Describe the process for handling outlier GNSS measurements within a Kalman filter?

• A. Apply residual-based gating (e.g., chi-square test); reject measurements that exceed a threshold; re-run the update with remaining measurements.
• B. Always include all measurements regardless of residuals.
• C. Increase the process noise to absorb outliers.
• D. Remove INS data and rely solely on GNSS.

Answer: (A)

When fusing GNSS data in a Kalman filter, you want the update to be driven by measurements that fit what the filter already predicts. Outliers produce large residuals, which can badly bias the state if they’re treated like normal measurements. The reliable way to handle this is to test each measurement against the predicted innovation.

Compute the residual between what the filter expects and what GNSS reports, and compare its squared Mahalanobis distance to a chi-square threshold based on the measurement dimension. If the residual is too large, that measurement is flagged as an outlier and rejected. After removing the bad measurements, you re-run the update with the remaining data. This keeps the state estimate consistent and accurate by preventing a single dubious measurement from skewing the result.

Choosing this approach over always including all measurements preserves integrity; simply cranking up process noise to absorb outliers degrades accuracy and undermines the model. Relying solely on GNSS while discarding INS data removes valuable information that the fused system would otherwise use, reducing robustness rather than addressing the outliers themselves.