Research Experiences for Undergraduates in Distributed Rational Agents

Open House 2004

Nine undergraduate participants of the REU summer program participated in the REU Open House 2004 to present their research projects.

• Simulation and Interfaces for Intelligent Environments

  The goal of this project was to develop a simulation environment for Intelligent Environments that facilitates the simple introduction of new components and devices (such as lamps, furniture, etc.) and their integration into the system without the need for restarting or programming effort. The purpose of the simulation is to allow for off-line experiments with prediction and decision-making algorithms as well as to serve as a graphical interface for the devices in the home. For the latter, the graphical simulation permits lights to be turned on and off via mouse clicks on the screen.
  To achieve the simulation goals, one REU students performed research in the development of graphical models and their integration using automatic service discovery tools. In particular, starting with results from the previous year's REU program, the student developed a set of tools that permit the easy creation of graphical models for objects in the environment and that integrate such a model into a Rendezvousä object. This permits devices to be added to the environment by simply running the corresponding object on a computer node in the network. The objects are then automatically discovered by the simulation server which reads their specifications (graphical model, location, and operation) and integrates them into the simulation interface. In this way, new devices can be incorporated into the interface by simply putting them onto the computer network. The figure shows the graphics design tool and the simulation environment developed.

• Anomaly Detection for In-Home Lighting Systems

  In this project, one REU participant performed research in anomaly detection using sensors and actuators in intelligent environments. The focus here was in the ability to detect problems in the lighting system in the house using the sensors network deployed throughout the environment. Here, the student developed models for the lighting and sensor system and used these to monitor the light levels to detect incorrect light commands or faulty light sources. In addition, the student integrated these models and verification mechanisms with an automatic service discovery framework to permit light sources to be added or removed from the environment. Evaluations of the anomaly detection scheme showed good results at determining faulty lights or errors in the switching of light sources. Combined with a set of action rules, the system was then used to successfully correct errors in the X10-based light control system in the MavHome laboratory.

• Data Visualization

  Another challenge arising in complex smart home environments is the monitoring and evaluation of the huge amounts of data created by all the sensor network nodes in the home. To address this, one REU participant performed research into data visualization techniques and developed a web-based interface to the sensor network in MavHome that permits the hierarchical analysis of the sensor data. To develop this interface, the student investigated different techniques as well as human factors involving the useful presentation of spatially and temporally distributed sensor data. The resulting system permits the monitoring of single or multiple sensors at different time scales. Moreover, it permits the selection of sensors by home regions and sensor types. The figure shows two examples of data from different sensors in the home. The data on the left shows the changes in temperature readings over a one month period while the graph on the left illustrates the activities on a light switch over a one hour time span.

• Biometrics for Key-less Entry and Security

  The goal of this project was to research biometric identification mechanisms for secure entry into a smart home. One REU participant here investigated a range of techniques including face recognition, voice profiling, and finger print analysis. He then developed a system that uses face recognition to determine the identity of a person. To do this, he augmented a system developed in a previous REU project by implementing image processing algorithms that permit the robust extraction of a face in the presence of a cluttered background. The system developed here uses research results that show that the color range of skin falls into a narrow range when viewed in a HSV color format. The figure shows the result of the face finding algorithm. Starting from the image on the left, the algorithm identifies skin regions and highlights them, leading to the coded image on the right. Using this image, the system developed by the student then determines the eye locations and uses a face recognition algorithm based on Eigenfaces to identify the person.

• Interfaces Specifications for Smart Appliances

  An important component of smart homes is the ability of the home to monitor and control devices in the environment. To make this possible, however, it is essential that ``smart'' devices use common interface descriptions that permits a home to interface with them. The goal of this project was to develop such interface specifications for a range of appliances that are common in home environments. Here one REU participant developed a hierarchical interface specifications for heating and cooking appliances in the kitchen. Different appliances, including stove tops, microwave ovens, and stoves were modeled as part of this project, interface specifications were created, and virtual appliance simulations were implemented. The interface specifications for all devices here inherit common traits (such as temperature regulation, timing for cooking tasks, etc.) from a heating device class and are then further augmented with components describing specific capabilities of the particular device. The figure shows two examples of virtual devices developed in this project. As indicated before, the interface specifications for both of these appliances share a common core that is then augmented to capture all capabilities of the final devices. The reason for such a hierarchical interface definition is to permit new appliances (with their basic capabilities) to be integrated into existing smart home environments without major changes to the surrounding software.

• Control of Service Robots

  To fully utilize the capabilities of smart homes, service robots are an important component. Over the summer one REU participant performed research into the control of basic service robots. In particular, he developed a control system that permits the precise control of a robot platform while providing for a simple means of integrating higher-level control and decision-making component. The figure shows the robot platform used for this project. The student further integrated this low-level control system with a previously developed system for shared robot/human autonomy to form a robot system that can navigate to designated locations. During this navigation it integrates user commands to form strategies that match the particular preferences of the user.

• Wireless Localization in In-Door Environments

  To make intelligent environments more competent it is important that the locations of inhabitants and of various mobile devices are known. The goal of this project was the implementation of a system that permits wireless nodes to be localized using the received signal strength (RSSI) readings from the wireless network. This permits such devices to be localized without the need for additional sensor infrastructure. Over the summer one REU student here, starting from a system that was developed in prior research, investigated wireless localization techniques and implemented a system that uses sequential Monte Carlo Sampling techniques (particle filters) to estimate the location of a wireless-enabled PDA in the {\em MavHome} laboratory. As part of this, he developed a wireless map that describes the signal strength throughout the environment and created an algorithm that fuses a series of readings into an estimate of the location. The figure shows the layout of the lab and the measured signal strength map. In this map, locations with a stronger signal are indicated with a darker shade of red. While the distribution of signal strengths throughout the lab proved to be very ambiguous, the resulting system performed reasonably well, allowing the determination of the approximate location of the device.

For additional information regarding this program please contact:

Manfred Huber
University of Texas at Arlington
Department of Computer Science and Engineering
Box 19015
Arlington, TX 76019-0015
Email: reu@cse.uta.edu

* This project is funded by NSF grant EIA-0139564