Self-Learning Game Playing System
Authors: Tom van Buskirk
Learning systems are an exciting area of research for various applications. A software package that can learn about and respond according to its environment has potential uses in a variety of products. Through learning, a general system could theoretically conform to the needs of many specific applications.
Existing game playing systems tend to be algorithmic in nature, with hard coded rules and strategies. These systems are designed to play one specific game and must be largely rebuilt for each new game. In this project, we develop a general system that learns the strategies for a variety of card games, specifically Poker and Golf.
Our learning system consists of two main components: a learning component that uses Artificial Neural Networks in conjunction with Genetic Algorithms, and a main component that governs the learning process in accordance with the game definitions. Over the course of the learning process, the neural networks guess at various game playing strategies. The main program gives performance feedback, which the genetic algorithms use to revise and improve these guesses. The result is that our system can learn to play Poker and Golf better than a player who acts at random.
Card games provide a specific application for a general learning system. Future versions of our software can be expanded to include more games and to try to achieve an optimal playing strategy. Moving beyond card games, similar systems could be used to meet the demands of a number of more complex environments.
View Poster in PDF, jpg
Using Near Infrared Spectroscopy to Detect Brain Function
Authors: Matthew Verghese
The human brain is often referred to as the “final frontier” of medical science. Much work has been done to study the cognitive function associated with its particular neural functions. Because of the limitations on physically measuring the changes in the brain, non-invasive methods are necessary in order to study the complexities of its behavior.
Near Infrared Spectroscopy (NIRS) is one such technique for safely measuring the changes in blood volume of the brain during cognitive functions for in vivo experiments. It has been suggested that certain cognitive functions induce changes in the blood volume levels of the prefrontal cortex area of the brain, and that observing these changes in blood volume levels can be used to predict these same cognitive thought processes.
Currently, the ability to accurately predict a particular cognitive brain function given a pattern in the change of blood volumes levels is limited. Thus, the goal of this project is to aid researchers in the study of cognitive brain function by providing them with a simple and cheap handheld brain sensor device that can detect changes in the blood volume levels of the brain.
The device, a handheld sensor that can be placed on the forehead, generates two infrared sinusoidal signals 180 degrees out of phase. These signals penetrate the brain and based on the blood volume content of the brain tissue an aggregate signal of particular amplitude is received by a photodiode.
Wirelessly transmitting the amplitude of this aggregate signal to a receiving station and plotting the results using computer software enables the researcher to associate particular blood volume patterns with certain cognitive functions.
Low Power Wireless Security Network
Authors: Lee Becker
Advisor: Van der Spiegel, Gruev
With the development of low power wireless technologies such as Motes and the Zigbee protocol, many applications involving wireless systems can be significantly improved. Home security systems could potentially benefit from this reduction in power, resulting in smaller and more modular devices.
SmartShelter is a low-cost, low-power wireless security system designed for people who frequently move homes, rent homes, and live in lower income housing. It uses the Zigbee protocol, implemented with PIC microcontrollers and Chipcon transceivers configured in a hub and spoke set-up. Multiple end-nodes perform the actual monitoring using a cluster of transducers and relay this information to a central coordinator that is responsible for overall network management and emergency response.
The system is designed to perform the following security functions in a home:
- Doorway Monitoring – Utilizes a combination of a magnetic trip and motion sensor to determine a forced entry through a door.
- Window Monitoring – Combines motion and sound detection with a magnetic trip to detect a broken window.
- Fire Detection – Searches for smoke and or excess heat to provide early warning for a home fire.
Neural Net Predicting Stock Price
Authors: Adam MacLeod
The use of neural networks in stock market prediction algorithms gained popularity in the mid-1990s with the advent of computers with processing power capable of running algorithms that attempt to approximate learning processes occurring in the human brain. Many companies have successfully implemented such algorithms for the purpose of predicting the behavior of a system, whether to predict rainfall based on space and time, to predict the identity of passengers with nothing more than a picture of their faces, to predict the localization of proteins based on their amino acid sequence, or to predict stock market returns. The major problem faced in the financial world is that if one of these models is successful, companies do not report what made the model successful, as by doing so, the company risks creating competition for the model created and for variants which could be derived from that model.
