Orazio Miglino

Phenotypic plasticity in evolving neural networks

We present a model based on genetic algorithm and neural networks. The neural networks develop on the basis of an inherited genotype but they show phenotypic plasticity, i.e. they develop in ways that are adapted to the specific environment The genotype-to-phenotype mapping is not abstractly conceived as taking place in a single instant but is a temporal process that takes a substantial portion of an individuals lifetime to complete and is sensitive to the particular environment in which the individual happens to develop. Furthermore, the respective roles of the genotype and of the environment are not decided a priori but are part of what evolves. We show how such a model is able to evolve control systems for autonomous robots that can adapt to different types of environments.

From simulated to real robots

Evolutionary robotics using genetic algorithms to evolve control systems for real robots is a powerful tool, since it allows an automatic evolution of control systems. However, evolutionary robotics has serious limitations because of the time involved. It is very time consuming to evolve whole populations of real robots for many generations. A simulated/physical approach where main parts of the evolution takes place in a simulator reduces the time consumption dramatically. We describe how the Khepera miniature mobile robot can be used to build its own simulator with a semi-autonomous process, how to evolve neural network control systems for the Khepera robot in the robots own simulator, and how to transfer the neural network control systems from the simulated to the real environment. By using this kind of simulator an expected gap in performance when transferring a robot control system from the simulator to the real environment is avoided.

Evolutionary robotics-a children's game

The authors explore the concept of development without programming by children. Especially, they look at the case of developing robot control systems. The evolutionary robotics approach has shown that in some cases, given a mathematically described fitness function, it is possible to achieve an automatic development of robot controllers. However, it is questionable how one is to construct the mathematical fitness function. So they applied an interactive genetic algorithm to the problem of developing robot controllers and achieved and evolutionary robotics approach that allows children without any programming knowledge to develop controller for LEGO robots. They used neural networks as robot controllers, and found that combining the interactive genetic algorithm with a kind of reinforcement learning-development at the evolutionary time scale combined with life-time development-reduces the development time drastically. Hence, they overcome one of the major drawbacks of the interactive genetic algorithm, namely the development time.

Evolving and Breeding Robots

Our experiences with a range of evolutionary robotic experiments have resulted in major changes to our set-up of artificial life experiments and our interpretation of observed phenomena. Initially, we investigated simulation-reality relationships in order to transfer our artificial life simulation work with evolution of neural network agents to real robots. This is a difficult task, but can, in a lot of cases, be solved with a carefully built simulator. By being able to evolve control mechanisms for physical robots, we were able to study biological hypotheses about animal behaviours by using exactly the same experimental set-ups as were used in the animal behavioural experiments. Evolutionary robotic experiments with rats open field box experiments and chick detours show how evolutionary robotics can be a powerful biological tool, and they also suggest that incremental learning might be fruitful for achieving complex robot behaviour in an evolutionary context. However, it is not enough to evolve controllers alone, and we argue that robot body plans and controllers should co-evolve, which leads to an alternative form of evolvable hardware. By combining all these experiences, we reach breeding robotics. Here, children can, as breeders, evolve e.g. LEGO robots through an interactive genetic algorithm in order to achieve desired behaviours, and then download the evolved behaviours to the physical (LEGO) robots.

Rats, nets, maps, and the emergence of place cells

We study through computer simulations the motion in space of small networks consisting of a few sensory, intermediate, and motor units linked by feedforward connections of initially random strengths. Evolutionary pressure, exerted through random differentiation and selective reproduction, can force such objects to adapt to perform elementary navigation tasks similar to those used in investigating hippocampal function in rats. The connection strengths resulting from the adaptation process are shown to provide intermediate units with response characteristics similar to those of place cells found in the rat hippocampus. These results illustrate the ease with which “place” units emerge in any minimal circuitry geared to solve simple navigation tasks, and highlight the importance of considering the complexity of the memory performance required, rather than the relatively trivial spatial “computations” involved, while using those tasks to explore hippocampal structure and function.

