Bayesian Modeling and Reasoning for Real World Robotics: Basics and Examples

David Bellot (1) , Roland Siegwart (1) , Pierre Bessiere (1) , Adriana Tapus (1) , Christophe Coué (1) , Julien Diard (1)

1 E-MOTION - Geometry and Probability for Motion and Action

Cognition and Reasoning with uncertain and partial knowledge is prob- ably the biggest challenge for autonomous mobile robotics. Previous robotics sys- tems based on a purely logical or geometrical paradigm are limited in their ability to deal with partial or uncertain knowledge, adaptation to new environments and noisy sensors. Representing knowledge as a joint probability distribution increases the possibility for robotics systems to increase their quality of perception on their environment and helps them to take the right actions towards a more realistic and robust behavior. Dealing with uncertainty is thus a ma jor challenge for robotics in a real and unconstrained environment. Here, we propose a new formalism and method- ology called Bayesian Programming which aims at the design of efficient robotics systems evolving in a real and uncontrolled environment. This original formalism will be exemplified by two interesting experiments where robots are driven by a Bayesian Program (BP). These examples represents situations where the robot can sense only a small part of its global environment using noisy sensors. The second fact about these environments is they cannot be constrained so that to ease the control of the robot.

Domaines: Autre [cs.OH]

David Bellot, Roland Siegwart, Pierre Bessiere, Adriana Tapus, Christophe Coué, et al.. Bayesian Modeling and Reasoning for Real World Robotics: Basics and Examples. Springer-Verlag. Embodied Artificial Intelligence Int. Seminar, 3139, Springer-Verlag, pp.186--201, 2004. ⟨inria-00182055⟩