This research aims to use insider trading data as well as historical price data to make predictions regarding the price of a given stock after an insider purchase is made for that particular stock. By analyzing each purchase over a given time period through the use of a system incorporating a genetic algorithm and a neural network, one can make a relatively accurate one month prediction of a given company's future price.
Our results indicate that over the time period from 2002-2004 using insider trade information from the gaming industry, we were able to achieve cumulative returns of 151.6%, a 36% annualized return. Considering that the average annual S&P 500 returns hover around 10%, we have clearly achieved our stated objective of exceeding the returns of the S&P 500 index.
Authors: Cathy Chen
Albert Christopher Purdy
The typical commercial battery has become the staple solution for portable power in today’s society of mobile electronic devices. Unfortunately, current commercial batteries possess low energy density, short life spans, and are harmful to the environment upon disposal. The goal of the micro-combustor project is to combat the aforementioned disadvantages of the commercial battery by providing a competitive, portable energy source.
A promising alternative to electrochemical batteries involves the combustion of liquid hydrocarbon based fuels. Since liquid hydrocarbon based fuels employ energy densities two orders of magnitude greater than commercial batteries, these fuels are able to provide an ideal source for mobile power generation. However, conventional methods of electrical generation require the combustion of hydrocarbons to drive a mechanical generator. The inclusion of the mechanical generator increases both the size and weight of such a system and reduces its efficiency due to intermediate energy conversions.
Unlike a conventional generator, the micro-combustor will harness the heat produced from combustion and directly convert it to electricity through a thermoelectric element. The absence of mechanical parts in a micro-combustor allows for a much smaller size and quiet operation while generating power. These features allow a micro-combustor to provide portable energy similar to a battery while allowing it to utilize greater energy density through hydrocarbon- based fuels.
Tactile Sensing Cockroach
Authors: Kelvin Hu
In the field of robotics, there has been a tremendous amount of resources dedicated to solving the problems of autonomous navigation and obstacle avoidance. Traditional approaches to robotic navigation have used “vision” systems that involve gathering data using cameras and processing the data using image recognition algorithms. The problem with this approach is that it is extremely computationally intensive and ineffective in low light conditions. From a biological standpoint, consider that almost one-third of the brain capacity of higher level mammals is dedicated to image/vision processing.
The chosen approach is inspired by nature, specifically the segmented antennae of cockroaches. Mini-Rhex, a hexapedal robot, extracts tactile information from the surrounding environment using flex sensors. When the sensors come into contact with any surface or obstacle, deformation of the flex sensors produces a corresponding voltage that is sampled by the robot and converted into digital form. This data is used in behavioral algorithms to govern the reaction of the robot to its environment.
The full implementation of this tactile-sensing system allows mini-Rhex to react dynamically to its environment while maintaining a velocity of up to 0.5 m/s. Wall following along with obstacle avoidance are implemented in mini-Rhex’s behavior. This system significantly enhances the range of mini-Rhex’s capabilities in both unfamiliar environments and conditions of poor lighting.
Digital Page Reconstruction
With so much of our personal information stored digitally these days, and an ever-growing risk of identify theft, people are quick to destroy physical documents containing any personally-identifying information. For most people, this destruction takes place at either the mercy of their own hands as they tear up the page, or through use of a mechanical shredder. In the case of accidental destruction, no quick method exists to recover the page except to painstakingly put it back together again by hand. The goal of this project is to take a digital representation of those pieces and rebuild the original page.
The system consists of a Graphical User Interface (GUI) to interact with the user, a database to store the current and previous project files, and two distinct matching systems. The first matching system addresses the shredded page problem using techniques analyzing colors along the edges of the shredded pieces. Using this information, a scoring algorithm merges the pieces until only one remains. This remaining piece is returned as a representation of the original image. The second matching system addresses the problem of the torn page by using code based on the color matching algorithms of the shredded page system and a shape matching algorithm. The software uses these algorithms to first reconstruct the border of the image, and then add the remaining pieces to the middle of the page. When all the pieces have been placed, a representation of the original image is returned.
