Odometry

See also:  Egomotion (vs odometry)

Source: https://en.wikipedia.org/wiki/Dead_reckoning

In navigation, dead reckoning is the process of calculating one’s current position by using a previously determined position, by using estimations of speed and course over elapsed time

• s. Brian Douglas video on sensor fusion

Source: Wikipedia Visual odometry

• Data can be generated from actuator movements, e.g. rotary encoders that measure motor shaft rotations
• This data can be used to estimate changes in position over time
• Usually has precision problems, e.g. due to wheels slipping and sliding, bumpy surfaces
• The errors are integrated over time and therefore get worse

Source: cometlabs

• Acceleration is obtained: integrate to get velocity, displacement [estimates]
• However, as the estimates drift over time and get integrated, this leads to increased errors
• Subject to
  • non-systematic errors e.g. human intervention
  • systematic errors, e.g. due to imperfections in robots' structure
• Examples:
  • wheel odometry, via encoders – error source: wheel slippage
  • IMU : measures lin and rotational acceleration, error sources: extensive drift, sensitive to bumpy terrain

Source: SLAM for Dummies

• Provides approximate position of the robot as measured by its relative movement
• Acts as an initial guess for the EKF

Problem: getting timing right between odometry data and laser data, so as not to compare outdated data to newer data Solution: extrapolate the data, easiest to extrapolate odometry data (since the controls are known)