Incorporating estimated feature descriptor information into Rao Blackwellized-PHD-SLAM

Abstract

Recently, various techniques which adopt Random Finite Set (RFS) based techniques for the solution of the fundamental, autonomous robotic, feature based, Simultaneous Localization and Mapping (SLAM) problem, have been proposed. In contrast to their vector based counterparts, these techniques offer the advantage that feature detection, as well as the usually considered spatial, statistics can be incorporated into the Bayesian recursion in a joint manner. Most of the proposed solutions are based on the Probability Hypothesis Density (PHD) filter approximation of an RFS estimator. With the aim of further improving such solutions, this article demonstrates the importance of modelling feature detection uncertainty, based on the commonly used range/bearing sensors such as laser range finders used in robotics. In particular, a feature descriptor is defined, based on the number of unoccluded range/bearing values which can be estimated via ray-tracing techniques from estimated SLAM robot pose/feature coordinates. A modified version of the PHD corrector equation is introduced, which incorporates this extra information. An example of such a descriptor, based on the center location and radii of trees in a park, is demonstrated, and statistical information obtained from such an environment is used in a SLAM simulation. This demonstrates the potential of achieving superior SLAM performance, when feature descriptor statistics are incorporated directly into the PHD filter update stage.