Introducing Vehicle Dynamic Models in Dynamic Networks for Navigation in UAVs

Estimation of the trajectory is a fundamental problem in robotics. Introduction of additional measurements in a robotic platform reduces the uncertainty in the trajectory estimate. The limitations on the power and payload in a UAV platform advocates for the usage of already existing inputs such as Vehicle Dynamic Model over the addition of new sensors. These inputs from the Vehicle Dynamic Model are independent from the lighting and texture of the surrounding environment. Different methods exist in the literature for multi-sensor fusion problems. The conventional methods of sensor fusion are outperformed by the graph-based methods such as Dynamic Network, which could model observations by constraining them in space or time. In this work we investigate the improvement gained by introducing the Vehicle Dynamic Model as measurements in a Dynamic Network for a Quadcopter. We investigate two separate studies - i) estimation of the trajectory with the known sensor parameters, ii) estimation of trajectory with unknown sensor parameters. In the second study the effect of correlation between the parameters on the estimated parameter value is studied. The observation model required for using Vehicle Dynamic Model as measurement was derived and was tested on simulated data.