Learning at the edge of chaos: temporal coupling of spiking neuron controller for autonomous robotics

  • Hedi Soula
  • , Aravind Alwan
  •  and Guillaume Beslon

Citation

  • Hedi Soula
  • , Aravind Alwan
  •  and Guillaume Beslon

, "Learning at the edge of chaos: temporal coupling of spiking neuron controller for autonomous robotics", AAAI Spring Symposium on Developmental Robotics, 2005

Abstract

In this paper, a recurrent spiking neural networks is trained on an robot to learn to avoid obstacles using visual flow. At the starting of the process, this network is initialized in a “chaotic” state and a STDP-like learning algorithm is used. We argue that a proper scaling variable can direct the network from chaos to synchronized state and back. This process allows us to train the robot because it links (external) temporal loops with (internal) neural activity. We use the scaling factor to have this coupling functional. Given an over-simplistic scaling, we managed to obtain very interesting resulting behaviors when tested on a real robot.

Link to the paper: soula2005learning.pdf