In-Time UAV Flight-Trajectory Estimation and Tracking Using Bayesian Filters

The expected increase of UAV operation in the next decade in areas of on-demand delivery, medical transportation services, law enforcement, traffic surveillance and several others pose potential risks to the low-altitude airspace above densely populated areas. Safety assessment of airspace demands the need for novel UAV traffic management frameworks for regulation and tracking of vehicles. Particularly for low-altitude UAV operations, quality of GPS measurements used for guidance, navigation, and control, is often compromised by loss of communication link caused by presence of trees or tall buildings in proximity to the UAV flight path. Inaccurate GPS measurements may yield unreliable monitoring and inaccurate prognosis of vehicle components such as remaining battery life and other safety metrics that rely on future expected trajectory of the UAV. This work therefore proposes a generalized trajectory for in-time monitoring and prediction of autonomous UAVs using GPS measurements. Firstly, a smooth 4D trajectory generation technique is presented using a series of waypoint locations with associated expected times-of-arrival based on B-spline curves. Initial uncertainty in the vehicle's expected cruise velocity is propagated through the trajectory to compute confidence intervals along the entire flight trajectory using error interval propagation approach. Further, the generated planned trajectory is considered as the prior knowledge that is updated during the UAV flight with incoming GPS measurements in order to refine its current location estimates and corresponding kinematic profiles. The estimation of the vehicle position is defined in a state-space representation such that the position at a future time step is derived from the position and velocity at the current time step and expected velocity at the future time step. A linear Bayesian filtering algorithm is employed to efficiently refine position estimation from noisy GPS measurements and update the confidence intervals. Further, a dynamic re-planning strategy is implemented to incorporate unexpected detour or delay scenarios. Finally, critical challenges related to uncertainty quantification in trajectory prognosis for autonomous vehicles are identified, and potential solutions are discussed at the end of the paper. The entire monitoring framework is demonstrated on real UAV flight experiments conducted at the NASA Langley Research Center.