Showing papers presented at "Simulation of Adaptive Behavior in 1994"

Automatic creation of an autonomous agent: genetic evolution of a neural-network driven robot

Abstract: The paper describes the results of the evolutionary development of a real, neural-network driven mobile robot. The evolutionary approach to the development of neural controllers for autonomous agents has been successfully used by many researchers, but most - if not all - studies have been carried out with computer simulations. Instead, in this research the whole evolutionary process takes places entirely on a real robot without human intervention. Although the experiments described here tackle a simple task of navigation and obstacle avoidance, we show a number of emergent phenomena that are characteristic of autonomous agents. The neural controllers of the evolved best individuals display a full exploitation of non-linear and recurrent connections that make them more efficient than analogous man-designed agents. In order to fully understand and describe the robot behavior, we have also employed quantitative ethological tools [13], and showed that the adaptation dynamics conform to predictions made for animals.

Seeing the light: artificial evolution, real vision

Abstract: This paper describes results from a specialised piece of visuo-robotic equipment which allows the arti cial evolution of control systems for visually guided autonomous agents acting in the real world. Preliminary experiments with the equipment are described in which dynamical recurrent networks and visual sampling morphologies are concurrently evolved to allow agents to robustly perform simple visually guided tasks. Some of these control systems are shown to exhibit a surprising degree of adaptiveness when tested against generalised versions of the task for which they were evolved.

Action-selection in hamsterdam: lessons from ethology

Abstract: A computational model of action-selection is presented, which by drawing on ideas from Ethology, addresses a number of problems which have been noted in models proposed to date including the need for greater control over the temporal aspects of behavior, the need for a loose hierarchical structure with information sharing, and the need for a flexible means of modeling the influence of internal and external factors. The paper draws on arguments from Ethology as well as on computational considerations to show why these are important aspects of any action-selection mechanism for animats which must satisfy multiple goals in a dynamic environment. The computational model is summarized, and its use in Hamsterdam, an object-oriented tool kit for modeling animal behavior is discussed briefly. Results are presented which demonstrate the power and usefulness of the novel features incorporated in the algorithm.

A case study in the behavior-oriented design of autonomous agents

Abstract: Trabajo presentado a la 3rd International Conference on Simulation of Adaptive Behavior (SAB), celebrada en Brighton (Inglaterra) del 8 al 12 de 1994.

Integrating reactive, sequential, and learning behavior using dynamical neural networks

Abstract: This paper explores the use of dynamical neural networks to control autonomous agents in tasks requiring reactive, sequential, and learning behavior. We use a genetic algorithm to evolve networks that can integrate these different types of behavior in a smooth, continuous manner. We apply this approach to three different task domains: landmark recognition using sonar on a real mobile robot, one-dimensional navigation using a simulated agent, and reinforcement-based sequence learning. A novel feature of the learning aspects of our approach is that we assume neither an a priori discretization of states or time nor an a priori learning algorithm that explicitly modifies network parameters during learning. Instead, we expose dynamical neural networks to tasks that require learning and allow the genetic algorithm to evolve network dynamics that generates the desired behavior.

Evolution of corridor following behavior in a noisy world

Abstract: Robust behavioral control programs for a simulated 2d vehicle can be constructed by artificial evolution. Corridor following serves here as an example of a behavior to be obtained through evolution. A controller’s fitness is judged by its ability to steer its vehicle along a collision free path through a simple corridor environment. The controller’s inputs are noisy range sensors and its output is a noisy steering mechanism. Evolution determines the quantity and placement of sensors. Noise in fitness tests discourages brittle strategies and leads to the evolution of robust, noise-tolerant controllers. Genetic Programming is used to model evolution, the controllers are represented as deterministic computer programs.

From SAB90 to SAB94: four years of animat research

Abstract: This paper builds on a previous review of significant research on adaptive behavior in animats. It summarizes the current state of the art and suggests some directions likely to provide interesting results in the near future.

Integration of reactive and telerobotic control in multi-agent robotic systems

Abstract: Multi-agent schema-based reactive robotic systems are complemented with the addition of a new behavior controlled by a teleoperator. This enables the whole society to be aaected as a group rather than forcing the operator to control each agent individually. The operator is viewed by the reactive control system as another behavior exerting his/her innuence on the society as a whole. Simulation results are presented for foraging, grazing, and herding tasks. Teleau-tonomous operation of multi-agent reactive systems was demonstrated to be signiicantly useful for some tasks, less so for others.

