Abstract | In this paper we describe the evolution of a discrete-time recurrent neural network to control a real mobile robot. In all our experiments the evolutionary procedure is carried out entirely on the physical robot without human intervention. We showthat the autonomous development of a set of behaviors for locating a battery charger and periodically returning to it can be achieved by lifting constraints in the design of the robot/environment interactions that were employed in a preliminary experiment. The emergent homing behavior is based on the autonomous development ofaninternal neural topographic map (which is not pre-designed) that allows the robot to choose the appropriate trajectory as function of location and remaining energy.

Using a simple computational model of visual perception and locomotion, obstacle avoidance behavior can emerge from evolution under selection pressure from an appropriate fitness measure. The Genetic Programming paradigm is used to model evolution. Both the structure of the visual sensor array, and the mapping from sensor data to motor action is determined by an evolved control program. The motor model assumes an innate constant forward velocity and limited steering. The agent can avoid collisions only by effective steering. Fitness is based on the number of simulation steps the agent can run before colliding with an obstacle.

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.

We have used an automatic programming method called Genetic Programming (GP) for control of a miniature robot. Our earlier work on real-time learning suffered from the drawback of the learning time being limited by the response dynamics of the robot's environment. In order to overcome this problem we have devised a new technique which allows learning from past experiences that are stored in memory. The new method shows its advantage when perfect behavior emerges in experiments quickly and reliably. It is tested on two control tasks, obstacle avoiding and wall following behavior, both in simulation and on the real robot platform Khepera.

Bedau et al.’s statistical classification system for long-term evolutionary dynamics provides a test for open-ended evolution. Making this test more rigorous, and passing it, are two of the most important open problems for research into systems of agents that actively process and transform their environment. This paper presents a detailed description of the application of this test to ‘Geb’, a system designed to verify and extend theories behind the generation of evolutionarily emergent systems. The result is that, according to these statistics, Geb exhibits unbounded evolutionary dynamics, making it the first autonomous artificial system to pass this test. However, having passed it, the most prudent course of action is to look for weaknesses in the test. The test is criticized, most significantly with regard to its normalization method for artificial systems. Furthermore, this paper presents a modified normalization method, based on component activity normalization, that overcomes these criticisms. The results of the revised test, when applied to Geb, indicate that this system does indeed exhibit open-ended evolution.

In autonomous agent systems, internal memory can be an important element to overcome the limitations of purely reactive agent behaviour. This paper presents an analysis of memory requirements for T-maze tasks well known as the road sign problem. In these tasks a robot agent should make a decision of turning left or right at the T-junction in the approach corridor, depending on a history of perceptions. The robot agent in simulation can sense the light intensity influenced by light lamps placed on the bank of the wall. We apply the evolutionary multiobjective optimization approach to finite state controllers with two objectives, behaviour performance and memory size. Then the internal memory is quantified by counting internal states needed for the T-maze tasks in noisy environments. In particular, we focused on the influence of noise on internal memory and behaviour performance, and it is shown that state machines with variable thresholds can improve the performance with a hysteresis effect to filter out noise. This paper also provides an analysis of noise effect on perceptions and its relevance on performance degradation in state machines. keywords: T-maze, delayed response task, evolutionary robotics, finite state machines, evolutionary multiobjective optimization, internal memory 1 1

The design of adaptive systems will be among the key research problems of the 21st century. This new field is emerging from several distinct lines of work. • Modern immunology is based on the theory of clonal selection and adaptive immunity. The remarkable recognition abilities of the vertebrate