Perceptual Science Series
Dual-Coding Representations for Robot Vision
Dr. Dave Touretzky
Monday, March 10, 2008, 01:00pm - 02:00pm
Carnegie Mellon University, Computer Science Department & Center for the Neural Basis of Cognition
Tekkotsu is an application development framework for mobile robots thatincludes extensive facilities for visual perception. One of Tekkotsu'skey ideas is the use of s "dual-coding" strategy inspired by AlanPaivio's theory of human mental representations. According to Paivio,people utilize both lexical (symbolic) and picture-based (iconic)representations, with extensive referential connections between them.Analogously, when Tekkotsu programmers write algorithms to parse thevisual world, they use both "shape" and "sketch" representations, andcan move freely between the two. Other aspects of Tekkotsu's perceptualsystem include the use of kinematic information to automatically mapbetween camera, body, and world coordinates, automated map construction,and the control of visual attention. Using this framework,undergraduates taking their first robotics course can easily constructsophisticated robot behaviors.
Background reading: Touretzky, D.S., Halelamien, N.S., Tira-Thompson, E.J., Wales, J.J., and Usui K. (2007) Dual-coding representations for robot vision in Tekkotsu. Autonomous Robots, 22(4):425-435. http://www.cs.cmu.edu/~dst/pubs/touretzky-autrob-2007.pdf