Learning to Behave in the Sensor-Motor Loop - An Animat Approach
Because animats - as well as animals - act in a sensorimotor loop, a learning process will have to generate solutions to problems which are posed by their environments under the physical conditions of their bodies and the neural predisposition of their control instances. In general the property of ``being a solution'' to these problems can not be predicted or explained at an abstract level, but has to be formed during continues interactions of internal with external processes. Thus, learning will be associated with processes driven by signals in the sensorimotor loop.
The project will generate and identify artificial neural structures and mechanisms which
endow animats with learning capacities, such that their behaviour resembles that of animals. Considered mechanisms may involve homeostatic properties of single neurons, proprioceptive signals and specific functional submodules. Assuming that neural structure is as relevant for learning as synaptic plasticity, evolutionary techniques for combining network structure generation with synaptic weight dynamics will be applied.
Thus, learning will be explored in the context of a dynamical systems approach to cognition, as-
suming that brains, as systems for behaviour regulation, can be understood as embodied adaptive
dynamical systems. The project will use physical simulations of anmats situated in
various changing environments, and will apply techniques from Neural Networks, Evolutionary
Robotics, Artificial Life, Dynamical and Complex Systems Theory.
More detailed information will be added soon! Please come back later!
- Prof. Dr. Frank Pasemann
- Dr. Christian Rempis
- Hazem Toutounji