One of the main limitations for the functional scalability of automated design systems is the representation used for encoding designs. We argue that generative representations, those which are capable of reusing elements of the encoded design in the translation to the actual artifact, are better suited for automated design because reuse of building blocks captures some design dependencies and improves the ability to make large changes in design space. To support this argument we compare a generative and non-generative representation on a table design problem and find that designs evolved with the generative representation have higher fitness and a more regular structure. Additionally the generative representation was found to better capture the height dependency between table legs and also produced a wider range of table designs.

Evolutionary designs based upon Artificial Ontogenies are beginning to cross from virtual to real environments. In such systems the evolved genotype is an indirect, procedural representation of the final structure. To date, most Artificial Ontogenies have relied upon an error-free development process to generate their phenotypic structure.

Artificial Ontogenies, which are inspired by biological development, have been used to automatically generate a wide array of novel objects, some of which have recently been manufactured in the real world. The majority of these evolved designs have been evaluated in simulation as completed objects, with no attention paid to how, or even if, they can be realistically built. As a consequence, significant human effort is required to transfer the designs to the real world. One way to reduce human involvement in this regard is to evolve how to build rather than what to build, by using prescriptive rather than descriptive representations. In the context of Artificial Ontogenies, this requires what we call Situated Development, in which an object’s development occurs in the same environment as its final evaluation. Not only does this produce sufficient information on how to build evolved designs, but it also ensures that only buildable designs are evolved. In this paper we explore the consequences of Situated Development, and demonstrate how it can be incorporated into Artificial Ontogenies in order to generate buildable objects, which can be sequentially assembled in a realistic 3-D physics environment.

Contemporary artificial life (also known as “ALife”) is an interdisciplinary study of life and life-like processes. Its two most important qualities are that it focuses on the essential rather than the contingent features of living systems and that it attempts to understand living systems by artificially synthesizing extremely simple forms of them. These two qualities are connected. By synthesizing simple systems that are very life-like and yet very unfamiliar, artificial life constructively explores the boundaries of what is possible for life. At the moment, artificial life uses three different kinds of synthetic methods. “Soft ” artificial life creates computer simulations or other purely digital constructions that exhibit life-like behavior. “Hard” artificial life produces hardware implementations of life-like systems. “Wet ” artificial life involves the creation of life-like systems in a laboratory using biochemical materials. Contemporary artificial life is vigorous and diverse. So this chapter’s first goal is to convey what artificial life is like. It first briefly reviews the history of artificial life and illustrates the current research thrusts in contemporary “soft”, “hard”, and

A current issue in evolutionary robotics involves the coevolution of robot controllers and body morphologies built from modular parts. As part of ongoing research, a model for the evolution of the morphologies and neural network controllers of robots is described. Several robots are evolved for locomotion in simulation built from modules representing cheap, preexisting parts and one is physically built that has comparable behaviour with its original simulated version. The behaviour in simulation of such example robots is described. A brief comparison is made between the behaviour of a simulated robot whose design and behaviour has been evolved and its physically instantiated counterpart.

Abstract. This article presents experiments that integrate competitive coevolution of neural robot controllers with ‘co-evolution ’ of robot morphologies and control systems. More specifically, the experiments investigate the influence of constraints on the evolved behavior of predator-prey robots, especially how task-dependent morphologies emerge as a result of competitive co-evolution. This is achieved by allowing the evolutionary process to evolve, in addition to the neural controllers, the view angle and range of the robot’s camera, and introducing dependencies between different parameters.

Many researchers in embodied cognitive science and AI, and evolutionary robotics in particular, emphasize the interaction of brain, body and environment as crucial to the emergence of intelligent, adaptive behavior. Accordingly, the interaction between agent and environment, as well as the co-adaptation of artificial brains and bodies has been the focus of much research in evolutionary robotics. Hence, there are plenty of studies of robotic agents/species adapting to a given environment. Many animals, on the other hand, in particular humans, to some extent can choose to adapt the environment to their own needs instead of adapting (only) themselves. That alternative has been studied relatively little in robot experiments. This paper therefore presents some simple initial simulation experiments, in a delayed response task setting, that illustrate how the evolution of environment adaptation can serve to provide cognitive scaffolding that reduces the requirements for individual agents. Furthermore, theoretical implications, open questions and future research directions for evolutionary robotics are discussed.

The majority of work in the field of evolutionary robotics concerns itself with evolving control strategies for human designed or bio-mimicked robot morphologies. However, there are reasons why co-evolving morphology along with control may provide a better path towards realizing intelligent agents. Towards this goal, a novel method for evolving three-dimensional physical structures using CPPN-NEAT is introduced which is capable of producing artifacts that capture the non-obvious yet close relationship between function and physical structure. Moreover, it is shown how more fit solutions can be achieved with less computational effort by using growth and environmental CPPN input parameters as well as incremental changes in resolution.

Manufacturing processes are a key source of faults in complex hardware systems. Minimizing this impact of manufacturing uncertainties is one way towards achieving fault tolerant systems. By treating manufacturing as a stochastic development process, we characterize some of the constraints limiting the levels of robustness that can be achieved with evolution. The analysis is by introducing a novel abstraction of development as a strategic decisionmaking process. Using this abstraction to analyze a toysystem that simulates a process of noisy assembly, we compare the maximum robustness achievable with adaptive and non-adaptive developmental strategies. Even in this highly simplified setup, the optimal adaptive and non-adaptive genotypes reveals a significant empirical difference in their robustness characteristics. This suggests that the choice of developmental strategy and the properties of the setup are major constraints on the robustness achievable, even prior to evolution-related considerations.

Evolutionary Design has demonstrated great potential to automatically generate a wide array of novel, interesting, and human-competitive designs. Few of these evolved designs, however, have in turn been physically manufacture. This is due largely to the fact that most evolved designs only specify what to build, and carry no information on how, or even if, a designed object can be assembled in the real world. When the goal is a physical object, rather than a mere schematic, substantial further effort, most often human-level, is subsequently required to develop a physical assembly process. Evolution of such descriptive representations therefore stands as an obstacle to the full automation of both design and assembly. In this paper we describe an alternative, the evolution of prescriptive representations, which offers to remove human effort from the design-andassembly loop.