Embryomorphic Engineering: Growing Architectures from Self-Positioning and Self-Identifying Agents
(Institut des Systèmes Complexes, Paris Ile-de-France CREA, Ecole Polytechnique & CNRS)
Multicellular organisms are rather unique examples of natural systems that exhibit both self-organization and a strong architecture. Similarly to other bio-inspired methodologies - in which principles derived from neural networks, genetics or ant colonies are routinely used in machine learning, stochastic optimization or combinatorial search problems - can we also export the precise self-formation capabilities of biological development to a new generation of algorithmic methods and technological systems? I present here two related studies in 'Embryomorphic Engineering' (an instance of a broader field called 'Morphogenetic Engineering'). First, a 2 D/3-D multi-agent model of biological development, based on virtual gene regulation networks, chemical gradients and mechanical forces - which can be applied to the self-aggregation or self-assembly of robotic swarms into specific and reproducible superstructures. Second, a related N-D multi-agent model of self-construction of complex but precise graph topologies by 'programmed attachment', based on dynamical opening of node ports, spread of gradients and creation of links - with potential applications in socio-technical systems composed of a myriad of peer-to-peer mobile devices and human users. In all cases, the challenge is to design - e.g., through an evolutionary search - the proper set of local rules followed by each agent of a complex system on how to interact with the other agents and the environment in order to produce global functional architectures.