$\delta$ -Generalised Labelled Multi-Bernoulli Simultaneous Localisation and Mapping
Diluka Moratuwage, Martin David Adams, Felipe Inostroza · 2018
Motivated by the need for simultaneous localisation and mapping (SLAM) algorithms which circumvent the requirement of external data association routines and map management heuristics, and account for realistic sensor detection uncertainty, recent literature has adopted Random Finite Set (RFS) based approaches. Solutions based on the Probability Hypothesis Density (PHD) filter and more recently the Labelled Multi-Bernoulli (LMB) filter have been demonstrated. The LMB filter was introduced as an efficient approximation of the computationally expensive δ-Generalised LMB ( δ-GLMB) filter. The LMB filter converts its representation of an LMB distribution to δ-GLMB form and back during the measurement update step. This conversion results in a loss of information and in general yields inferior results compared to the δ-GLMB filter. To address this issue, we present a SLAM solution using an efficient variant of the δ-GLMB filter ( δ-GLMB-SLAM) based on Gibbs sampling, which is computationally comparable to LMB-SLAM, yet more accurate and robust against sensor noise, measurement clutter and feature detection uncertainty. The performance of the proposed δ-GLMB-SLAM algorithm is compared to the LMB-SLAM algorithm with a Gibbs sampling based joint map prediction and update approach using a series of simulations.