Probabilistic Algorithms for Modeling Building Interiors With Mobile Robots

Sebastian Thrun
School of Computer Science
Carnegie Mellon University

Time and Place

1305 Newell-Simon Hall
Refreshments 3:15 pm
Talk 3:30 pm

Abstract

Wouldn't it be neat if we had low-cost robots that could autonomously explore buildings and generate accurate volumetric maps of their interiors? Such technology could provide great benefits to realtors, architects, disaster relief units, the military, and video game designers. This talk presents a probabilistic approach to autonomous multi-robot exploration, localization, and mapping. Building on Bayes rule, our approach uses fast probabilistic particle filtering methods for localizing robots during mapping, and the expectation-maximization algorithm for extracting low-complexity 3D models from range and camera data. Results obtained in Wean Hall and various DARPA demonstrations will illustrate some of the advantages and shortcomings of this work.

Speaker Biography

Sebastian Thrun has been a student all his life. He studied computer science and statistics at the Universities of Hildesheim and Bonn (Germany). Presently, he is with Carnegie Mellon University where he studies robotics, machine learning, and probabilistic AI.

Speaker Appointments

For appointments, please contact the speaker (thrun@cs.cmu.edu).