Using a morphogenetic system (see Roggen et al. "Evolutionary morphogenesis for multi-cellular systems" in Genetic Programming and Evolvable Machines, 2007), we evolve a multi-cellular circuit that controls the navigation of a robot in an environment with moving visual cues.
The task of the robot is to learn one of the moving visual cues, and afterwards home toward the cue that was learned.
Each cell is a spiking neural network with spike timing dependent plasticity (STDP) learning rules.
Therefore learning at the synaptic level of the
Results show that the multi-cellular controller can be successfully evolved with the morphogenetic system, and behavioral tests show that the robot is able to home towards the learned cue from most of the locations in the arena.
This demonstration combines evolution, development, and learning, thus combining all perspectives envisioned in the "POE" model of bio-inspiration.
Loading more stuff…
Hmm…it looks like things are taking a while to load. Try again?