Robo-Salamander – an approach for the benefit of both robotics and biology

Even if the character of robotics is primarily technological, it was always closely connected with biology right from the beginning. However, most of the time this was only a one-way relationship, for biological insights were often used as a pool of approved ideas and methods to find solutions for rudimentary problems in robotics (walking machines in particular) with little in return. In contrast to this general habit, our project “Robo-Salamander” started with the attempt to gather evidence for the answer to an old query in biology from a technical point of view – the first step in the evolution of vertebrate walking. We found a simple principle which indeed helps to understand the locomotion of the first tetrapodes (that can still be observed today by watching salamanders and certain lizards), and its continuous development into more existing walking schemes. By transferring this quite simple idea to a first prototype robot, we could also find its interesting technological advantages referring to movability, efficiency, flexibility, robustness and redundancy of walking machines and their capability to cope with rough and demanding environments. These characteristics were very useful for testing our latest learning algorithms based on a combination of reinforcement learning and neural networks with very promising results. Furthermore they turned out to be the ideal tool for our research on evolving neural networks as controllers for complex walking machines by using incremental evolution. For the Robo-Salamander combines at least two independent, different complex methods of locomotion, it allows to evolve controlling networks stepwise beginning by few actuators and sensors with simple movements and rising up to the final complex robot with many degrees of freedom. Although the idea of using incremental evolution for handling high complexities is not new, it is still not completely known how previously gained functionalities of a network can be preserved while adding new actuators, sensors or even neurons in the next steps. This paper firstly discusses the properties of the Robo-Salamander that make it the ideal platform for our research on solving these problems and secondly shows the first results.