Encoding geometric and non-geometric information: a study with evolved agents

Vertebrate species use geometric information and non-geometric or featural cues to orient. Under some circumstances, when both geometric and non-geometric information are available, the geometric information overwhelms non-geometric cues (geometric primacy). In other cases, we observe the inverse tendency or the successful integration of both cues. In past years, modular explanations have been proposed for the geometric primacy: geometric and non-geometric information are processed separately, with the geometry module playing a dominant role. The modularity issue is related to the recent debate on the encoding of geometric information: is it innate or does it depend on environmental experience? In order to get insight into the mechanisms that cause the wide variety of behaviors observed in nature, we used Artificial Life experiments. We demonstrated that agents trained mainly with a single class of information oriented efficiently when they were exposed to one class of information (geometric or non-geometric). When they were tested in environments that contained both classes of information, they displayed a primacy for the information that they had experienced more during their training phase. Encoding and processing geometric and non-geometric information was run in a single cognitive neuro-representation. These findings represent a theoretical proof that the exposure frequency to different spatial information during a learning/adaptive history could produce agents with no modular neuro-cognitive systems that are able to process different types of spatial information and display various orientation behaviors (geometric primacy, non-geometric primacy, no primacy at all).

BRIDGING DIGITAL AND PHYSICAL EDUCATIONAL GAMES USING RFID/NFC TECHNOLOGIES

The physical educational games and the traditional psycho-pedagogical methodologies are deeply based on the manipulation of objects. The opportunity derived by some low-cost technologies could join the physical world with the digital tools creating Augmented Reality Environments based on the concepts of Internet of Things (IoT). This connection has all the capacities to enhance the traditional educational games played in the schools or at home with the digital tools in order to create more exiting learning activities and more appropriate for the new digital natives. In this field, the RFID/NFC technology seems to be a natural candidate due to its natural predisposition to be heavily connected to real objects and send the signal to the digital devices. In this paper, we describe how the RFID/NFC technology could be used to connect digital and physical didactic materials in this hybrid approach. We present three different applications and prototypes: a) Block-Magic, it is an educational games based on well-known Logic Blocks material in the framework of an European project (I); b) Walden PECS Communicator (II) a platform based on Picture Exchange Communication System (PECS) a worldwide methodology to enhance communication skills in autistic persons; c) WandBot (III, IV), a learning environment that combines toy robots, RFID-technology and serious games for scientific dissemination in science centres

Breedbot: an evolutionary robotics application in digital content

Purpose – This paper aims to describe an integrated hardware/software system based on evolutionary robotics and its application in an edutainment context.Design/methodology/approach – The system is based on a wide variety of artificial life techniques (artificial neural networks, genetic algorithms, user‐guided evolutionary design and evolutionary robotics). A user without any computer programming skill can determine the robots behavior in two different ways: artificial breeding or artificial evolution. Breedbot has been used as a didactic tool in teaching evolutionary biology and as a “futuristic” toy by several science centers. The digital side of Breedbot can be downloaded on the web site: www.isl.unina.it/breedbot Findings – The results in this pilot study suggest that using Breedbot in an educational context can be useful to improve learning in biology.Research limitations/implications – As this is a pilot study, one limitation is the small sample considered. The issue will be investigated further w...

Place cognition and active perception: a study with evolved robots

A study of place cognition and ‘place units’ in robots produced via artificial evolution is described. Previous studies have investigated the possible role of place cells as building blocks for ‘cognitive maps’ representing place, distance and direction. Studies also show, however, that when animals are restrained, the spatial selectivity of place cells is partially or completely lost. This suggests that the role of place cells in spatial cognition depends not only on the place cells themselves, but also on representations of the animals physical interactions with its environment. This hypothesis is tested in a population of evolved robots. The results suggest that successful place cognition requires not only the ability to process spatial information, but also the ability to select the environmental stimuli to which the agent is exposed. If this is so, theories of active perception can make a useful contribution to explaining the role of place cells in spatial cognition.

Breedbot: An Edutainment Robotics System to Link Digital and Real World

The paper describes Breedbot an edutainment software and hardware system that could be used to evolve autonomous agents in digital (software) world and to transfer the evolved minds in physical agents (robots). The system is based on a wide variety of Artificial Life techniques (Artificial Neural Networks, Genetic Algorithms, User Guided Evolutionary Design and Evolutionary Robotics). An user without any computer programming skill can determine the robot behaviour. Breedbot was used as a didactic tool in teaching Evolutionary Biology and as a futuristic toy by several Science Centers. The digital side of Breedbot is downloadable from www.isl.unina.it/breedbot.