A deep chasm separates Artificial Life work that uses robotics models [37,36,95,22] from the one in virtual worlds [123,124,114,101,81]. Robots today lack plasticity in their design; they need to be built by hand, molded using expensive methods. The number of generations and the number of configurations tested is several orders of magnitude smaller than those that can be reached with simulated environments. Evolutionary Robotics does not address morphology, although the idea was around from the beginning . Experiments generally focus on evolving behaviors within a fixed morphology -- a robotic ``platform''. Occasionally we see shape variables, but limited to a few parameters, such as wheel diameter or sensor orientation [26,88,96].
Evolution in Virtual Worlds on the other hand, is often morphological[123,81]. Virtual worlds are constrained neither by the fixed ``speed'' of real time, nor by physical laws such as conservation of mass. The drawback is in the level of detail and the complexity of reality: simulating everything would require an infinite amount of computation.
The Evolution of Buildable Structures project aims at bridging the reality gap between virtual worlds and robotics by evolving agents in simulation under adequate constraints, and then transferring the results, constructing the ``reality equivalent'' of the agent.
Lego2.1 bricks are popular construction blocks, commonly used for educational, recreation and research purposes. We chose these commercially available bricks because they have proven to be adequate for so many uses, suggesting that they have an appropriate combination of size, tightness, resistance, modularity and price. These characteristics led us to expect to be able to evolve interesting, complex structures that can be built, used and recycled.