Solving Sudoku puzzles with the Cougaar agent architecture

Abstract

The Cougaar distributed agent architecture, originally a DARPA-funded research project, provides a platform for developing distributed agent systems. Testing its ability to solve complex problems using a large number of agents in an interesting topic of research. In this thesis, the Cougaar distributed agent architecture is studied from the standpoint of Sudoku. Through analysis and experimentation, insight is gained into both the properties and weaknesses of Cougaar. Cougaar's performance when solving Sudoku puzzles is then compared with other Sudoku solving techniques. The Cougaar agent approach solves Sudoku puzzles in a human-like fashion, reaching solutions using both analysis and guessing. Cougaar is shown to be capable of solving Sudoku puzzles in a distributed agent architecture. Although not as fast as traditional techniques, the Cougaar distributed agent approach is able to provide solutions to Sudoku puzzles using fewer guesses. Additionally, solving Sudoku puzzles with Cougaar exposed some reliability issues and demonstrated the overhead required for communication. These results open the way for additional study of Cougaar.