Time of Flight Trigger for PET
Author: Robert Yu
Advisor: Newcomer, Van Berg
Recent advancements in processing technology and crystal research have allowed improvements in PET scanner technology. These improvements include the addition of time-of-flight feature. Conventional scanners do not use the physical properties of gamma rays to filter out noise. Consequently, refracted rays that occur due to Compton scattering and rays from cosmic radiation get recorded in the picture and result in blurring. However, with the advent of time-of-flight, these negative effects can be accounted for.
Time-of-flight technology requires the appropriate crystals to meet its timing demands. Current research in Lanthanum Bromide (LaBr3) crystals has made the goal viable. The Medical Physics department has decided to use LaBr3 because it has high light output, high stopping power, good linearity, and fast decay time. The crystals will be packaged in the anger-logic arrangement.
Along with the suitable crystals, a data processing entity needs to do the filtration process. This is where the time-of-flight trigger board comes in. The trigger board examines the geometrical mapping of the incident gamma rays and filters out the pairs that create noise, thereby logically identifying only those that can be potential coincidences. The filtration is done in real time with a Virtex FPGA Xilinx chip. The signals are filtered out based on a coincidence angle (referred to as “alpha”), where the angle is determined through some physical aspects of the patient being scanned, e.g. height and weight.
The algorithms used in the board are scalable, so that scanners of a larger size can potentially be used. However, the current algorithm was designed for a two dimensional scanner, while delay times for larger scanners were simulated.
Voice to Drum Real-Time Recognition
Authors: Aryeh Levin
Speech recognition is one of the key tools that developers hope to use in making the next generation of computer applications more user-friendly and in using computers to simplify the lives of everyday users. A number of obstacles remain in the way of this goal and must be corrected before speech recognition can become a truly mainstream technology.
One of these obstacles is the difficulty in distinguishing stop consonants, those sounds created by stopping the flow of air in the mouth and letting it go in a burst (e.g. p, t, k). Since stop consonants all follow a similar pattern of a stop followed by a burst, and are often unvoiced when pronounced, telling them apart is a particularly difficult problem. Building a system that can distinguish stop consonants may help bring the world of speech recognition one step closer to its larger goals of an ideal user interface.
A rudimentary speech-recognition system has been developed that uses the fundamentals of Artificial Intelligence to recognize a limited number of ‘words.’ These ‘words’ are based on the problematic stop consonants. This system is applied and demonstrated with a ‘beat-box’ system, allowing a user to speak specific sounds into a microphone and have a computer parse the sounds and play a corresponding drumbeat.
In the chosen approach, voice is recorded from the user and immediately processed using a barrage of Digital Signal Processing techniques. These techniques are designed to simulate the operation of the ear, the best working speech-recognition package available, which emphasizes certain frequencies over others and hears in a logarithmic frequency scale. These DSP routines ultimately output a set of cepstral coefficients, containing the key recurring frequency components of the input signal.
These cepstral coefficients are passed along to an Artificial Intelligence system, which is based on the Hidden Markov Model, and uses one model for each sound in the ‘dictionary’ of sounds being detected. The model outputs a probability that the recorded sound came from a specific intended letter. The model with the highest probability is selected, and its corresponding drumbeat is played.
Authors: Dong Tran
The decrease of physical activity in children has significant health consequences for society because it is directly linked to an increase of obesity in children and osteoporosis in adults. Research has shown that the intensity of weight-bearing activities is crucial in developing bone mass and strength. Last year’s senior design team developed a Pediatric Step Monitor that measures the magnitude and duration of forces exerted on the soles of children’s feet. This device was intended for use in current bone health studies to detect trends and quantify the correlation between load-bearing activity and bone development. However, because of the bulkiness of the electronic box and the necessity to transmit data in real-time via a Bluetooth, the device was still not ready for use on children on a daily basis.
The Pediatric Dynamometer that will be designed this year consists of a piezoelectric polyvinylidene fluoride sensor, a signal conditioning circuit, a microcontroller for data processing, a memory chip for data storage and a lithium cell battery. The whole device, sensors and electronic box inclusive, will be embedded in a child’s shoe. At the end of the day, the device will be removed and will be connected to an external circuit board containing an RS232 driver. The data from the memory chip is downloaded via RS232 serial protocol onto a computer to a simple graphical interface, displaying information such as force magnitude and energy expenditure.
