event
Thursday 31 May 2012: Evolving Modularity in Robot Behaviour Using Gene Expression Programming
Jonathan Mwaura - University of Exeter
Harrison 170 15:00-16:00
An important issue in behaviour based robotics (BBR) and ethology (study of animal behaviour) is behaviour coordination or action selection (Prescott et el 2009, Wahde 2004). This refers to the process of determining which behaviour, from a set of behaviours, to activate at a particular moment in time. In BBR, the mechanisms of robot behaviour coordination are categorised based on the method used in behaviour selection. Two major categories have been identified: (a) Arbitration methods; these are competitive based methods where the dominant behaviour affects the motor output. The selection of which behaviour is dominant is determined by the sensor readings and the internal state of the robot at a given time. b) Cooperative methods; in these methods the different behaviours contribute towards a single motor action. In BBR, subsumption architecture and potential fields’ methods have been suggested in order to deal with behaviour organisation.
This talk will explore how behaviour modularity and behaviour organisation can be developed automatically using evolutionary techniques. In the presented work, techniques to evolve behaviour modularity shall be investigated using gene expression programming (GEP). In the presented work, a biological gene analogy is used to model a new variant of GEP. This new GEP variant is then used to evolve modular behaviour and the obtained results compared to similar behaviour evolved using the standard GEP.
