Department |
People |
Curriculum |
Research |
Resources |
Faculty Current Students Alumnae |
|
|
||||||||||
|
|||||||||||
|
Course Navigation
Tim Ambroggi
Larry BombackMy final project will be a research paper and presentation on the RoboCup Initiative. I will cover the history, the current status, and future goals of this initiative. The ultimate goal of the RoboCup initiative is as follows: "By the mid-twenty-first century, a team of fully autonomous humanoid robot soccer players shall win a soccer game, complying with the official rules of FIFA, against the winner of the most recent World Cup." AI researchers see it as the next major step in the field, following Deep Blue's victory over Gary Kasparov, and the success of the Mars Pathfinder mission. If you want to see some videos of two of the many different robot soccer leagues you can click the following URL: www.cs.kuleuven.ac.be/~nico/demo/pages/rcvideo.html
Jackie Chew
Catherine ChiuMy final project is going to be a palm pilot controlled robot. I am hoping that it will be able to use iconic memory to map out the room and follow a path designated buy a path drawn on the palm with the stylus.The chassis that we have for the robot has three infra-red sensors and the palm just docs into the the chassis. The software that is used to compile the program is CodeWarrior so that won't be fun, but it will be interesting.
Jason ColemanI'm now going to add the learning algorithm I talked about once in my weekly reaction to my Konane game. Basically, the program will remember every move played against it, and if it loses, it will consider whoever beat it to be a good player and consider those moves while playing future games. I'll leave my old project idea below... I want to make a competitive robot game similar to lasertag, but using bumpers instead. Each robot will have two target bumpers, one on the left and right. When one target has been hit, the robot will be limited to using one motor at a time--simulating damage. When both targets are hit, the robot will be immobolized, and out of the competition. Each robot will have a light in the front, so other robots can find them... with the added complexity that the robot will have the option to turn off that light for ten second intervals, but only once every minute. To make things interesting, I'm going to try to program a simlutor for the robots on the pc, and use the simulator with some genetic algorithms to evolve playing strategies. I'll design some strategies myself, and evolve others, and pit the robots against each other in royal rumble style matches (with more than two robots competing at once).
Ian HarrisonI've incorporated a genetic algorithm that evolves a statc evaluation function into the konane game. The ga doesn't actually evolve the entire evaluation function, it evolves constants used to modify a previously set function, essentially determining the weighting each component of the function is given in determining the value of any board configuration. Currently, the ga can be evolved in either of two ways. An entire population can be forced to play round robin games amongst itself (using a board size of 6 and a search depth of 4 to keep the process from taking forever), with fitness being the number of times one gene beats any other gene, or with the population pitted against a standarized player (one that never changes through the generations), with fitness being the number of times the gene beats the standarized player. Eventually, I will combine the two approaches so that two populations co-evolve, alternately playing against each other and playing against one of the original two options.
Nick Kerr
Ananya MisraMy plan is to extend the Konane project into a generic game engine that plays any two-agent zero-sum perfect information game, given the rules of the game. I could illustrate it with a few other example board games, like connect-4, tic-tac-toe, or checkers (but not chess). Actually, it would be nice to have an interface that let the user define rules for her own game and then play it (only to lose), but that is not part of my final project!
Jamie RacanelliNick and I want to program two robots to play capture the flag. We will have to sides (one plain and one covered with dark construction paper) two a long enclosed rectangle, and two robots will try to get to the other's flag without getting hit while in enemy territory. We're not sure yet wheter the robots will have to turn around and get to there own side once they have found the flag (whatever that flag will be) or if they just have to get to the end of the enemy's territory. It will be mostly an exercise in subsumption architecture, but we will make it really cool.
Juan RamosMy final project will consist in creating a semantic network with the program SNePS that can interpret information about a given topic, possibly the animal kingdom, and develop a question-answer system based on the information that the network has been given. For example, I could tell the system that "Tweety is a bird", then "Birds can fly". The I could ask the system "Can Tweety fly?", and the network should respond "yes, Tweety can fly because Tweety is a bird and birds fly" (or something to this extent). However, the system would not be able to answer "Does Tweety eat seeds?" because the network has no knowledge about seeds. This project is the first step in my thesis project. For this class, I will focus on building the question-answer system, where the input and output is in the syntax for SNePS. The final product should be able to interact with the user using natural language. To extend it further, I plan to extend the natural language interface so that the system can interpret both English and Spanish, and answer each question in the desired language.
Matt RushtonFor my final project I wish to create a genetic algorithm that predicts the outcomes of professional football games. I would use the following statistics: Points Per Game, Points Allowed Per Game, Total Yards, Total Yards Allowed, Turnover Ration, Win Percentage, Home/Away Game, and Opponent Win Percentage. I would store these stats in structs or classes for each of the 32 NFL teams. I would then find league averages and compare how many standard deviations each stat is from the average. I would then use weights to figure out the importance of each stat. These weights would be what the genetic algorithm evolved. For example, would Points Per Game be a more important stat than whether the game was home or away. By using the weights a numerical value could be generated as a sort of index to score each team. These scores could be compared to one another when each team plays and the team with the higher score would be predicted to win. Obviously weights with a higher winning percentage would be deemed more fit and they would survive into the next generation. I think I could evolve a fairly accurate predictor, one that even outperforms most of the experts on espn.com.
Tina TanViewWiki | EditWiki | Webmaster@wiki.cs |
Course Home Page