La carte bayésienne -- Un modèle probabiliste hiérarchique pour la navigation en robotique mobile

Abstract

What is a map? What is its utility? What is a location, a behaviour? What<br />are navigation, localization and prediction for a mobile robot facing a<br />given task? <br /><br />These questions have neither unique nor straightforward answer to this day,<br />and are still the core of numerous research domains.<br /><br />Robotics, for instance, aim at answering them for creating successful<br />sensori-motor artefacts. Cognitive sciences use these questions as<br />intermediate goals on the road to understanding living beings, their skills,<br />and furthermore, their intelligence.<br /><br />Our study lies between these two domains. We first study classical<br />probabilistic approaches (Markov localization, POMDPs, HMMs, etc.), then<br />some biomimetic approaches (Berthoz, Franz, Kuipers). We analyze their<br />respective advantages and drawbacks in light of a general formalism for<br />robot programming based on bayesian inference (BRP).<br /><br />We propose a new probabilistic formalism for modelling the interaction<br />between a robot and its environment: the Bayesian map.<br /><br />In this framework, defining a map is done by specifying a particular<br />probability distribution. Some of the questions above then amount to solving<br />inference problems.<br /><br />We define operators for putting maps together, so that "hierarchies of maps"<br />and incremental development play a central role in our formalism, as in<br />biomimetic approaches. By using the bayesian formalism, we also benefit both<br />from a unified means of dealing with uncertainties, and from clear and<br />rigorous mathematical foundations. Our formalism is illustrated by<br />experiments that have been implemented on a Koala mobile robot.