Peter Eggenberger

Evolving the morphology of a compound eye on a robot

Reports on an experiment in evolving the morphology of an artificial compound eye with 16 light sensors on a robot. A special robot was designed and constructed that is able to autonomously modify the angular positions of the individual light sensors within the compound eye. The task of the robot was to employ motion parallax to estimate a critical distance to obstacles. This task was achieved by adapting the morphology of the compound eye by an evolutionary algorithm while using a fixed neural network to control the robot.

Adaptation on the Evolutionary Time Scale: A Working Hypothesis and Basic Experiments

In the pertinent literature, an ongoing discussion can be found about whether evolutionary algorithms are better suited for optimization or adaptation. Unfortunately, the pertinent literature does not offer a definition of the difference between adaptation and optimization. As a working hypothesis, this paper proposes adaptation as tracking the moving optimum of a dynamically changing fitness function as opposed to optimization as finding the optimum of a static fitness function. The results presented in this paper suggest that providing enough variation among the population members and applying a selection scheme is sufficient for adaptation. The resulting performance, however, depends on the problem, the selection scheme, the variation operators, as well as possibly other factors.

Toward Seamless Transfer from Simulated to Real Worlds: A Dynamically-Rearranging Neural Network Approach

In the field of evolutionary robotics artificial neural networks are often used to construct controllers for autonomous agents, because they have useful properties such as the ability to generalize or to be noise-tolerant. Since the process to evolve such controllers in the real-world is very time-consuming, one usually uses simulators to speed up the evolutionary process. By doing so a new problem arises: The controllers evolved in the simulator show not the same fitness as those in the real-world. A gap between the simulated and real environments exists. In order to alleviate this problem we introduce the concept of neuromodulators, which allows to evolve neural networks which can adjust not only the synaptic weights, but also the structure of the neural network by blocking and/or activating synapses or neurons. We apply this concept to a peg-pushing problem for KheperaTM and compare our method to a conventional one, which evolves directly the synaptic weights. Simulation and real experimental results show that the proposed approach is highly promising.

Toward seamless transplantation from simulated to real worlds: a dynamically rearranging neural network approach

Toward seamless transplantation from simulated to real worlds: a dynamically rearranging neural network approach Akio Ishiguro a , Seiji Tokura b , Yoshiki Uchikawa c & Peter Eggenberger d a Department of Computational Science and Engineering, Graduate School of Engineering, Nagoya University, Furocho, Chikusa-ku, Nagoya 464-8603, Japan b Department of Computational Science and Engineering, Graduate School of Engineering, Nagoya University, Furocho, Chikusa-ku, Nagoya 464-8603, Japan c Department of Computational Science and Engineering, Graduate School of Engineering, Nagoya University, Furocho, Chikusa-ku, Nagoya 464-8603, Japan d Evolutionary Systems Department, ATR Human Information Processing Research Laboratories, Seika-cho, Soraku-gun, Kyoto 619-0288, Japan Published online: 02 Apr 2012.

Generation of an adaptive controller CPG for a quadruped robot with neuromodulation mechanism

In order to create agile locomotion for legged robots, so far various methods have been proposed with the concept of neural circuits, or so-called Central Pattern Generators (CPG). In contrast to these approaches in which monolithic CPG circuits are employed to control locomotion, in this study a polymorphic neural circuit is employed instead, allowing the dynamic change of its properties according to the current situation in real time. To this end, the concept of neuromodulation has been introduced. To verify the feasibility of this approach, this concept has been applied to the control of a 3-D quadruped robot. Simulations have been carried out and the results have showed that the proposed method appropriately changes the walking pattern according to the current situation in real time.

Reduction of the gap between simulated and real environments in evolutionary robotics: a dynamically-rearranging neural network approach

The evolutionary robotics approach has been attracting a lot of concern in robotics and artificial life communities. In this approach, neural networks are widely used to construct controllers for autonomous mobile agents, since they intrinsically have nonlinear mapping, generalization, noise-tolerant abilities and so on. However, the following are still open questions: 1) the gap between simulated and real environments, 2) the evolutionary and learning phase are completely separated, and 3) the conflict between stability and evolvability/adaptability. We particularly focus on the gap problem, and try to alleviate this by incorporating the concept of dynamic rearrangement function of biological neural networks with the use of neuromodulators. Simulation and real experimental results show that the proposed approach is highly promising.

Evolving an adaptive controller for a quadruped-robot with dynamically-rearranging neural networks

As highly complicated interaction dynamics exist, it is therefore extremely difficult to design controllers for legged robots. Evolutionary robotics is one of the most promising approaches, but there still exist several problems that have to be solved. One of the critical problems is that evolved agents generally tend to over adapt to their given environments through the evolutionary process. In other words, they lack rich adaptability. Therefore, it is highly necessary to establish a method that enables one to efficiently construct adaptive controllers that can cope with different situations. For this purpose we introduce the concept of neuromodulators, allowing the evolvement of neural networks which can adjust not only the synaptic weights, but also the structure of the neural network by blocking and/or activating synapses or neurons. We apply this concept to create an adaptive controller for a quadruped robot which not only can walk forward but also regulate the torque output applied to each joint, according to the current situation.

Evolving Bipedal Locomotion with a Dynamically-Rearranging Neural Network

Since highly complicated interaction dynamics exist, it is in general extremely difficult to design controllers for legged robots. So far various methods have been proposed with the concept of neural circuits, so-called Central Pattern Generators(CPG). In contrast to these approaches in this article we use a polymorphic neural circuit instead, allowing the dynamic change of its properties according to the current situation in real time. To do so, we introduce the concept of neuromodulation with a diffusion-reaction mechanism of neuromodulators. Since there is currently no theory about how such dynamic neural networks can be created, the evolutionary approach is the method of choice to explore the interaction among the neuromodulators, receptors, synapses and neurons. We apply this neural network to the control of a 3-D biped robot which is intrinsically unstable. In this article, we will show our simulation results and provide some interesting points derived from the obtained results.

The Effect of Neuromodulations on the Adaptability of Evolved Neurocontrollers

One of the serious drawbacks in Evolutionary Robotics approaches is that evolved agents in simulated environments often show significantly different behavior in real environments due to unforeseen perturbations. This is sometimes referred to as the gap problem. In order to alleviate this problem, we have so far proposed Dynamically-Rearranging Neural Networks(DRNN) by introducing the concept of neuromodulations with a diffusion-reaction mechanism of signaling molecules to so-called neuromodulators. In this study, an analysis of the evolved DRNN and a quantitative comparison with standard neural networks are presented. Through this analysis, we discuss the effect of neuromodulation on the adaptability of the evolved neurocontrollers.

Evolving Morphologies and Neural Controllers Based on the Same Underlying Principle: Specific Ligand-Receptor Interactions

Our research is motivated by the fact that in Nature macroscopic forms are often shaped with high precision (the shape of the cornea, the bones in an articulation etc.), this paper investigates the possible use of biological mechanisms to evolve shapes and functions for a given task. The concurrent evolution of shape and neural controller of an agent creates a new kind of problem: As the neural nets are not independent of the body structure (often the sensors are distributed over the body, the effectors have their positions), a change in body shape will often decrease the overall fitness. This will flaw any artificial evolutionary system, unless the evolutionary process is robust against changes of the positions of the cells. Therefore, this paper proposes an evolutionary system able to explore shape and neural networks of an agent independently based on specific receptor-ligand interactions.