Gaussian process occupancy maps

Abstract

We introduce a new statistical modelling technique for building occupancy maps. The problem of mapping is addressed as a classification task where the robot's environment is classified into regions of occupancy and free space. This is obtained by employing a modified Gaussian process as a non-parametric Bayesian learning technique to exploit the fact that realworld environments inherently possess structure. This structure introduces dependencies between points on the map which are not accounted for by many common mapping techniques such as occupancy grids. Our approach is an 'anytime' algorithm that is capable of generating accurate representations of large environments at arbitrary resolutions to suit many applications. It also provides inferences with associated variances into occluded regions and between sensor beams, even with relatively few observations. Crucially, the technique can handle noisy data, potentially from multiple sources, and fuse it into a robust common probabilistic representation of the robot's surroundings. We demonstrate the benefits of our approach on simulated datasets with known ground truth and in outdoor urban environments.

Mapping, Planning and Exploration with Pose SLAM

“…The fact that entropy reduction in the path can be computed online and at high frame rate is thanks to the use of Pose SLAM as the estimation workhorse, but unfortunately, in Pose SLAM, the map posterior is marginalized out and needs to be computed to evaluate exploration candidates. Doing so at a high frame rate can be achieved with a more scalabe alternative to gridmaps, such as the one presented in [125].…”

Exploration on continuous Gaussian process frontier maps

“…Recent developments in Bayesian regression and classification methods, particularly from the machine learning community, are providing strong mathematical tools for continuous learning and inference in complex data sets. Nonparametric kernel models, such as Gaussian processes (GP), have proven particularly powerful to represent the affinity of spatially correlated data, overcoming the assumption of independency between cells [9]. The GP associated variance surface equates to a continuous representation of uncertainty in the environment, which can be used to highlight unexplored regions and optimize a robot's search plan.…”

“…However, training a unique GP for both occupied and free areas results in a mixed variance surface and it is not possible to disambiguate between boundaries of occupiedunknown and free-unknown space without thresholding of the continuous map (see Fig. 6 in [9]). Moreover, it limits selection of an appropriate kernel and results in extrapolated obstacles or low quality free areas.…”

“…Training points for the free area GP map are sampled along the laser beam between the robot and the sensed obstacles as in [9]. Computing the obstacles GP map is more straight forward, as it is possible to use the global measured points directly.…”

Learning spatial correlations for Bayesian fusion in pipe thickness mapping

“…In terrain modelling GPs have been used before to model and use the spatial correlation of the given data to estimate the elevation values for other unknown points of interest [8], [9] and more recently [10]. A recent approach for occupancy mapping with GPs has also been presented in [11].…”