Recently, a new approach that involves a form of simulated evolution has been proposed for the building of autonomous robots. However, it is still not clear if this approach may be adequate to face real life problems. In this paper we show how control systems that perform a non-trivial sequence of behaviors can be obtained with this methodology by carefully designing the conditions in which the evolutionary process operates. In the experiment described in the paper, a mobile robot is trained to locate, recognize, and grasp a target object. The controller of the robot has been evolved in simulation and then downloaded and tested on the real robot.

An evolutionary algorithm is used to evolve gaits with the Sony entertainment robot, AIBO. All processing is handled by the robot's on-board computer with individuals evaluated using the robot's hardware. By sculpting the experimental environment, we increase robustness to different surface types and different AIBOs. Evolved gaits are faster than those created by hand. Using this technique we evolve a gait used in the consumer version of AIBO.

A trend in robotics is towards legged robots. One of the issues with legged robots is the development of gaits. Typically gaits are developed manually. In this paper we report our results of autonomous evolution of dynamic gaits for the Sony Quadruped Robot. Fitness is determined using the robot's digital camera and infrared sensors. Using this system we evolve faster dynamic gaits than previously manually developed. 1

It is well known that individual learning can speed up artificial evolution enormously. However both supervised learning and reinforcement learning require specific learning goals which usually are not available or difficult to find. We introduce a new principle -- homeokinesis -- which is completely unspecific and yet induces specific, seemingly goal--oriented behaviors of an agent in a complex external world. The principle is based on the assumption that the agent is equipped with an adaptive model of its behavior. A learning signal for both the model and the controller is derived from the misfit between the real behavior of the agent in the world and that predicted by the model. If the structural complexity of the model is chosen adequately, this misfit is minimized if the agent exhibits a smooth controlled behavior. The principle is explicated by two examples. We moreover discuss how functional modularization emerges in a natural way in a structured system from a mechanism of competition for the best internal representation.

This thesis presents a study of the provision of emotions for artificial agents with the ultimate aim of enhancing their autonomy, i.e. making them more flexible, robust and self-sufficient. In recent years, the importance of emotions and their assistance to cognition has been increasingly acknowledged. Emotions are no longer considered undesirable or simply useless. Their role in various aspects of human and animal cognition like perception, attention, memory, decision-making and social interaction has been recognised as essential. The importance of emotions is much more evident insocial interaction and therefore much of the emotions research done in artificial systems focuses on the expression and recognition of emotions. However, recent neurophysiological research suggests that emotions also play a crucial part in cognition itself. This thesis investigates ways in which artificial emotions can improve autonomous behaviour in the domain of a simple, but complete, solitary learning agent. For this purpose, a non-symbolic emotion model was designed and implemented. It takes the form of a recurrent artificial neural network where emotions influence the perception

. Complex robots inspired by biological systems usually consist of many dependent actuators and are difficult to control. If no model is available automatic learning and adaptation methods have to be applied. The aim of this contribution is twofold: (1) To present an easy to maintain and cheap test platform, which fulfils the requirements of a complex control problem. (2) To discuss the application of Genetic Programming for evolution of control programs in real time. An extensive number of experiments with two real robots has been carried out. Keywords genetic programming, real-time robotics, random morphology robot, hardware evolution 1 Complex Bio-Inspired Robots Conventional industrial robots are designed in such a way that a model can be derived easily and the inverse kinematic can be calculated. In operation, the inverse kinematics is used to compute the trajectory for movement between given points in the working area of the robot. Connections between actuators are made as stic...

