Approximating UAV and Vision Feature Point Correlations in a Simplified SLAM problem

Navigation with a range sensor and vision aided inertial measurement unit (IMU) estimation is difficult in Global Positioning System (GPS) denied environments. Ignoring vision feature point and vehicle state correlations contributes to inaccuracy and filter inconsistency. Approximation of feature point and vehicle cross correlation terms would allow the accuracy and consistency comparable to a correlated solution whilst reducing operation count and allowing for decoupled filter design. A Monte-Carlo simulation for a two dimensional bearing to feature point approximation of the simultaneous localization and mapping (SLAM) problem was developed. The results of a least absolute shrinkage and selection operator (LASSO) regression were then used to estimate cross covariance terms. A 1000 trial simulation showed that the regression solution was comparable in accuracy and consistency to the fully correlated solution. Future developments have the potential to provide a more accurate, approximately correlated SLAM solution to bound IMU drift for UAVs operating in a GPS denied environment.