Prospects for Machine Embryogenesis

In Nature, the embryogenesis process proceeds from a single fertilized cell through division, migration, specialization and apoptosis. Although a lot is known about development, we still have a long way to go from theories of pattern formation towards understanding the intelligence within an unsupervised manufacturing process which robustly assembles complex biological forms. Our approach has been to co-evolve bodies and brains in simulation and then convert them into reality using commercial manufacturing technology. I will review several generations of robots which were automatically designed using co-evolutionary techniques. The goal has been the fully automated design and construction of artificial lifeforms. The first generation was based on genetic programming and a simulation of LEGO rod adhesion. The second generation used direct evolution on a iterative simulation of truss structures and used 3D printing for the output. A third generation was based on generative representations using L-systems. In each of these cases, we assumed a perfect factory which could accept an evolved specification and then manufacture the desired result. In reality, there is no perfect factory, except for the science fiction Star Trek replicator. All manufacturing and assembly systems are subject to error. Each primitive manufacturing action results not in a deterministic new state, but a probability distribution of outcomes. In later work, we replaced the idea of a perfect factory with one subject to noise and error. Even the smallest bit of error ruins the outcome of deterministic construction plans. We first evolved construction plans which could overcome errors through redundancy, and then this led to a new model for machine embryogenesis as a process which continuously optimizes assembly processes in a game against Nature.