Suboptimal Multiple Model Filter for Mobile Robot Localization

Abstract

The problem of mobile robot localization by using sensor information appeals to different communities since the need for accurate position has become crucial for many robot subtasks. The Kalman filter (KF) has been acknowledged as an appropriate tool for a suitable dynamic combination of the different measurements using the state and measurement models. However, when there are discrete uncertainties about the models, without additional restrictions, the performance of KF degrades drastically as the predicted estimate tends to be updated by a wrong measurement. This paper focuses on a suboptimal multiple model filter to deal with the associated measurement/model. Experimental and simulated results are achieved, which prove the feasibility of the proposal. The experimental setup consists of a structured environment constituted of elementary features such as walls and corners, while a possibly unmodeled obstacle may be encountered. The robot is equipped with odometry and a set of ultrasonic sensors.

Mobility Assistance for Older People

“…By means of an interacting multiple model (IMM) algorithm [2,4,5,19,27] running several Kalman filters [14] representing the different systems, it is possible to achieve a seamless localisation system. Figure 4 shows the implemented architecture in this work for the IMM algorithm.…”

Robust Position Estimation of an Autonomous Mobile Robot

“…The main idea of the approach proposed here is to consider the localization process as a hybrid process which evolves according to a model among a set of models with jumps between these models according to a Markov chain (Djamaa & Amirat, 1999) (Djamaa, 2001). A close approach for multiple model filtering is proposed in (Oussalah, 2001). In our approach, models refer here to both state and observation processes.…”

Appearance-based hierarchical localization and environment modeling for mobile robot

“…However, when there are discrete state and measurement models, without additional restrictions, the performance of KF degrades drastically as the predicted estimate tends to be updated by wrong measurements. Oussalah [75] uses more than one single model to increase the robustness of tracking.…”