Fine grained artificial development for body-controller coevolution of soft-bodied animats

We apply the concept of multicellular development to evolve bodies and controllers of soft-bodied animats, evaluated within simulated physics environment. Growth of an embryo is driven by a neural network-like controller in every cell. We show how, through an evolutionary process, this simple approach produces rich and complex morphologies of ani-mats made of hundreds of cells and exhibiting life-like gaits without centralized control. We discuss design decisions and issues that needed addressing to make evolution of moving bodies possible and investigate how some of these decisions impact quality of obtained solutions. We then show how the system allows us to evolve animats that optimize both for dis-tance and the use of actuating material. Finally, we demon-strate how it can be used to evolve developmentally plastic in-dividuals, whose genomes encode more than one phenotype, each adapted to a different type of environment