Sensorimotor Categorization

For autonomous agents, whether human or machine, sensorimotor categories provide an essential means for recognizing situations, for predicting the effects of actions, and for building concepts that form the basis of higher-level reasoning and planning. Mobile robots, for example, require a level of abstraction away from raw sensor readings when operating in environments of even modest complexity. The objective of this work is that agents discover such categories for themselves, without supervision, so they are better equipped to handle complex and possibly changing environments. My proposed solution was inspired by the method of delays, a nonlinear dynamics technique for producing spatial representations of time-based data.

Presentations

Concepts, Clusters and Time Series

Symbol Grounding With Delay Coordinates

Activity Maps

Related Publications
Continuous categories for a mobile robot
	M.T. Rosenstein and P.R. Cohen. In Proceedings of the Sixteenth National Conference on Artificial Intelligence, 634-640, 1999. [pdf] [ps.gz]
Symbol grounding with delay coordinates
	M.T. Rosenstein and P.R. Cohen. In AAAI Technical Report WS-98-06, The Grounding of Word Meaning: Data and Models, 20-21, 1998. [pdf] [ps.gz]
Concepts from time series
	M.T. Rosenstein and P.R. Cohen. In Proceedings of the Fifteenth National Conference on Artificial Intelligence, 739-745, 1998. [pdf] [ps.gz]
Learning what is relevant to the effects of actions for a mobile robot
	M.D. Schmill, M.T. Rosenstein, P.R. Cohen, and P.E. Utgoff. In Proceedings of the Second International Conference on Autonomous Agents, 247-253, 1998. [pdf] [ps.gz]
Action representation, prediction and concepts
	M.T. Rosenstein, P.R. Cohen, M.D. Schmill, and M.S. Atkin. In Working Notes of the AAAI Workshop on Robots, Softbots, Immobots: Theories of Action, Planning and Control, 1997. [pdf] [ps.gz]

updated 06-Aug-2002 mtr@cs.umass.edu