home · mobile · calendar · colloquia · 2008-2009 · 

Colloquium - Beeson

From Sensors to Symbols: Building Cognitive Maps for Mobile Robots
University of Texas at Austin
3/10/2009
3:30pm-4:30pm

Autonomous robots need to generate and maintain their own models of the uncertain world. Of particular interest to many mobile robot researchers is the domain of spatial reasoning. The overarching issue here is how a robot can build and maintain a model of its environment (a map) that facilitates reliable navigation. That is, how can the robot build a useful "cognitive map"?

There are two major paradigms in mobile robot mapping research. Metrical mapping methods are good at overcoming noise and errors in modern range sensors (lidar, sonar, stereo vision). These methods are computationally expensive; thus, they do not scale well to large environments, especially those with loops. Topological methods generate compact, efficient hierarchical maps that support multi-level symbolic reasoning for robust navigation, planning, and communication. The major hurdle for topological map-building has been the reliable abstraction of symbolic places and paths from the continuous, noisy "pixel-level" perception of the robot.

In this talk, I will discuss the Hybrid Spatial Semantic Hierarchy (HSSH). This framework for spatial reasoning handles qualitatively different types of uncertainty by combining the strengths of topological and metrical approaches. The HSSH factors the mapping problem into natural sub-goals: maintaining a metrical representation for the local space surrounding the robot; generating the symbolic description of the current local space from the metrical model; combining the local symbolic descriptions gathered during exploration into a topological map that represents the qualitative structure of the entire, large-scale space; and (when necessary) constructing a large, global metrical representation for the environment using the skeleton provided by the topological map.

Hosted by James Martin.

Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
webmaster@cs.colorado.edu
www.cs.colorado.edu
May 5, 2012 (14:13)
XHTML 1.0/CSS2
©2012