Autonomous Self-Reconfiguration of Modular Robots by Evolving a Hierarchical Mechanochemical Model

In this paper, we present a two-layer hierarchical mechanochemical model for self-reconfiguration of modular robots in changing environments. The model, which is inspired by the embryonic development of multi-cellular organisms and chemical morphogenesis, can autonomously generate and form different patterns for modular robots to adapt to environmental changes. Layer 1 of the model utilizes a virtual-cell based mechanochemical model to generate appropriate target patterns (i.e., chemical blueprints) for current environment. Layer 2 is a gene regulatory network (GRN) based controller to -coordinate the modules of modular robots for physically realizing the chemical target pattern defined by the first layer. This hierarchical mechanochemical framework is a distributed system in that each module makes decisions based on its local perceptions. To optimize pattern de-sign of modular robots, the covariance matrix adaptation evolution strategy (CMA-ES) is adopted to evolve the pattern parameters of the mechanochemical model. Simulation results demonstrate that the proposed system is effective and robust in autonomously reconfiguring modular robots to adapt to environmental changes.