Brief
GNSS (GPS and similar satellite systems) locates a receiver from signal travel times. Start from raw Android GNSS data from the Google Smartphone Decimeter Challenge 2022: pseudorange (measured satellite-to-phone distance, including clock error), Doppler (frequency shift, which gives velocity) and carrier-phase data (a much finer but ambiguous distance measurement). Compute satellite positions and corrections, such as satellite clock and atmospheric delay. Then compute weighted least-squares position fixes with robust outlier rejection: Huber weighting (reduces the influence of large errors) and chi-square gating (drops measurements that are statistically too far off). Add an EKF plus RTS smoother that fuses Doppler/TDCP (time-differenced carrier phase, which measures movement between epochs very precisely) and IMU (accelerometer and gyroscope) data, with noise parameters fit by maximising innovation likelihood. No RTKLIB or other positioning library in the solution path.
VICTORY CONDITIONS
- Late-submission private score inside the top 10% of the final leaderboard. Confirm late submissions are open before starting; otherwise score on a held-out training trace with released ground truth. (Official competition metric on the private leaderboard (late submission)).
- A public repository with a README that shows your result and how to reproduce it.
Terms in plain words
- GNSS
- Global navigation satellite systems such as GPS, Galileo and BeiDou, which position a receiver from satellite signal timing.
- Pseudorange
- The distance to a satellite computed from signal travel time, called pseudo because it includes the receiver's unknown clock error.
- Carrier phase / TDCP
- Distance measured from the phase of the radio wave itself, precise to millimetres but with an unknown whole number of cycles; differencing it over time cancels that unknown.
- Huber loss
- A loss that is quadratic for small errors and linear for large ones, so outliers pull the fit less.
- Chi-square gating
- Rejecting a measurement when its normalised squared error is larger than a chi-square threshold.
- Sensor fusion
- Combining several kinds of sensors, here satellites and the phone's motion sensors, into one estimate.
Why it rates ★★★★☆
A top-10% result on a real sensor-fusion problem using hand-built estimators is concrete and checkable.