Information distribution between mobile agents

 

In this project, you will implement and test communication strategies for ad-hoc communication between mobile agents. The idea is to evaluate which strategies are suitable to distribute information between different agents. Rather than just flooding the network with information, you will evaluate how much communication is needed between agents in order to sustain certain information in a network of agents. Different agents will join and leave the simulation at intervals, but information should never get lost. So it is important that agents communicate enough to spread information and to keep it alive but at the same time limit communication as much as possible, to preserve resources. 

 

For these simulations you will use a multi-agent simulator such as MASON. Since the simulator is in Java you will also be required to develop your project in the Java language. We will provide you with datasets of realistic movement patterns for agents in different scenarios. An example of one such dataset is the from the UCI Network Data Repository. It contains information about people attending a conference. The movement of all conference attendees was tracked using RFID tags, and this data can be used as a realistic model of mobile agents. 

 

The project will require some implementation work in Java, but you will spend a lot of your time performing simulations. You will compare different communication strategies and the impact of different parameters. For your experimental results to be valid it is important that you setup a good experiment.

 

Project goals:

  • Implement multiple communication strategies (specified by supervisors) in MASON
  • Perform experiments on different datasets
  • Evaluate impact of parameters on the data dissemination within the network of agents

 

Depending on your interests we can further elaborate our goals, we could look at more complex algorithms or analyze the impact of differences between different datasets.

Contact

Maarten Devillé

Peter Vrancx