Introduction
This chapter describes our work in evolution of buildable designs using miniature plastic bricks as modular components. Lego
1
bricks are well known for their flexibility when it comes to creating low cost, handy designs of vehicles and structures. Their simple modular concept make toy bricks a good ground for doing evolution of computer simulated structures which can be built and deployed. Instead of incorporating an expert system of engineering knowledge into the program, which would result in more familiar structures, we combined an evolutionary algorithm with a model of the physical reality and a purely utilitarian fitness function, providing measures of feasibility and functionality. Our algorithms integrate a model of the physical properties of Lego structures with an evolutionary process that freely combines bricks of different shape and size into structures that are evaluated by how well they perform a desired function. The evolutionary process runs in an environment that has not been unnecessarily constrained by our own preconceptions on how to solve the problem.
The results are encouraging. The evolved structures have a surprisingly alien look: they are not based in common knowledge on how to build with brick toys; instead, the computer found ways of its own through the evolutionary search process. We were able to assemble the final designs manually and confirm that they accomplish the objectives introduced with our fitness functions.
This chapter discusses background and related work first (section 2), then goes on to describe our methods; first the model we use to simulate Lego structures (sections 3-4), then the representation and evolutionary algorithms (section 5). The results sections (6-7) discuss applications, showing the results of several evolutionary runs and illustrating with pictures of the final assembled
Lego artifacts. Finally, on sections 8-9, current and future lines of work and conclusions are drawn.
2 Background
In order to evolve both the morphology and behavior of autonomous devices which can be manufactured, one must have adequate representations and simulations. The representation must provide the computer with ways to create, manipulate and modify an infinite variety of virtual architectures to be tested in simulation. The objects being simulated need to be adaptive enough to cover the gap between simulated and real world, so they will perform correctly when built. Desirable features of a software engine for evolving morphology are:
• Universal - the simulator should cover an infinite general space of mechanisms.
• Conservative - because simulation is never perfect, it should preserve a margin of safety.
• Efficient - it should be quicker to test in simulation than through physical production and test. • Buildable - results should be convertible from a simulation to a real object