Multi-robot Specialization: Build a Java-based Soccer Team


Sponsor Tucker Balch
tucker@cc.gatech.edu
216 MARC
Area Intelligent Systems

Problem
Multi-robot cooperation is a new and expanding field in robotics research. Open questions include issues like the benefits of communication, cooperation and learning, and whether team members should specialize or be homogeneous. In this project, you will explore the last question, and also contribute to ongoing research.

Soccer (Football outside the U.S.) is a good task for multi-robot investigations. First, it involves cooperation and competition. Second, compared to many robot tasks, performance is easy to measure. One research project in the Mobile Robot Laboratory at Georgia Tech is investigating multi-robot issues using a Java-based simulation called JavaSoccer.

Your task for this project is to develop two multi-robot soccer teams for the JavaSoccer simulation: one in which all the robots' behaviors are identical, and another in which at least one robot plays a specialized role (like a goalie for instance). You should develop the homogeneous team first, then implement the specialized team as a modification of the homogeneous one. If the teams are substantially similar except for the specialization, we can make stronger statements about the performance advantage or disadvanage of specialization.

After you've implemented the teams, compare their performance against each other and against one of the benchmark teams included in the JavaSoccer package. Is specialization a good thing?

Your work on this project will also contribute directly to ongoing multi-robot learning research. Here's how: In many robotics tasks the performance of a system is compared against a known optimal solution, but in complex robot tasks like this one it is difficult or impossible to develop provably optimal solutions. Still, we would like to make quantitative assessments of our experimental system. One approach is to compare performance of experimental teams against human-coded teams. We'll use the teams you develop as adversaries for our learning soccer teams.

Background

Deliverables
Evaluation
Evaluation is based on the quality of your report and how well your team performs against the SchemaDemo team. Since the SchemaDemo team was coded in only 20 minutes, you should have an easy time defeating it.


updated by tucker, 9/7/97, 5:45pm.