Construction of a terrain traversability map from UAV imagery for route planning of a screw-propelled robot

Abstract

A screw propulsor provides high bearing traversability over water-saturated, marshy and snow-covered ground, yet it loses mobility on dry compacted and stony surfaces; consequently, conventional traversability maps calibrated for wheeled and tracked platforms become unsuitable for it, since the cells penalised by the usual model as hard to traverse are precisely the ones preferable for the screw-propelled vehicle. The mismatch between the standard cost labelling of terrain and the inverse mobility of the screw propulsor yields routes that are formally short but physically infeasible. The aim of the work is to construct a traversability map from aerial survey imagery captured by an unmanned aerial vehicle, with a cost labelling tuned to the mobility profile of the screw-propelled robot, and to plan a route over such a map. An author's computational model was built: an orthophoto and a digital elevation model recovered photogrammetrically are converted into a movement cost raster through surface-type classification and local slope estimation, after which the route is found by a weighted search algorithm on the resulting raster. A model computation over a scene of 1.24 km² showed that the route laid out along the inverse traversability map is about 18,6% longer than the geometrically shortest one, whereas its integral movement cost is 41,3% lower, and the share of the track passing over soft ground favourable to the screw rises from 31,7 to 78,4%. The result is most sensitive to the weight of slope in the cost function and to the resolution of the digital elevation model; when the elevation grid is coarsened from 0.25 to 1.0 m, the share of soft ground along the track decreases non-monotonically. The obtained relations show that the suitability of a route for a screw-propelled vehicle is governed by the agreement of the cost labelling with the inverse character of the propulsor's bearing traversability, whereas geometric length plays a secondary role.