Luiz Marcos Garcia Gonçalves

Increasing Students' Interest With Low-Cost CellBots

This paper introduces the use of a flexible and affordable educational robot specifically developed for the practical experimentation inherent to technological disciplines. The robot has been designed to be reconfigurable and extendible, serving as an experimental platform across several undergraduate courses. As most students have a mobile cell phone, this was used as the main control computer for the so-called CellBot, thus avoiding any need to deal with the details of microcontrollers or other embedded computing devices. Assessment results are also presented, based on a pre- and post-survey of student opinion administered to 204 science and engineering students from several universities. Among the conclusions are that 83% of the students prefer to use these low-cost robots as tools to improve their learning of the theory in several disciplines, and 71% of the students stated that they prefer to have their own robot to experiment with, instead of using a didactic kit loaned to them by the university.

Learning Coordination in Multi-Agent Systems Using Influence Value Reinforcement Learning

In this paper authors propose a new paradigm for learning coordination in multi-agent systems. This approach is based on social interaction of people, specially in the fact that people communicate each other what they think about their actions and this opinion can influence the behavior of each other. It is proposed a model in which agents, into a multi-agent system, learns to coordinate actions giving opinions about actions of other agents and also being influenced with opinions of other agents about their actions. This paradigm was used to develop a modified version of the Q-learning algorithm. This algorithm was tested and compared with independent learning (IL) and joint action learning (JAL) in two single state problems with two agents. This approach shows to have more probability to converge to an optimal equilibrium than IL and JAL Q-learning algorithms. Also, it does not need to make an entire model of all joint actions like JAL algorithms.

A comparative analysis of dissimilarity measures for clustering categorical data

Similarity and dissimilarity (distance) between objects is an important aspect that must be considered when clustering data. When clustering categorical data, for instance, these distance (similarity or dissimilarity) measures need to address properly the real particularities of categorical data. In this paper, we perform a comparative analysis with four different dissimilarity measures used as a distance metric for clustering categorical data. The first one is the Simple Matching Dissimilarity Measure (SMDM), which is one of the simplest and the most used metric for categorical attribute. The other two are context-based approaches (DIstance Learning in Categorical Attributes - DILCA and Domain Value Dissimilarity-DVD), and the last one is an extension of the SMDM, which is proposed in this paper. All four dissimilarities are applied as distance metrics in two well known clustering algorithms, k-means and agglomerative hierarchical clustering algorithms. In this analysis, we also use internal and external cluster validity measures, aiming to compare the effectiveness of all four distance measures in both clustering algorithms.

3D Probabilistic Occupancy Grid to Robotic Mapping with Stereo Vision

Environment mapping is considered an essential skill for a mobile robot in order to actually reach autonomy [1]. The robotic mapping can be defined as the process of acquiring a spa‐ tial model of the environment through sensory information. The environment map allows mobile robots to interact coherently with objects and people in this environment. The robot can safely navigate, identify surrounding objects and have flexibility to dealing with unex‐ pected situations. Without a map some important operations could be complex as the deter‐ mination of objects position in the surroundings of the robot and the definition of the path to be followed. These issues involve the importance of the mapping task be performed cor‐ rectly, since the acquisition of inaccurate maps can lead to errors in the inference of correct positioning of the robot, resulting in an imperfect implementation of these operations. Therefore there is a mutual dependence between inferring the exact localization of the robot and building an accurate map.

N-BOAT: An Autonomous Robotic Sailboat

This article presents the ongoing development of an autonomous sailboat control architecture and a prototype sailboat constructed for experimental validation of the proposed architecture. The main goal of the project is to allow long endurance autonomous missions for ocean monitoring. In order to accomplish such objective, the system relies on wind forces propulsion instead of motors. We present the mathematical model using PID and Fuzzy controllers to control the sail and the rudder. Furthermore, we also present a study of the hardware architecture that enables better control performance of the system.

Real-Time Localization of Mobile Robots in Indoor Environments Using a Ceiling Camera Structure

The localization of mobile robots in indoor environments finds lots of problems such as accumulated errors and the constant changes that occur at these places. A technique called global vision intends to localize robots using images acquired by cameras placed in such a way that covers the place where the robots movement takes place. Localization is obtained by marks put on top of the robot. This work proposes a mobile robot localization system at indoor environments using a technique called global vision to estimate the position of a robot in real time. Since the method relies only on information extracted from the current image, accurate position estimates can be calculated.

