Experience with wired networks has provides guidance about what level of detail is appropriate for simulationbased protocol studies. Wireless simulations raise many new questions about approriate levels of detail in simulation models for radio propagation and energy consumption. This paper describes the trade-offs associated with adding detail to simulation models. We evaluate the effects of detail in five case studies of wireless simulations for protocol design. Ultimately the researcher must judge what level of detail is required for a given question, but we suggest two approaches to cope with varying levels of detail. When error is not correlated, networking algorithms that are robust to a range of errors are often stressed in similar ways by random error as by detailed models. We also suggest visualization techniques that can help pinpoint incorrect details and manage detail overload.

Adaptation to inversion of the visual eld is studied in a simple simulated model of phototactic behaviour. Inspired by recent ndings in neuroscience, a novel neural architecture based on continuous dynamical neural networks is implemented. Individual cells behave homeostatically by facilitating local plasticity whenever their activity goes out of bounds. Robots are evolved to perform long-term phototaxis on a series of light sources while trying to keep neurons behaving homeostatically. Robots are then tested under the condition of left/right inversion of vision. Initially, their phototactic capability is lost, which in most cases causes neurons to lose homeostasis and trigger plastic changes. After long periods of maladaptation, robots adapt to the new sensorimotor situation, and phototactic behaviour is recovered. The introduction of other disruptions such as radical perturbations to motor and sensor gains also results in eventual adaptation. The model intends t...

On behalf of:

Using evolutionary simulations we develop autonomous agents controlled by artificial neural networks (ANNs). In simple life-like tasks of foraging and navigation, high performance levels are attained by agents equipped with fully-recurrent ANN controllers. In a set of experiments sharing the same behavioural task but differing in the sensory input available to the agents, we find a common structure of a command neuron switching the dynamics of the network between radically di erent behavioural modes. When sensory position information is available the command neuron reects a map of the environment, acting as a location-dependent cell sensitive to the location and orientation of the agent. When such information is unavailable the command neuron's activity is based on a spontaneously evolving short-term memory mechanism, which underlies its apparent place-sensitive activity. A two-parameter stochastic model for this memory mechanism is proposed. We show that the parameter values emerging via the evolutionary simulations are near optimal; evolution takes advantage of seemingly harmful features of the environment to maximize the agent's foraging efficiency. The accessibility of evolved ANNs for a detailed inspection, together with the resemblance of some of the results to known findings from neurobiology places evolved ANNs as an excellent candidate model for the study of structure and function relationship in complex nervous systems.

Abstract. This paper explores the capabilities of continuous time recurrent neural networks (CTRNNs) to display reinforcement learning-like abilities on a set of T-Maze and double T-Maze navigation tasks, where the robot has to locate and “remember ” the position of a reward-zone. The “learning ” comes about without modifications of synapse strengths, but simply from internal network dynamics, as proposed by [12]. Neural controllers are evolved in simulation and in the simple case evaluated on a real robot. The evolved controllers are analyzed and the results obtained are discussed. 1

Plastic spiking neural networks are synthesized for phototactic robots using evolutionary techniques. Synaptic plasticity asymmetrically depends on the precise relative timing between presynaptic and postsynaptic spikes at the millisecond range and on longer-term activity-dependent regulatory scaling. Comparative studies have been carried out for dierent kinds of plastic neural networks with low and high level of neural noise. In all cases, the evolved controllers are highly robust against internal synaptic decay and other perturbations.

cognitive systems, biologicallyinspired computing, artificial life, artificial intelligence, autonomous agents This paper presents a review of the academic literature on biologically-inspired computing approaches to the science and engineering of cognitive systems. This review is intended as a rapid tour through the area (rather than a leisurely wander); and it should be readable in a few hours. The tour is partial in both senses of the word: it is only partially complete, and it is biased (i.e., it is not an

The problem of how to acquire a model of a physical robot, which is fit for evolution of controllers that can subsequently be used to control that robot, is considered in the context of racing a radio-controlled toy car around a randomised track. Several modelling techniques are compared, and the specific properties of the acquired models that influence the quality of the evolved controller are discussed. As we aim to minimise the amount of domain knowledge used, we further investigate the relation between the assumptions about the modelled system made by particular modelling techniques and the suitability of the acquired models as bases for controller evolution. We find that none of the models acquired is good enough on its own, and that a key to evolving robust behaviour is to evaluate controllers simultaneously on multiple models during evolution. Examples of successfully evolved racing control for the physical car are analysed.

Using evolutionary simulations, we develop autonomous agents controlled by artificial neural networks (ANNs). In simple life-like tasks of foraging and navigation, high performance levels are attained by agents equipped with fully-recurrent ANN controllers. Examining several experimental settings, differing in the sensory input available to the agents, we find a common structure of a "command neuron" switching the dynamics of the network between radically different behavioural modes. In some of the models the command neuron reects a map of the environment, acting as a "place cell". In others it is based on a spontaneously evolving short-term memory mechanism. The resemblance to known findings from neurobiology places Evolved ANNs as an excellent candidate model for the study of structure and function relation in complex nervous systems.

Traditionally sensors have been assumed to behave independently of one another. In this paper evidence is presented that shows that for certain types of sensors this assumption of independence is incorrect. In fact, in some cases groups of sensors respond in a highly correlated fashion. A new model of sensor noise is introduced which combines independent noise with dependent noise to produce sensor responses with varying degrees of correlation. This new model is then compared to the standard model in a set of evolutionary computation experiments. The results reveal that by adopting the new model transfer of simulation results to reality is improved. 1 Introduction In evaluating possible control structures for evolutionary robotics one has three options: use a physical robot, use a simulated robot, or use a hybrid of the two (Nolfi et al., 1994). Using a physical robot is obviously the most desirable choice but is often too time consuming. Using a simulated robot leads to the correspo...