Encoding/Decoding System for PAPPA Experiment
Authors: Angelos Stamatakis
Jose Gabriel Ramos
There has been a significant amount of research in the field of physical cosmology that attempts to explain the origins of the universe through scientific observation. A topic of interest of Dr. Mark Devlin of the University of Pennsylvania is the search for anisotropy in the Cosmic Microwave Background (CMB). The Primordial Anisotropy Polarization Pathfinder Array (PAPPA) balloon flight experiment, which uses a wide array of instruments to perform precise measurements of incident radiation from remote, high-altitude locations, is a major component of Dr. Devlin’s work.
In order to analyze the large amount of data that is generated from these measurements, the data must be transmitted wirelessly from the remote collection points to stations on the ground. The transmission system used to send this data must include an encoding and decoding system, a component which the PAPPA experiment currently lacks.
The chosen approach is to extract raw binary data from a PC and encode it for transmission, which will be accomplished using existing transmitter and receiver hardware. The Manchester encoding scheme provides an added layer of protection from idle states and long sequences of high or low voltage levels which are undesirable in digital systems. The encoding process will also efficiently separate the data into frames and implement an error control scheme that will correct many transmission errors. The encoded signal will be properly decoded and restored to its original form after transmission so that it can be recorded in a PC at the receiving station. Error correction is then performed so that the received data exactly matches the data that was extracted from the data acquisition station.
The design incorporates two layers. The hardware layer is implemented on an FPGA through the use of the VHDL programming language, and consists of the Manchester encoding/decoding and a design of a FIFO buffer. The software layer is implemented in the C++ language and consists of the framing and error correction schemes.
A fully functional encoder module hardware design that implements a Manchester encoder has been designed and tested. The design and implementation of the basic features of the encoder was successful. Advanced encoder and decoder features have been completed and tested, including framing/ de-framing. Work is currently focused on finishing advanced error correction features and overall integration, where significant advances have been made so far.
Artificial Transmission Line
Authors: Dilip Ramachandran
The RF world is one of the most studied topics in Electrical Engineering. This projects aims at designing RF hardware that may be used in a laboratory to analyze and understand the response of an artificial transmission line. Laboratory function generators max out at 15MHz and are unable to reach high frequencies (around 300MHz) where the electrical wavelength is short enough that it reaches a physically acceptable length. The device will show students that it is possible to fabricate a transmission line using hand selected capacitors and inductors. The electrical parameters of the line are representative of the behavior of the center conductor of a coaxial cable. Measurements can be taken at different points of the line and therefore students will have an understanding of the source and reflected signals present in the line, and also experience how they interact with each other at different points of the line. To demonstrate the properties of transmission lines, numerous precision mismatched loads will be used to study the artificial transmission line.
The goal of this project is to design a 50ohm, 1 MHz, 40 tap transmission line in which performance will be characterized as high as 100 MHz. In order to acquire accurate measurements, a digital counter circuit and a 40PST switching circuit will be designed. On the software side, LabView will be programmed to acquire these values and display them on a chart. LabView is used to perform additional mathematical calculations on the measurement data.
Programmable Photoresist Spinner
Authors: Ann Chempakaseril
Advisor: Van der Spiegel, Gruev
Photolithography is a critical process in the manufacturing of silicon integrated circuits (ICs). The spinner is an essential device for the deposition of a uniform and well-controlled photoresist layer on top of the silicon wafer. To attain uniform layer thickness, it is crucial to have a high angular speed, which leads to a high centrifugal force that acts on the resist fluid to move away from the center of the substrate. The Programmable Photoresist Spinner improves the precision control of the angular speed and acceleration. This component will aid in the development of discrete silicon devices such as the ICs for the microfabrication purposes.
The program prompts the user for inputs and the Motorola 68HC11 processes high-level commands and outputs signals to the motor control subsystem. The system is designed to perform the following functions:
- Program Recipes: User inputs key parameters such as speed, acceleration and spin times for a given load.
- Access Pre-programmed Recipes: Provides user with standard recipes for various substrates and corresponding thickness options.
