Emergent behaviours are a welcome feature in a robot that is expected to work in the real world. However they are too often the result of careful engineering, thus lacking a ``true emergent'' component. Here we explore the emergence of behaviours in a chaos driven robot. A random recurrent neural network drives the activation of several behaviours. Attractors in the network are created by using an unsupervised Hebbian learning mechanism. These attractors live at the edge between chaos and low-period dynamics. We show, with three experiments conducted on a real robot, that emergent behaviours arise when online learning is adopted. An analysis of results proves that this is the case when attractor learning is adopted. This suggests a methodology to obtain surprising emergent behaviours.
|Title of host publication||Unknown Host Publication|
|Number of pages||7|
|Publication status||Published - 2010|
|Event||Proc. of Towards Autonomous Robotic Systems, TAROS 10 - |
Duration: 1 Jan 2010 → …
|Conference||Proc. of Towards Autonomous Robotic Systems, TAROS 10|
|Period||1/01/10 → …|