DAVID TOURETZKY
Research Professor, Neural Basis of Cognition and Computer Science
www

My two current research areas are cognitive robotics and computational neuroscience.  

Tekkotsu Framework for Cognitive Robotics. I'm trying to change the way robotics is taught to CS undergraduates. Most robot programming today occurs at a very low level. Tekkotsu ("framework" in Japanese) is a software architecture for mobile robots that draws on ideas from cognitive science to provide high level primitives for perception, manipulation, navigation, and control. We've made the most progress in the area of vision, taking inspiration from Ullman's "visual routines" and Paivio's "dual coding theory" to create a unique approach to parsing the visual world. We hope to similarly operationalize other cognitive science concepts, such "affordances" and "motor schemas".   I teach a cognitive robotics course here at CMU, and I'm collaborating with Spelman College in Atlanta to help introduce cognitive robotics at several other historically black colleges. Working with my graduate student Ethan Tira-Thompson and a stream of undergraduates, we are continuing to develop Tekkotsu's primitives and have begun creating our own robot designs.  

Representations of Space in the Rodent Brain. My students and I are attempting to understand how information is represented in neural tissue by studying the place and head direction systems in the rodent brain. Place cells, found in the hippocampus, fire when a rat is in a particular region of the environment. They appear to be driven by a combination of sensory cues and an internal path integration mechanism in entorhinal cortex. The population activity of a collection of place cells implements a sort of "cognitive map". Head direction cells in thalamus and postsubiculum, which fire when the animal faces in a particular direction, provide an internal compass. Both these systems are thought to be implemented as recurrent neural networks, using attractor dynamics to maintain stability. My work uses computer modeling of neurophysiological recording data in order to understand how these spatial representations are created and updated as the animal moves.   Rats sometimes alter their cognitive maps, or create new ones, in response to changes in their environment. By understanding what underlies these representational changes, we can gain insight into the way that animals conceptualize their world.