Characteristics of Indoor Disaster Environments and their impact on Simultaneous Localization and Mapping for Small Unmanned Aerial Systems

Abstract

This thesis explores the use of small unmanned aerial systems (SUASs) for mapping of unknown disaster environments and investigates the impact of characteristics of such challenging environments on simultaneous localization and mapping (SLAM) algorithm. It provides a formal analysis of indoor disaster environments and identifies four characteristics of a region of space: scale, degree of deconstruction, location of obstacles, and tortuosity. The analysis compares the value of these characteristics for Prop 133 at Disaster City and develops computer simulated environments. Furthermore, a SLAM algorithm for SUAS flying in indoor disaster environments is developed and the system is tested in these virtual environments. Three different environments with increasing deconstruction are designed. For each type of environment, 10 different maps with a common floor plan are simulated with randomly placed obstacles. For each map, three trials with varying flight paths are run, thus conducting 90 trials of experimentation. As verified from the statistical testing, there is a convincing increase of 26.36% in the average value of RMSE as the deconstruction changes from Group 1 to Group 3. But, the change in value of error is not statistically convincing when Group 1 and 2 and, Group 2 and 3 are respectively compared. Hence, though the result suggest that the value of error increases between different groups, it cannot be claimed that the RMSE in localization will always increase with deconstruction. The tortuosity increases with deconstruction and this value is empirically calculated. The average RMSE in localization does not change as the Agent to Environment ratio changes. These results can help identify the remaining gaps in the state of the art indoor SUAS for disasters.