Next-Best-View Planning for Exploration and Autonomous Object Modelling in Static Environments with Irregular Depth Noise using Interval Probabilities

Abstract

This thesis examines Next-Best-View planning for exploration and autonomous 3D-modelling in static environments with depth noise using Interval Probabilities. An autonomous exploration algorithm is extended to cope better with depth noise from non-diffusely reflecting environments. The measured data is interpreted by a novel voxel space update which combines a state-of-the-art Bayes Update on three dimensional occupancy maps with the concept of Interval Probabilities. Thereby an additional informational level of uncertainty is introduced. An adapted next-best-view criterion, which calculates the maximum entropy of each volume element, aids the exploration process. Experiments are conducted in a simulation to examine how to take advantage of the possibilities of Interval Probabilities. The results are compared to an implementation of the Bayes Update at the DLR-RMC. The method is then transferred to object modelling. A robotic industrial arm named KUKA KR16 is used to conduct further experiments to verify whether the results from the simulation are applicable to real-life environments. It is concluded that the novel update is especially advantageous in dealing with specular reflections.