Alexander Maye - eSMCs computational models
Computational models of sensorimotor contingencies
Alexander Maye, University Medical Center Hamburg-Eppendorf
The tutorial gives an overview of different approaches to formalize sensorimotor contingencies and to use them for robot control. First an introduction to the concept of Sensori-Motor Contingencies (SMCs) as an account of perception and sensory awareness (O'Regan and No√´, 2001) is given. It is shown that SMCs can be conceived as probabilities over joint action and sensory observations, conditioned on the history of an agent's sensorimotor flow. A corresponding model that was developed in the EU-funded project "Extending sensorimotor contingencies to cognition (eSMCs)" is presented in detail together with results from robotic experiments. In the second part an overview of previous approaches to ground the concept of SMCs in artificial agents is given, comprising short introductions to predictive state representations, Braitenberg vehicles, sensorimotor invariances, and simulation based approaches.