SCS Undergraduate Thesis Topics
|Neil Halelamien||David Touretzky||Visual Routines for Spatial Cognition on a Mobile Robot|
In his studies on visual cognition, Ullman (1984) proposed that high-level vision makes use of reusable elemental operators which may be sequentially composed into a "visual routine" specialized for a particular visual task. Seeking to explore Ullman's ideas in the domain of robotics, we have created a framework for robot vision programming which uses reusable operators to parse the visual world. These visual operators act on and transform between two different representations of the world: iconic (pixel-based) and symbolic (shape-based). The framework provides a powerful set of primitives for the construction of visual behaviors, while also providing a GUI which allows for the run-time visualization of intermediate representations. The framework has been implemented in simulation and on a Sony AIBO robot, to perform tasks such as tic-tac-toe board parsing and character recognition. Current research is focused on using camera information to generate a persistent top-down world representation, allowing for more elaborate spatial behaviors.