

A real-world example

An agricultural robot weeds rows of vegetables using RTK GPS, which normally gives centimeter-level positions. At the end of each row it passes under a line of tall trees, where the satellite fix degrades for about 15 to 20 seconds. During these gaps the robot switches to dead reckoning from wheel encoders and a gyroscope.

At a speed of 0.5 m/s, each gap covers about 10 m. Field tests show a lateral error of roughly 10-20 cm by the time RTK accuracy returns, which is acceptable for turning at the headland but not for weeding inside the row. The operators therefore configured the robot to stop weeding tools during outages longer than 20 seconds, which prevented crop damage while keeping the robot moving.