Eduardo Zalama

A real-time, unsupervised neural network for the low-level control of a mobile robot in a nonstationary environment

Abstract This article introduces a real-time, unsupervised neural network that learns to control a two-degree-of-freedom mobile robot in a nonstationary environment. The neural controller, which is termed neural NETwork MObile Robot Controller (NETMORC), combines associative learning and Vector Associative Map (VAM) learning to generate transformations between spatial and velocity coordinates. As a result, the controller learns the wheel velocities required to reach a target at an arbitrary distance and angle. The transformations are learned in an unsupervised training phase, during which the robot moves as a result of randomly selected wheel velocities. The robot learns the relationship between these velocities and the resulting incremental movements. Aside from being able to reach stationary or moving targets, the NETMORC structure also enables the robot to perform successfully in spite of disturbances in the environment, such as wheel slippage, or changes in the robots plant, including changes in wheel radius, changes in interwheel distance, or changes in the internal time step of the system. Finally, the controller is extended to include a module that learns an internal odometric transformation, allowing the robot to reach targets when visual input is sporadic or unreliable.

Adaptive behavior navigation of a mobile robot

Describes a neural network model for the reactive behavioral navigation of a mobile robot. From the information received through the sensors the robot can elicit one of several behaviors (e.g., stop, avoid, stroll, wall following), through a competitive neural network. The robot is able to develop a control strategy depending on sensor information and learning operation. Reinforcement learning improves the navigation of the robot by adapting the eligibility of the behaviors and determining the linear and angular robot velocities.

Road Crack Detection Using Visual Features Extracted by Gabor Filters

Data has been acquired and used to train and test pavement management systems (PMS) and methods. The article discusses how PMS require detailed information on the current state of the roads in order to take appropriate actions to optimize expenditures on maintenance and rehabilitation. The presence of cracks is a crucial aspect to be considered. A solution based on an instrumented vehicle that is equipped with an imaging system, two Inertial Profilers, a Differential Global Positioning System (DGPS), and a webcam is presented. Information about the state of the road is acquired at normal road speed. A method that is based on the use of Gabor filters is used to detect longitudinal and transverse cracks. The methodologies used to create Gabor filter banks and the use of the filtered images as descriptors for subsequent classifiers are discussed in detail in the article. The article also evaluates three different methodologies for setting the threshold of the classifiers. Finally, an AdaBoost algorithm is used for selecting and combining the classifiers, which improves the results provided by a single classifier. The article discusses how suitable results have been obtained in comparison with other reference works.

An Effective Texture Mapping Approach for 3D Models Obtained from Laser Scanner Data to Building Documentation

: Obtaining virtual models from real buildings, terrains, or building works is a matter of increased interest in construction. The application of such models ranges from technical use in architecture and civil engineering, to multimedia presentation, or remote visits through the web. This is becoming possible thanks to recent advances in laser scanning technology and related 3D processing algorithms. Moreover, real texture mapped onto 3D models is often required for communication, cataloguing, or digital documentation projects. In this article, an effective methodology to obtain digital building documentation based on 3D textured models is presented. First of all, a brief presentation of laser scanners is given as their data are used. An approach for mapping photographic images onto 3D models is also presented. The proposed approach, based on a camera registration method, offers high flexibility as it is based on hand-held cameras and can be implemented in a computing-effective way. A method for automatic image selection in overlapped areas is also presented. Finally, some hints are given concerning the automatic extraction of sections, orthophotos, and feature lines from the models. Experimental results focused on heritage buildings are shown, which demonstrate the suitability of the proposed techniques.

A realistic, virtual head for human-computer interaction

In this paper an interactive and realistic virtual head oriented to human-computer interaction and social robotics is presented. It has been designed following a hybrid approach, taking robotic characteristics into account and searching for a convergence between these characteristics, real facial actions and animation techniques. An initial head model is first obtained from a real person using a laser scanner. Then the model is animated using a hierarchical skeleton based procedure. The proposed rig structure is close to real facial muscular anatomy and its behaviour follows the Facial Action Coding System. Speech synthesis and visual human-face tracking capabilities are also integrated for providing the head with further interaction ability. Using the said hybrid approach, the head can be readily linked to a social-robot architecture. The opinions of a number of persons interacting with this social avatar have been evaluated and are reported in the paper, as against their reactions when interacting with a social robot with a mechatronic face. Results show the suitability of the avatar for on-screen, real-time interfacing in human-computer interaction. The proposed technique could also be helpful in the future for designing and parameterizing mechatronic human-like heads for social robots.

