Probabilistic 3D object model generation from noisy measurements for manipulator motion planning

U ovom radu prikazan je postupak dobivanja 3D modela iz kontaktnih mjerenja optičkog taktilnog senzora generiranih u simulaciji. Model opne senzora je izrađen i objašnjen je postupak dobivanja istog. Simulacija opne senzora napravljena je u prilagođenoj verziji Unreal Enginea s NVIDIA Flex proširenjem. Prikazan je put od modela, preko deformacije modela do skupljanja mjerenja. Prikupljena su dva skupa podataka: oblak točaka kao kontaktna mjerenja predmeta s pozicijom i orijentacijom, te slike opne pod deformacijom prizmatičnog predmeta u različitim pozicijama i orijentacijama. Oblaci točaka su prikazani u trodimenzionalnom prostoru, te su kreirani odgovarajući 3D modeli iz tih mjerenja koji su približni originalnim predmetima istraživanja. Prikazano je i poklapanje različitih mjerenja oblaka točaka.In this paper, the procedure of obtaining a 3D model from contact measurements of optical tactile sensor generated in the simulation is presented. A model of the sensor membrane was made and the procedure for obtaining it was explained. The sensor membrane simulation was made with a customized version of Unreal Engine with NVIDIA Flex extension. The journey from the model, through the deformation of the model, to the measurement collection is described. Two data sets were collected: point clouds as contact measurements of an object with positions and orientations, and images of the membrane under the deformation of a prismatic object in different positions and orientations. Point clouds were displayed in three-dimensional space, and corresponding 3D models were created from these measurements. Point cloud matching of different measurements was also shown