Establishing Prior Probability Distributions for Inline Sensor Calibration
Establishing accurate prior probability distributions for inline sensors prevents false recalibration trips and isolates optical fouling from true fluid drift.
Kalman Filter Covariance
Meaning
Estimation error metrics quantify the uncertainty of the state estimates produced by a recursive state estimation algorithm. Guidance and navigation systems use the kalman filter covariance to determine the level of confidence in estimated positions and velocities on the factory floor. This matrix does not represent the actual physical errors of the system.
Mathematical Update
Computing the uncertainty changes at each time step involves projecting the previous error state forward and combining it with new measurements. The kalman filter covariance shrinks when high-accuracy sensor updates are received and grows during periods without measurements. Relying on these estimates without verifying the noise matrices can cause the filter to diverge from the true physical state.
The cost of incorrect tuning is the collision of automated guided vehicles.
Filter Tuning
Process and measurement noise covariance values are adjusted to match actual operating environments. Tracking the kalman filter covariance reveals if the model is overconfident or too sluggish to respond.
Sensor Integration
Multi-sensor platforms combine data from encoders and laser scanners. Calculating the kalman filter covariance allows the fusion algorithm to weight each sensor according to its current accuracy. This dynamic weighting maintains steering accuracy under varying lighting and surface conditions.