Giovanni Adorni

Robot self-localization by means of vision

We present an application of vision-based object recognition capabilities to the self-positioning-problem of an autonomous robot. Alphanumeric signs are placed in the robot environment as position markers and perceived through an on-board CCD camera on a pan-tilt head. Sign recognition is performed by a neural network based system, driven by some a-priori knowledge about the characteristics of the objects used as markers (signs). When given a map of the location of markers, the robot is able to estimate its position from the information extracted through perceived images. Marker distances and angular displacements allow the computation of a position uncertainty region for the mobile robot. Even using common, human readable markers, localization is performed with an average position accuracy within a few centimeters.

Landmark-based robot self-localization: a case study for the RoboCup goal-keeper

Robot soccer competitions provide an interesting opportunity for robotics and artificial intelligence research. In particular, robot soccer players must navigate in a dynamic, partially unknown environment, co-operate with team-mates and compete with opponents, track moving objects to protect the goal and to kick the ball in the right direction, while performing real time visual perception tasks. Furthermore, all these capabilities must be integrated into a single and complete autonomous system. The paper describes our experience in building one of the players of ART (Azzurra Robot Team), designed to participate in the RoboCup competition. Such a player, the goal-keeper, is based on a Pioneer 1 robot controlled by means of a binocular vision system able to perform its tasks in real time.

A Cellular-Programming Approach to Pattern Classification

In this paper we discuss the capability of the cellular programming approach to produce non-uniform cellular automata performing two-dimensional pattern classification. More precisely, after an introduction to the evolutionary cellular automata model, we describe a general approach suitable for designing cellular classifiers. The approach is based on a set of non-uniform cellular automata performing specific classification tasks, which have been designed by means of a cellular evolutionary algorithm.

A Multi Language Environment to Develop Multi Agent Applications

HOMAGE is an environment for the development of multi agent systems integrating agent and object-oriented programming paradigms and offering two different programming levels: object and agent. The object level allows the use of three object-oriented programming languages (C++, Common Lisp and Java) to develop new agent models as well as the components that will be used to build the body of agents. The agent level allows the development of new agents by defining their “brain” and by composing the components defined at the object level, and allows the development of multi agent systems by distributing and interconnecting agent instances. Moreover, these multi agent systems can be distributed on a net of heterogeneous machines connected through internet taking advantage of a set of communication and distribution libraries allowing the communication between agents through different protocols. The paper includes a brief description of a robotic application that we implemented during the experimentation of the environment.

Genetic Programming of a Goal-Keeper Control Strategy for the RoboCup Middle Size Competition

In this paper we describe a genetic programming approach to the design of a motion-control strategy for a goalkeeper robot created to compete in the RoboCup99, the robot soccer world championships which have been held yearly since 1997, as part of the Italian middle size robot team (ART, Azzurra Robot Team). The evolved program sends a motion command to the robot, based on the analysis of information received from a human-coded vision sub-system. The preliminary results obtained on a simulator are encouraging. They suggest that even using very simple fitness functions and training sets including only a small sub-set of the situations that the goalkeeper is required to tackle, it is possible to evolve a complex behavior that permits the goalkeeper to perform well also in more challenging real-world conditions.

Team/Goal-Keeper Coordination in the RoboCup Mid-Size League

In this paper we describe the coordination strategies that were designed to achieve effective cooperation between a robot goal-keeper and the rest of the ART (Azzurra Robot Team) team that participates in the RoboCup F-2000 (mid-size) competitions. The paper introduces the multi-robot environment on which cooperation in ART is based and, in particular, its communication sub-system. Some case studies and the results of their application in the ART team are then described.

Design of Explicitly or Implicitly Parallel Low-resolution Character Recognition Algorithms by Means of Genetic Programming

The paper describes two approaches to low-resolution character recognition that are either implicitly or explicitly based on the SI MD (Single Instruction Multiple Data) computation paradigm. In both approaches, a set of binary classifiers have been designed by means of evolutionary computation techniques.

ART'00 - Azzurra Robot Team for the Year 2000

Robotic soccer is a challenging research domain that can be used to explore new problems and test new techniques/solutions in the fields of Artificial Intelligence and Autonomous Robotics, as well as for training and educational purposes.

Cellular Automata Based Inverse Perspective Transform as a Tool for Indoor Robot Navigation

In this paper we describe a system for perspective-effect removal using the cellular automata paradigm. The system has been developed as a tool to be included in the vision system of an autonomous robot designed to operate in indoor environments. The navigation of such a robot is guided by traffic signs by means of a neural-network based system. By using such a tool as part of the vision system, it is possible to instruct the neural networks with a training set including only frontal views of the signs, thus simplifying training and making it computationally much lighter. Furthermore, using a low-cost massively parallel architecture implementing the cellular automata paradigm, makes the algorithm much more computationally efficient with respect to sequential implementations, thus making it possible for the robot to perform just-in-time operations.

MAP - a Language for the Modelling of Multi-Agent Systems

This paper presents a distributed object-oriented language, called Multi-Agent Programming language, whose features are suitable to develop multi-agent systems. This language is based on an object, called agent, (i) performing private actions, (ii) communicating with other agents, and (iii) re-configuring system structure through the creation of other agents and changing its behaviour. The main feature of this language is the use of a large set of communication primitives, defined on the spirit of Speech Act theory, which are suitable to model agent interactions and which can be specialised to implement specific communication protocols. In particular, the paper shows how these primitives are suitable to model negotiation protocols.