Research

My initial experimentations were focused on bio-inspired evolution of virtuals plants of various scales: from mitosis-based developmental models of a single entity to complex colonies in dynamical environments. Following the obtention of my PhD, I broadened the scope of my research to include Artificial Neural Networks evolved through a scalable method with indirect encoding. These were, first, applied to 2D robots with morphological components and, later on, to modular 3D robots.

NeuroEvolution

NeuroEvolution is a research field regrouping all techniques by which an Artificial Neural Network (ANN) can be made “better” via an Evolutionary approach with either direct or indirect encodings. Better, refers to the comparison, through a (multi-objective) fitness function, of performance between individuals, either on a population or species level.
Read more

Evolutionary Robotics

This is an item about both evolutionary robotics platforms I have worked with.
Read more

AMaze

AMaze is a benchmark maker library meant to provide a fast prototyping platform. In the midst of my combining NeuroEvolution with Reinforcement Learning on an Evolutionary Robotics platform, the cost of using the latter became obviously hindering. From this stemmed the core concepts of the library: lightweight computations, controllable environmental complexity and easy human interaction.
Read more

Open-Ended Evolution

This is an item fronting my OEE research
Read more