Biological Motivation

At the origin of life, simple reproductive molecular “machines” emerged, which were shaped from the very beginning by ecological relationships (competition, co-operation and mutual consumption). In the process called “biological evolution”, billions of different life forms developed since then, the successful ones still exist today.

In biology, the topic of Evolution has always be a crucial one. Although the concepts of Darwin’s explanation of the origin of biological species are widely accepted, his thesis is nevertheless a theory and not a “proven law”. Especially nowadays, evolutionary theory is frequently challenged by creationists, who argue that very little “hard” evidence for biological evolution is brought forward by biological research. Serious biological research approaches this problem from several directions: Palaeontologists search for fossil records of speciation and of adaptation, ecologists investigate the possible selection criteria that can drive evolution, and theoretical biologists analyze the general properties of the evolutionary processes in mathematical terms. Within the last two decades, the field of Artificial Life followed the approach to evolve life-like structures. Most of these works were performed in computer simulation, which sometimes allowed whole ecosystems to evolve. These studies showed that complex behavioural patterns and stable ecological relationships could emerge purely by evolutionary forces. With the progress in robotics, Artificial Life got more and more interest in real embodied, physically existent animats that can prove the feasibility to develop creatures by the means of pure Darwinian Evolution.

The research presented in this project is aimed to extend the current state of knowledge significantly further and (at the end) provide significant new insights into the emergent phenomena of evolution in physically embodied agents. It addresses not only the question whether or not sophisticated high-level behaviour can emerge by artificial evolution, it specifically investigates this question in a field of evolutionary research that still has many unresolved questions. These questions can be formulated (in short) as follows. Some of these questions can (will) be investigated by analyzing the evolutionary pathway of features within one single robot; others will be investigated by observing groups of (aggregated) robots:

Individual level
Collective level
1) Evolution of robust concepts: Can “robustness” be an intrinsic shaping factor in evolution without being explicitly coded into a sort of fitness function? 1) Evolution of Cooperation: How can egoists, who serve as the primary unit of selection, decide to cooperate?
2) Evolution of flexible concepts: Can “flexibility” be an intrinsic shaping factor in evolution without being explicitly coded into a sort of fitness function? 2) Evolution of Communication: How can a sophisticated way of signalling (or even a “language”) evolve as an emergent phenomenon from Artificial Evolution?
3) Evolution of simple concepts: Can “simplicity” be an intrinsic shaping factor in evolution without being explicitly coded into a sort of fitness function? 3) Emergence of Multi-cellular Organisms: How can a higher-level (multi-component) organism autonomously assemble via Artificial Evolution?
4) Non-target-focussed behaviour: Can unintentional (on the first sight) behaviour emerge via Artificial Evolution? E.g., “playing behaviour” to serve as a training phase for the organism during the developmental period.
5) Evolution of sensor characteristics: What solutions can be found by Artificial Evolution to provide sensor data exactly in an optimal format for the controller that relies on these data (Evolution of secondary sensory units).
6) Evolution of spin-off functionality: Can evolution produce functionality that was not explicitely designed for that purpose? E.g. evolve a method to communicate by the noise produced by the motors. Or achieve collision avoidance by actively producing sound with the gear box.

These topics are of high relevance for evolutionary biologists and they are very hard to be investigated with real animals. One reason for that is that neuronal systems of animals are not fully understood yet, another one is that evolution is a slow-acting process that changes the morphology and the physiology of organisms over the run of hundreds or thousands of generations.

The approach taken in this project will be able to provide insights into these questions by using “artificial animals” (animats). These animats can be shaped by evolution within much shorter time spans and their relevant structures (morphology, physiology and controllers) are much easier accessible and interpretable.

Although these animats are not “real” animals, they are similarly constrained by the rules of physics and chemistry (as are real organisms), the used approach to control their behaviour will be very nature-near and by implementing a simulated physiology, they will closely resemble real animals. Like real animals, the features of the habitats they “live” in will affect them and simulated sexuality (e.g., partner choice) will also be incorporated into those algorithms that drive the Artificial Evolution. Thus, the observed dynamics of the evolutionary process and the achieved end products of this process will be highly comparable to systems found in real animals. This way, our suggested population of SYMBRION robots will also serve the Artificial Life community as a test platform for their future concepts and ideas.

Biological Background. In nature organisms often have controllers with a high degree of redundancy concerning their morphological structure. In higher organisms these controllers are mostly neuronal systems, which (from a certain level of organisation on) show complex structures called “brains” (Fig. 6). In more simple organisms, for example in crayfish or insects, we find ventral nerve cords that are structured similar to a rope ladder. These nervous structures extend through the whole body of the animal on the ventral side, having one or two ganglia per segment. The development and structure of these nervous systems can be defined or explained very good by certain mathematical systems, for example by L-systems. The processes involved in developing L-Systems are highly suitable to create symmetric, repetitive (segmented, cellular) and redundant structures. We plan to exploit these abilities to generate such controllers in our approach to generate symmetric and modular higher-order organisms out of our small SYMBRION modules (robots). We find these high structured morphology not only in rope-ladder-like ventral nerve cords, but also in simple brains and in their evolutionary antecessors, the upper and lower pharyngeal ganglia. To adapt such systems to the micro robotic world, we have to find a system that allows us to automatically evolve these bio-inspired controllers. The blueprint must be stored in some gene-like code, evaluated, and evolved from generation to generation. The second important unit we have to implement is the developmental engine. This unit translates the genetic code into the final controller structure that is finally steering the robot. In the robot, this controller is encoded by normal processor or microcontroller commands.

The translation of the genetic code into a usable controller can be understood as a decoding process. That means, that syntax and the grammar of the genetic code and the functions of the developmental engine are not independent from each other. From a biological point of view the reading process of the developmental engine is comparable with the transcription process, the protein synthesis and the morphologic development. The part of the Developmental Engine that creates (encodes) the final controller structure can be compared to the biological process of Embryogenesis.

Scientific questions

  • How will the robots adapt to their environments?
  • Which methods will the robots develop to interact?
  • Will certain protocols of communication develop?
  • Will the robots develop into self-aggregating super-organisms, depending on the environment ?
  • Will the robots compete on the level of super-organisms?
  • Will the robots increase their energetic gain over time?
  • Will environmental lead to novel developments in behaviour and ecology?
  • How will sexual selection affect the system?

Created by admin. Last Modification: Friday 30 of November, 2007 13:30:31 CET by admin.