View Poster in pdf, jpg
Generic USB 2.0 Custom Video Sensor Platform
Authors: Mark Dweck
Advisors: Van der Spiegel, Gruev
High Speed Universal Serial Bus (USB) 2.0 is a recently developed transfer protocol that is well suited for real time data acquisition from analog or digital sensors. Capable of transfer rates up to 480 Mb/s, USB 2.0 provides a data pipeline robust enough for most data acquisition applications, making it an ideal platform for sensor to PC interface. While USB sensor interface systems are available for many sensor types such as audio, economical data acquisition boards are not currently available for video.
Webcams and digital video cameras have successfully used USB 2.0 to interface to computers. However, existing USB cameras are not open systems and are composed of hardware and software designs tailored only for their single specific application. It is the goal of this project to create a USB 2.0 interface that can be generalized for various custom video sensors, providing a platform that can easily be adapted for multiple sensor technologies.
The design approach for the platform divides the implementation into three primary elements—camera, data acquisition microcontroller, and PC host. The camera serves as the source of the system data, providing the pixel information and synchronization signals necessary to produce video. This data is collected and packaged by an Orange Tree ZestSC1 FPGA microcontroller board, which is responsible for converting the raw digital data into a USB 2.0 data stream. The resulting stream is routed into the PC host, which is responsible for coordinating the system operation and displaying the data via a Graphical User Interface (GUI).
This successful development of a USB video interface allows various camera sensors to be attached. The data acquired from the camera sensors is relayed and properly displayed according to the camera type. This generic capability of the USB system is demonstrated via the implementation of two different grayscale custom video sensors, one of which provides frame limitation and focus capabilities. The result is a robust platform that leverages the utility of USB 2.0 to provide an efficient and accessible means of interfacing custom video sensors with personal computers.
e-Parking Meter Management System
Authors: Stephen Dabideen
Advisor: Guerin, Kassam
As the number of cars grow, it becomes increasingly difficult for the parking authorities of busy cities to effectively and efficiently monitor the parking resources within the city. This project presents a system to efficiently operate and monitor parking meters. The system assumes that there is a city-wide wireless network that can be used as a communication medium. It also assumes that each parking meter has sensing and communications capabilities.
The goal was to get information about each meter’s current status to some central office. At the central office, administrators can see the state of each parking meter and whether or not there are any violations. The main task was to design and implement a communication protocol, including packet formats and handling of errors and failures, to report monitoring/sensing information from the parking meters. Evaluation metrics include reliability, energy consumption at the meters, timeliness of transmission, and ability to recover from various failures.
The SAFE (Synchronized Adaptive-Forwarding Efficient) Routing Protocol was developed for this purpose. This is an on-demand routing protocol takes into account the highly variable nature of wireless networks and allows the customer to decide on the best trade-off between energy efficiency and reliability. The level of reliability was increased from about 20% data loss with single-path routing to 2% with SAFE’s probabilistic multi-path routing.
Although the SAFE Routing Protocol was designed for this particular application, it can be easily adapted for any multi-path wireless network where there is a trade-off between energy efficiency and reliability.
IntelliCam: An Intelligent Visual Tracking System
Authors: Zhan Chen
Advisor: Van der Spiegel, Gruev
Visual tracking systems currently exist which are able to track one moving object at a time. The IntelliCam extends this concept, implementing region-priority tracking and color-priority tracking, to increase the robustness of the motion tracking system.
The IntelliCam system is designed to track motion in an indoor environment. An algorithm is used to extract motion information from the webcam video input. This algorithm performs edge detection and temporal processing on each pixel of a video frame, in order to compute a velocity center of mass. The system can then compensate for moving objects according to this velocity center of mass.
After motion is detected, a command will be sent from the PC to the HC11 microcontroller via a RS-232 connection. The microcontroller then interprets the commands and generates a pulse-width modulated output to control the angle of the servomotor.
By default, the IntelliCam system tracks the velocity center of mass, thus following the largest moving object. The system controls the motor at a variable speed according to the degree of motion detected. Additionally, two extra control parameters are available to the user:
Region-Priority Tracking: Select a region of interest
Color-Priority Tracking: Select a color to be tracked
These options are controlled via a Windows Visual C++ program, which interfaces the various components of our system: Webcam, PC, microcontroller + servomotor. The goal of the system is to be able to process images at a rate of at least 15fps, and successfully operate in two test scenarios: surveillance of a living room, and recording a lecturer in a classroom.