A hierarchical classifier system implementing a motivationally autonomous animat

Abstract: This work describes a control architecture based on a hierarchical classi er system This architec ture which uses both reactive and planning rules implements a motivationally autonomous animat that chooses the actions it will perform accord ing to the expected consequences of the alterna tives The adaptive faculties of this animat are illustrated through various examples

Learning efficient reactive behavioral sequences from basic reflexes in a goal-directed autonomous robot

Abstract: This paper describes a reinforcement connectionist learning mechanism that allows a goaldirected autonomous mobile robot to adapt to an unknown indoor environment in a few trials. As a result, the robot learns efficient reactive behavioral sequences. In addition to quick convergence, the learning mechanism satisfies two further requirements. First, the robot improves its performance incrementally and permanently. Second, the robot is operational from the very beginning, what reduces the risk of catastrophic failures (collisions). The learning mechanism is based on three main ideas. The first idea applies when the neural network does not react properly to the current situation: a fixed set of basic reflexes suggests where to search for a suitable action. The second is to use a resource-allocating procedure to build automatically a modular network with a suitable structure and size. Each module codifies a similar set of reaction rules. The third idea consists on concentrating the exploration of the action space around the best actions currently known. The paper also reports experimental results obtained with a real mobile robot that demonstrate the feasibility of our approach.

Achieving rapid adaptations in robots by means of external tuition

Abstract: Experiments with the Edinburgh R2 mobile robot are presented that show how robots can be taught to accomplish various diierent tasks, without the need for reprogramming the controller, and without using self-tuition. In an externally supervised teaching process | not unlike the process of \shaping" known in animal learning | the robot acquires compe-tences such as obstacle avoidance, contour following , box pushing, phototaxis or route learning (mazes). The learning is fast.

MINIMEME: of life and death in the noosphere

Abstract: This paper deals with artificial animals able to communicate beliefs about their environment’s properties to each other. In order to study the relations existing between the information exchanged and the emergence of behaviors or organizations, we propose a model called MINIMEME. This model exploits Dawkins’ paradigm stating that ideas, or memes, can be compared to parasites infecting their host and trying to duplicate themselves in other hosts’ memories. MINIMEME models the interactions occurring between the world of the memes and the animal societies. It is shown that some ecological concepts can be applied to memes, as if they were some kind of animats, and how these concepts relate to the simulated animals. The way this model can be used as a tool to help objectively qualifying emergent behaviors in simulated societies is then discussed.

The blind breeding the blind: adaptive behavior without looking

Abstract: Sensors and internal states are often considered necessary components of any adaptively behaving organism, providing the information needed to adapt a creature’s behavior in response to conditions in its external or internal environment. But adaptive, survivalenhancing behavior is possible even in simple simulated creatures lacking all direct contact with their environment — evolutionarily shaped blind action may suffice to keep a population of creatures alive and reproducing. In this paper, we consider the evolution of the behavioral repertoires of such sensor-less creatures in response to environments of various types. Different spatial and temporal distributions of food result in the evolution of very different behavioral strategies, including the use of looping movements as time-keepers in these otherwise cognitively-challenged creatures. Exploring the level of adaptiveness available in even such simple creatures as these serves to establish a baseline to which the adaptive behavior of animats with sensors and internal states can be compared.

Reinforcement learning with dynamic covering of state-action space: partitioning Q-learning

Abstract: An adjustable X-ray collimator is disclosed which has two web assemblies. Each web assembly has a pair of spaced and connected webs which form a continuous loop reaved over a pair of rollers. The assemblies are positioned near and parallel with one another with the axes of the rollers on one assembly being perpendicular to the other so that one assembly defines the sides and the other assembly the ends of a rectangular X-ray beam opening. The size of the opening is adjusted by rotating the rollers so as to move the interconnected webs to adjust the amount of space between ends of the webs.

The Interval Reduction Strategy for Monitoring Cupcake Problems

Abstract: Monitoring is the process by which agents assess their environments. Most AI applications rely on periodic monitoring, but for a large class of problems this is inefficient. The interval reduction monitoring strategy is better. It also appears in humans and artificial agents when they are given the same set of monitoring problems. We implemented two genetic algorithms to evolve monitoring strategies and a dynamic programming algorithm to find an optimum strategy. We also developed a simple mathematical model of monitoring. We tested all these strategies in simulations, and we tested human strategies in a "video game." Interval reduction always emerged. Environmental factors such as error and monitoring costs had the same qualitative effects on the strategies, irrespective of their genesis. Interval reduction appears to be a general monitoring strategy.

What is cognitive and what is not cognitive

Abstract: The ubiquitous contemporary use of the term 'cognitive' brings certain drawbacks. In the tradition of Tolman (1932), closer attention might be paid to its meaning in a way that can demarcate cognitive from non-cognitive processes. This has implications for understanding learning, motivation and behavioural hierarchies. It suggests different evolutionary processes, reveals similarities in information processing in rats and humans and is relevant to the design of automatons.