One of the biggest challenges in obtaining truly autonomic, selfmanaged networks is to automate the process of software evolution, and in particular, the evolution of protocol implementations and configurations. Such networks ultimately require self-modifying, evolving protocol software. Otherwise humans must intervene in every situation that has not been anticipated at design time. For this to become feasible autonomic systems must ensure non-disruptive, resilient on-line software evolution. We are starting to explore approaches to network evolution that operate directly at the code level. We investigate related code steering techniques in two directions: One is the fully automatic selection of protocol service elements where, depending on device characteristics and current operation environment, each communication entity has to select among a potentially wide variety of protocol implementations providing similar services. The other direction relates to the automatic synthesis of new protocol elements which are the result of optimizing existing implementations for a specific context. In both cases we look at genetic programming as a tool to generate new code and software configurations automatically. We propose a framework for such a resilient protocol evolution and report on first exploratory results on the adaptation and re-adaptation to environmental conditions, and the elimination of superfluous code.

We present a complete architecture for behavioral control of locomotion for both real and simulated agents and provide a design methodology for building the locomotion control systems that embody the architecture. A low-level locomotion engine controls an agent's actions directly based on intermediate-level reactive behaviors such as attraction and avoidance. High-level state machines schedule and control the reactive behaviors allowing for more "intelligent" decision processes, and an agent model provides a mechanism for varying locomotion according to agent state and personality attributes. In addition to providing specifications for a locomotion engine, we address the problem of selecting and organizing an appropriate set of behaviors. We present selection criteria and a method for partitioning the behaviors to aid in implementation. We discuss the challenges specific to human locomotion and explain how to...

This article introduces the developmental approach to artificial intelligence, which is different from other existing major approaches: knowledge-based, behavior-based, learning-based, and evolutionary approaches. The developmental approach is motivated by human cognitive development from infancy to adulthood, during which human individuals develop their intelligence through interactions with the environment. A developmental algorithm of a species, either natural or artificial, starts to run at the "birth" of the individual and it runs continuously through the entire life span. It automates the process of system development. The developmental approach does not mean just from small to big and from simple to complex. It requires the system to learn new tasks and new aspects of each complex task without a need of reprogramming. We introduce AA-learning as a basic mode for developmental learning. This paper introduces the basic concepts, the architecture, some developmental algorithms, and...

this paper? : : : : : : : : : : : : : : : : : : : : : : : : vi LIST OF TABLES Table 2.1: One point crossover. : : : : : : : : : : : : : : : : : : : : : : : 14 Table 3.1: Average of the two large clusters, no walls. : : : : : : : : : : 22 Table 3.2: Average of the five clusters, reflecting walls. : : : : : : : : : : 28 Table 3.3: Average of the four clusters, lethal walls. : : : : : : : : : : : 31 Table 3.4: Cluster averages with no walls, multiple sources. : : : : : : : 38 Table 3.5: Average of the two large clusters, five targets, reflecting walls. 43 Table 3.6: Average of the two large clusters, five targets, lethal walls. : : vii LIST OF FIGURES Figure 1.1: Summary of the genetic algorithm. : : : : : : : : : : : : : : : 7 Figure 2.1: Overhead view of a symbot. : : : : : : : : : : : : : : : : : : 10 Figure 2.2: The Symbot World : : : : : : : : : : : : : : : : : : : : : : : 11 Figure 2.3: The symbot motion loop. : : : : : : : : : : : : : : : : : : : : 12 Figure 2.4: Summary of the genetic algorithm. : : : : : : : : : : : : : : : 15 Figure 3.1: Typical symbot path in the single source environment. : : : : 22 Figure 3.2: Path fitness over generations for a single, centered source, no walls. : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 23 Figure 3.3: The two major clusters for path fitness with no walls : : : : : 24 Figure 3.4: Path fitness over generations for a single, centered source, reflecting walls. : : : : : : : : : : : : : : : : : : : : : : : : : : 25 Figure 3.5: Path plots of the first and fifth clusters in the reflecting walls environment. : : : : : : : : : : : : : : : : : : : : : : : : : : : 27 Figure 3.6: Path fitness over generations for a single, centered source, lethal walls. : : : : : : : : : : : : : : : : : : : : : : : : : : : 30 Figu...