Generation of Non-Coordinate Navigation Knowledge from a Flow of Input Views
Josué Figueroa-González, Georgii Khachaturov, Juan Manuel Martínez-Hernández · Research in Computing Science · 2019
An approach to automatic generation of a network of reference 'views' specifically structured for non-coordinate robot navigation is presented.The reference views are selected from the input flow while the robot moves along a random continuous trajectory in its environment.The network is organized as an atlas that represents inner model of the robot environment.It accumulates a 'rich' navigation knowledge extracted from the flow of 'poor' views.No coordinate transformation is involved, but fundamental topological concepts: the continuity of a robot trajectory and discontinuities ('heavy changes') detected in the input flow.Avoiding any explicit coordinate transformation becomes possible using a local inversion of the knowledge hidden in the relation 'robot control -change of view'.Both techniques -the inversion and detection of heavy changes -are specific for a particular robot architecture.The approach was verified on a virtual static 2D-workspace.