Two Methods ‘for Synthesis of State and Disturbance Observers for an Unmanned Aerial Vehicle

Simultaneous Localization and Mapping (SLAM) and Data Fusion in Unmanned Aerial Vehicles: Recent Advances and Challenges

“…The UAV trajectory and the environment map are referred to as the state [74]. When a set of longitude-latitude pairs describe the altitude of a UAV corresponding to the geographical coordinates, noise (N ) may be carried forward from one timestep to the next, degrading the accuracy of the current estimation [75]. In order to use noisy input to estimate the state of a UAV under uncertainty, evidential belief reasoning techniques such as the Dempster-Shafer theory are used to assign beliefs to the data, as well as to the combination rules to fuse the sensor data [76].…”

“…SLAM for a single UAV finds the solution of the maximum a posteriori (MAP) based on prior distribution obtained by UAV odometry over the trajectory q. The posterior distribution obtained with Bayes' theorem as given in Equations ( 7) and ( 8) is the SLAM likelihood given a certain prior distribution of the UAV motion state [75]. Enough memory is required to maintain the trajectory of the UAV in a given state [59].…”

“…In full connectivity or multiple-hop connectivity, each UAV can directly communicate with other UAVs at any time, or only share information with one UAV at a time [62]. When all the UAVs' initial states can be estimated, the C-SLAM problem is an extension of the single-UAV problem and includes UAVs' states as well as inter-UAV measurements that link multiple UAV maps [75]. As a result, each UAV's neighbors need multiple neighbor-toneighbor transmissions to reach all UAVs [5,61].…”

“…The Equation (3) describes how SLAM addresses the problem of estimating the UAV trajectory q and landmark mapping l. The mapping and localization are done in the presence of noise N in the sensor data d [75]. When the UAV sensors capture landmark data l j at time n and UAV trajectory position q n , the landmark pixel information is Gaussian distributed with the distribution function d n, j , d(•):…”

“…The sensors learn the geometric features, the appearance, and visual context of the landmark scenes in unknown environments. Each observation is a 360 • set of points measured during one sensor rotation, and the SLAM algorithm needs to predict the transformation T = [∆s, ∆θ], where ∆s represents the UAV distance traversed when the UAV coordinates change from s 1 = [s 1x , s 1y , s 1z ] to s 2 = [s 2x , s 2y , s 2z ] and ∆θ is the orientation difference between two consecutive sensor scans for UAV states [75]. The displacement of the UAV relies on the mapping of states to T = [∆s, ∆θ] at time n and the position q x , q y , θ n of the UAV [74,129].…”