Web-Based Robot Programming Environment and Control Architecture

We propose a novel web-based programming architecture for robots. The proposed system offers a web interface that allows users to type their robot control programs, so no development environments and tools need to be installed in the users computer. From the web control panel, the user can also save, load and execute programs stored in the robots memory. For the implementation we use a smart phone as the robots main control computer, so the web server is embedded in the phone, which is fixed to the robot. The system allows different programming languages to be used and also takes advantage of sensors available in the phone.

Fuzzy Logic on a Polygenic Multi-Agent System for Steganalysis of Digital Images

Digital cryptography has being a solution for protecting transmission of data in applications such as electronic commerce (Luciano 2003), electronic vote (Kofler 2003), and digital Television (Macq 1995). However, an interceptor monitoring network flow could easily break purely encoded data and clear the contents of cryptographed messages. Steganography techniques came up in order to help improving this protection. The goal of steganography is to hide data into a covering message (envelop) in such a way that an interceptor has no way to notice the presence of a hidden message in its covering envelop. Note that one can combine both cryptography and steganography in order to achieve better security. For example an image can be enriched with visually imperceptible extra information that, when eventually noticed, could be understood as an eventual noise. This damaged image could serve thus as a camouflaging body that brings protected data to the other side of the communication process. Any media object can be used as the covering message, such as text, audio, video, network packages, and file systems. Digital images are known to be the most used media objects for this purpose due to its inherent artistic appeal.

Influence Value Q-Learning: A Reinforcement Learning Algorithm for Multi Agent Systems 1

The idea of using agents that can learn to solve problems became popular in the artificial intelligence field, specifically, in machine learning technics. Reinforcement learning (RL) is part of a kind of algorithms called Reward based learning. The idea of these algorithms is not to say to the agent what the best response or strategy, but, indicate what the expected result is, thus, the agent must discover what is the best strategy for obtaining the desired result. Reinforcement learning algorithms calculate a value function for state predicates or for state-action pairs, having as goal the definition of a policy that best take advantage of these values. Q-learning (Watkins, 1989) is one of the most used reinforcement learning algorithms. It was widely applied in several problems like learning in robotics (Suh et al., 1997; Gu & Hu, 2005), channel assignment in mobile communication systems (Junhong & Haykin, 1999), in the block-pushing problem (Laurent & Piat, 2001), creation of electricity supplier bidding strategies (Xiong et al. 2002), design of intelligent stock trading agents (Lee et al., 2004), design of a dynamic path guidance system based on electronic maps (Zou et al., 2005), mobile robots navigation (Barrios-Aranibar & Alsina, 2004; Tanaka et al., 2007), energy conservation and comfort in buildings (Dalamagkidis et al., 2007), resource allocation (Usaha & Barria, 2007; Vengerov, 2007), and others. In the other hand, the use of multi-agent systems became popular in the solution of computacional problems like e-commerce (Chen et al., 2008), scheduling in transportation problems (Mes et al., 2007), estimation of energy demand (Toksari, 2007), content based image retrieval (Dimitriadis et al., 2007), between others; and in the solution of problems involving robots like mail sending using robots (Carrascosa et al., 2008), rescue missions (Rooker & Birk, 2005), mapping of structured environments (Rocha et al., 2005), and others. Also, Q-learning and derived algorithms were applied in multi-agent problems too. For example a fuzzy Q-learning was applied to a multi-player non-cooperative repeated game (Ishibuchi et al., 1997), a hierarchical version of Q-learning (HQL) was applied to learn both the elementary swing and stance movements of individual legs as well as the overall coordination scheme to perform forward movements on a six legged walking machine

A Fuzzy Approach to Stereo Vision Using Pyramidal Images with Different Starting Level

We propose a stereo matching algorithm based on multiresolution correlation that varies the depth for the resolution level with which to start stereo calculation for each image pixel (or block of pixels). The initial depth depends on the images local characteristics. We propose to use a neural fuzzy approach to calculate the desirable depth for each pixel of one of the matching images and then use this starting depth to proceed with the multiresolution approach. Variable depth correlation reduces the errors caused by coarse levels. At the same time, the new fuzzy heuristic that we propose for calculating the desired depth keeps most of the blocks at a coarse level, thus having little impact on execution time. Variable depth correlation is expected to have little problems with very plain surfaces and borders, but is rather faster than usual algorithms. In the tests, the multiresolution algorithm proposed here performed faster than plain correlation, with much better results