An unsupervised neural network for low-level control of a wheeled mobile robot: noise resistance, stability, and hardware implementation

We have recently introduced a neural network mobile robot controller (NETMORC). This controller, based on previously developed neural network models of biological sensory-motor control, autonomously learns the forward and inverse odometry of a differential drive robot through an unsupervised learning-by-doing cycle. After an initial learning phase, the controller can move the robot to an arbitrary stationary or moving target while compensating for noise and other forms of disturbance, such as wheel slippage or changes in the robots plant. In addition, the forward odometric map allows the robot to reach targets in the absence of sensory feedback. The controller is also able to adapt in response to long-term changes in the robots plant, such as a change in the radius of the wheels. In this article we review the NETMORC architecture and describe its simplified algorithmic implementation, we present new, quantitative results on NETMORCs performance and adaptability under noise-free and noisy conditions, we compare NETMORCs performance on a trajectory-following task with the performance of an alternative controller, and we describe preliminary results on the hardware implementation of NETMORC with the mobile robot ROBUTER.

A framework for building mobile single and multi-robot applications

The complexity of robot software systems calls for the use of a well-conceived architecture together with programming tools to support it. One common feature of robot architectures is the modular decomposition of systems into simpler and largely independent components. These components implement primitive actions and report events about their state. The robot programming framework proposed here includes a tool (RoboGraph) to program and coordinate the activity (tasks) of these middleware modules. Project developers use the same task programming IDE (RoboGraph) on two different levels. The first is to program tasks that must be executed autonomously by one robot and the second is to program tasks that can include several robots and building elements. Tasks are described using a Signal Interpreted Petri Net (SIPN) editor and stored in an xml file. A dispatcher loads these files and executes the different Petri nets as needed. A monitor that shows the state of all the running nets is very useful for debugging and tracing purposes. The whole system has been used in several applications: A tour-guide robot (GuideBot), a multi-robot surveillance project (WatchBot) and a hospital food and laundry transportation system based on mobile robots.

A realistic facial animation suitable for human-robot interfacing

In this paper an interactive and realistic facial animation oriented to human-computer interaction and social robotics is proposed. An initial model is obtained from a real face using a range scanner, and is then animated using a hierarchical skeleton oriented procedure. The rig structure is close to that of real facial muscular anatomy and the behavior follows the Facial Action Coding System. The resulting animated face is suitable for both real-time interfacing in human-machine interaction such as a social robot, and for helping in design and parametrization of android heads.

Robot learning in a social robot

In this paper, research work on Arisco is described Arisco is a social robot built around a robotic head with gesture ability, visual and auditive perception and learning It is intended for interacting with people The general architecture is first described in the paper Then, the learning capacity of Arisco is addressed It learns and performs associations between different stimulus responses through several dynamic neural networks, guided by motivational drives Main contribution of this paper is the integration in a real robot of conditioning learning models based on a neural competitive network A number of experiments are discussed, covering stimulus competition, habituation and first and second order conditioning.

Fast 6D Odometry Based on Visual Features and Depth

The availability of affordable RGB-D cameras which provide color and depth data at high data rates, such as Microsoft MS Kinect, poses a challenge to the limited resources of the computers onboard autonomous robots. Estimating the sensor trajectory, for example, is a key ingredient for robot localization and SLAM (Simultaneous Localization And Mapping), but current computers can hardly handle the stream of measurements. In this paper, we propose an efficient and reliable method to estimate the 6D movement of an RGB-D camera (3 linear translations and 3 rotation angles) of a moving RGB-D camera. Our approach is based on visual features that are mapped to the three Cartesian coordinates (3D) using measured depth. The features of consecutive frames are associated in 3D and the sensor pose increments are obtained by solving the resulting linear least square minimization system. The main contribution of our approach is the definition of a filter setup that produces the most reliable features that allows for keeping track of the sensor pose with a limited number of feature points. We systematically evaluate our approach using ground truth from an external measurement systems.