While the theory and practice of evolutionary robotics is well established, most work to date has been concerned with evolving a robot’s control system – it’s software. This chapter is concerned instead with the more difficult problem of how to evolve robot hardware – a robot’s physical body shape (morphology) and the arrangement of sensors and actuators within that body. We focus on the problems of which aspects of a robot’s hardware can be evolved, how they might be coded in the genome and, perhaps most importantly, how (and if) the evolved robot can be physically constructed with currently available materials and processes. We examine both non-modular and modular (multi-cellular) approaches to evolvable robot hardware, which we liken to engineering and artificial life approaches respectively. And we propose a number of possible directions for evolvable robot hardware, notably including the need for self-healing within complex evolved robots. We conclude the chapter by identifying three major challenges in evolvable robot hardware: when and how often to physically instantiate and fitness test, brain-body coevolution and genotype-phenotype mapping.
Winfield, A. F., & Timmis, J. (2015). Evolvable robot hardware. In M. A. Trefzer, & A. M. Tyrrell (Eds.), Evolvable Hardware, 331-348. Springer. https://doi.org/10.1007/978-3-662-44616-4_13