Controlling legged robots with coupled artificial biochemical networks

Abstract

Artificial biochemical networks (ABNs) are computational architectures motivated by the organisation of cells and tissues at a biochemical level. In previous work, we have shown how artificial biochemical networks can be used to control trajectories in discrete and continuous dynamical systems. In this work, we extend the approach to the control of a hybrid dynamical system: a legged robot. Taking inspiration from biological cells, in which complex behaviours come about through the interaction of different classes of biochemical network, we develop the notion of a coupled artificial biochemical network, in which an artificial genetic network controls the configuration of an artificial metabolic network. Using a higher-level robotic control task, we show how the coupled network finds solutions which can not be readily expressed using the artificial genetic network or artificial metabolic network alone. Our results also show the important role that non-linear maps can play as a natural source of complex dynamics.