Promotors: Prof. Dr. G. Eiben, Prof. Dr. J. Ellers
Co-Promotors: Prof. Dr. G. Meynen, Dr. E. Ferrante
Imagine a future where robots are not designed by humans, but are reproducing and evolving, and have become an integral part of the ecosystem. It may be on Earth or another planet. In this future robots are not built, they are “born”. Being part of the ecosystem, they serve vital functions for otherwise delicate environments, like maintaining a balance between multiple species of plants and animals. Imagine robots deployed at a global scale to monitor and maintain the temperature and CO2 levels of the atmosphere. Small and large ecosystems that have robots integrated in their delicate balance sound like science fiction today, but recent progress in the field of Evolutionary Robotics are pushing us towards this reality. Evolutionary Robotics is a field of Artificial Intelligence that borrows tools and algorithms from Evolutionary Computing and Genetic Algorithms and applies them to robot design. Designing and producing a robot has always been a human task; this has some advantages, like absolute control over the result, and disadvantages, i.e. really high development costs and low scalability. In the field of Evolutionary Robotics we imagine robots that are not be designed by a team 1 Chapter 1. Introduction of engineers but by an evolutionary algorithm. This algorithm will evolve robots within an environment, optionally with a task for the robots to solve, and replicate those robots that have proved to be more adapted to the environment. This strategy has three major advantages: one is that humans don’t have to tweak individual details of the robots, but only to design a good environment. Secondly, the adaptation process does not have to stop; within a dynamic environment, robots can keep adapting to new changes and become integral part of the ecosystem of such environment. Thirdly, it is possible that the evolutionary process comes up with some unconventional design modifications that would not have been considered by a group of engineers. While Evolutionary Robotics is certainly a novel way to design robots, it can also be used as a model to study evolutionary processes. Ideally a running evolutionary robotics system can recreate complex conditions and surpass simple agent based evolutionary models. On the opposite side, it can be a simpler model than real biological evolution, and it can be tested with several encoding types, while we would otherwise be limited to only DNA-based organisms. While this all may sound interesting and promising, there are still many challenges within the field. The number of resources required to start such a system is one of them; the time required to develop useful solutions is a second one. These obstacles pushed researchers to evolve robots within simulations, where resources are cheap and time can be sped up considerably. In turn, the heavy use of simulators skew research towards evolving simple robots in simple isolated environments with limited interactions.