This thesis investigates efficient forms of Simultaneous Localization and Mapping (SLAM) that do not require explicit identification, tracking or association of map features. The specific application considered here is subsea robotic bathymetric mapping. In this context, SLAM allows a GPS-denied robot operating near the sea floor to create a self-consistent bathymetric map. This is accomplished using a Rao-Blackwellized Particle Filter (RBPF) whereby each particle maintains a hypothesis of the current vehicle state and map that is efficiently maintained using Distributed Particle Mapping. Through particle weighting and resampling, successive observations of the seafloor structure are used to improve the estimated trajectory and resulting map by enforcing map self consistency.
The main contributions of this thesis are two novel map representations, either of which can be paired with the RBPF to perform SLAM. The first is a grid-based 2D depth map that is efficiently stored by exploiting redundancies between different maps. The second is a trajectory map representation that, instead of directly storing estimates of seabed depth, records the trajectory of each particle and synchronises it to a common log of bathymetric observations. Upon detecting a loop closure each particle is weighted by matching new observations to the current predictions. For the grid map approach this is done by extracting the predictions stored in the observed cells. For the trajectory map approach predictions are instead generated from a local reconstruction of their map using Gaussian Process Regression. While the former allows for faster map access the latter requires less memory and fully exploits the spatial correlation in the environment, allowing predictions of seabed depth to be generated in areas that were not directly observed previously. In this case particle resampling therefore not only enforces self-consistency in overlapping sections of the map but additionally enforces self-consistency between neighboring map borders.
Both approaches are validated using multibeam sonar data collected from several missions of varying scale by a variety of different Unmanned Underwater Vehicles. These trials demonstrate how the corrections provided by both approaches improve the trajectory and map when compared to dead reckoning fused with Ultra Short Baseline or Long Baseline observations. Furthermore, results are compared with a pre-existing state of the art bathymetric SLAM technique, confirming that similar results can be achieved at a fraction of the computation cost.
Lastly the added capabilities of the trajectory map are validated using two different bathymetric datasets. These demonstrate how navigation and mapping corrections can still be achieved when only sparse bathymetry is available (e.g. from a four beam Doppler Velocity Log sensor) or in missions where map overlap is minimal or even non-existent.