Unknown Area Exploration with an Autonomous Robot using Markov Decision Processes

Simon Le Gloannec (1) , Laurent Jeanpierre (2) , Abdel-Illah Mouaddib (2)

1 Cassidian
2 Equipe MAD - Laboratoire GREYC - UMR6072

This paper addresses the problem of exploration of an unknown area by an autonomous robot. The robot decides at each step where it has to move, without any human intervention. The Goal is to gather the maximum information in a minimum time. This work is based on the Markov Decision Processes framework. We focus on the decision problem : the robot must automatically select points where it can maximize its information gain. Even if the goal of the robot is to produce a map of the environment, the localization aspect is not treated in this paper. We divide the environment into three layers. One Layer where the robot moves, another where the map is constructed and a last layer where decisions are taken. As this model is suitable for small robots with cheap sensors we present some experiments on real robots, with very good results.

Domaines: Intelligence artificielle [cs.AI]; Théorie de l'information [cs.IT]; Robotique [cs.RO]; Informatique et théorie des jeux [cs.GT]

Simon Le Gloannec, Laurent Jeanpierre, Abdel-Illah Mouaddib. Unknown Area Exploration with an Autonomous Robot using Markov Decision Processes. TAROS, Guido Bugmann; Tony Belpaeme, Aug 2010, Plymouth, United Kingdom. ⟨hal-01438198⟩