2015 Project Final

Overview

In this project you should implement a multi agent system, then assess it experimentally. You will first submit a proposal, which, once approved you can start work on.

The Proposal

Create a one to two page proposal that contains the following sections:

1. Overview: Big picture of what you're planning to do.
2. Research Question: What question or hypothesis are you going to explore? Try to make this clear and quantitative. Example: "Does schooling increase prey survival rates?"
3. Implementation Approach: Which algorithms or methods are you going to use? Which simulation environment?
4. Quantitative Assessment: How will you measure your results?
5. Milestones: List 3 milestones for your project that you will have to meet in order to get done on time. First cut of behavior implemented in single agent; Multiagent behavior working; Project Complete. Note that each of these should include an increment of work, so if you've already completed some of it previously, you need to do NEW work.
6. Fallback: What if you can't build the full system you want to, what's the fallback?
7. Citations: Citations to the previous work relevant to your proposal.

Milestone Dates

• Proposal Due: Thursday 12 March
• Milestone 1: Thursday 2 April (3 weeks)
• Milestone 2: Thursday 16 April (2 weeks)
• Final: 1 May (2 weeks)

Final Project Ideas

You are not limited to these ideas. These are just potential starting points.

• Implement a real animal behavior and compare its performance to the real animal model (examples: porpoise herding of schooling fish in shallow water, honey bee communication)
• Implement an adversarial game in 3d using PyBioSim. Perhaps 3d soccer or quid ditch.
• Implement 2d or 3d herding. First implement the sheep, then program your sheepdogs. Investigate different herding strategies and measure their performance.
• Is schooling (or herding) an effective behavior for prey fish (land animals)? You can explore this by programming a fixed set of predator behaviors, then program a baseline prey behavior, then vary a parameter that turns up or down the tendency to school. Measure how many prey are killed over a fixed time period.
• Use a machine learning approach (RL or GAs) to evolve a group behavior such as foraging, predator behavior or prey behavior.
• Implement ant navigation using pheromones.
• Does how does prey behavior evolve as a consequence of predator behavior? When does schooling evolve or not? In this project, program several different predator behaviors that you believe would be countered by different pre behaviors, then create a method to evolve prey behavior (e.g., GAs). Compare and contrast the prey behaviors that result.
• Implement a 2d or 3d modular self assembly algorithm. Can you improve it?