Evaluation of Cooperative UAV-UGV Navigation Strategies with Maneuvering UAVs

Abstract

This paper presents and evaluates cooperative navigation methods used to reduce navigation solution error growth between members of an unmanned aerial vehicle - unmanned ground vehicle (UAV-UGV) team when Global Position System (GPS) measurements are partially or completely unavailable to the group. A scenario in which two UAVs and one UGV travel in a cooperative team was simulated with a centralized navigation algorithm based on the Extended Kalman Filter (EKF) and with a decentralized navigation algorithm based on the Covariance Intersection (CI) filter. Measurements including relative range, relative range-rate, and relative bearing were made available to the vehicles in different simulation runs to compare their impact on navigation state observability and navigation state estimation accuracy. The UAVs were also guided along varied trajectories of a “spiral" class during different simulation runs to investigate whether estimation accuracy can be improved by varying intervehicle dynamics and geometry. To analyze the observability of the studied scenarios, a condition number test was performed on the observability Gramian matrix. This study indicates that the navigation state observability in cooperative navigation scenarios where a kinematic vehicle model is aided with relative measurements can be improved by the proposed vehicle maneuver. As the rate of the proposed spiral maneuver is increased, this analysis suggests an improvement in observability. This result is further validated in the simulated results which show that with relative bearing only, even low rates of inter-vehicle spiral motion allow for estimates of relative position with less than 3 meters of error. As the spiral rate increases, accurate relative positioning is shown to be possible with only relative range measurements. The loss of inter-vehicle covariance information encountered in the decentralized case resulted in a significant reduction of positioning accuracy that varied depending on vehicle maneuver and available relative measurements. Lastly, an experimental hardware implementation of the simulated scenarios was performed to validate the simulation results. This experiment demonstrated similar results to the simulated scenarios. Relative position error was reduced from over 100 meters to sub-meter accuracy, depending on relative measurement availability. Absolute error was also reduced from over 70 meters (in the IMU-only case) to meter-level accuracy depending on measurement availability.