Iterated Conditional Modes to solve Simultaneous Localization and Mapping in Markov Random Fields context

GIMENEZ J; A. N. AMICARELLI; JM TOIBERO; DI SCIASCIO FERNANDO AGUSTÍN

International Journal of Automation and Computing

Springer

Año: 2018 vol. 15 p. 310 - 324

1476-8186

This paper models the complex SLAM problem through a very fexible Markov Random Field and then solves it by using the Iterated Conditional Modes algorithm. Markovian models allow to incorporate: any motion model; any observation modelregardless of the type of sensor being chosen; prior information of the map through a map model; maps of diverse natures; sensor fusion weighted according to the accuracy. On the other hand, the Iterated Conditional Modes algorithm is a probabilistic optimizer widely used for image processing which has not yet been used to solve the SLAM problem. This iterative solver has theoretical convergence regardless of the Markov Random Field chosen to model. Its Initialization can be performed on-line and improved by parallel iterationswhenever deemed appropriate. It can be used as a post-processing methodology if it is initialized with estimates obtained from another SLAM solver. The applied methodology can be easily implemented in other versions of the SLAM problem, such as for instance the multi-robot version or the SLAM with dynamic environment. Simulations and real experiments show the flexibility and the excellent results of this proposal.