SCHOOL OF ENGINEERING AND APPLIED SCIENCES
HARVARD UNIVERSITY

CS 266: Biologically-inspired Distributed and Multi-agent Systems Prof. Radhika Nagpal Fall 2007

Final Projects

Evolution Inspired Generation of Ant Foraging Algorithms, Nicolas Hoff, Amelia Sagoff, Stefan J. Wernli
Lessons from Regenerative Systems in Biology, Alex Shpunt and Seth Frey
Collective Construction: theoretical and evolutionary approaches, Andrei Munteanu, Thomas Carriero, Taro Narahara
Title: Dynamic Shape Formation, Zain Khalid and Vaidya Rajagopalan
ACO Routing in Wireless Sensor Networks, Jason Waterman and Atanu Chowdhury
Understanding Godsib, Rohan Murty
Lazy Calibration for Wireless Sensor Networks , Billy Lau and Michael Lyons

---

Evolution Inspired Generation of Ant Foraging Algorithms
Nicolas Hoff, Amelia Sagoff, Stefan J. Wernli

Ant foraging is a sufficiently interesting problem to explore multi-agent strategies, and an area where emergent group behavior can make drastic differences in how effectively and efficiently the ants of a colony gather food. We used this problem as an experimental space within which to test ideas about evolutionary algorithms and genetic programming strategies. To that end, we created a s-expression based decision tree language we call Baobab that represents a simple set of actions and tests that represent ant actions in their environment and comprise the genome we evolve across. Using the tree based evolutionary operators of cross-over, copying, and mutation drawn from the work of Koza, J. R. and an existing multithreaded java ant foraging simulator, we designed a system to test two kinds of evolutionary growth: one where every ant within a colony used the same decision tree and evolutionary operators are applied across a population of colonies, and another where each ant in a colony used its own tree and operators were only applied to individuals within the colony. (pdf)

Lessons from Regenerative Systems in Biology
Alex Shpunt and Seth Frey

Regenerative systems in Biology are analyzed and general principles of regeneration are articulated. The classification of all regenerative behaviors into epimorphosis and morphallaxis is found to be inaccurate and an alternative set of principles is proposed (following Agata-Saito-Nakajima 2007), "distalization" and "intercalation". To illustrate these principles, a simple canonical system (the Flag system) is exhibited, capable of a continuum of regenerative behaviors. The full implementation of the system listing in NetLogo is included and is available by request from the authors. Further work would include coupling between the intercalation and morphogenetic field self-regulation. For the latter, a framework of distributed consensus is proposed, that circumvents the absolute values of morphogens and their gradients as the drivers of development and regeneration. (pdf)

Collective Construction: theoretical and evolutionary approaches
Andrei Munteanu, Thomas Carriero, Taro Narahara

We studied the problem of collective building in a setting similar to the one studied in *Coordination in Distributed Building* by Theraulaz and Bonabeau. This project was split into two parts: 1) On the theoretical front we obtained results to support some of the claims which Theraulaz and Bonabeau made without proof. We restricted our study to construction along one dimension. We categorized building rules into three categories: immediately deterministic, eventually deterministic and nondeterministic. We showed that the family of strings generated by all three kinds of rules are regular languages. We then focused on immediately deterministic rules and proved a number of facts about them. 2)On the non-theoretical side of things, we first developed a standard Genetic Algorithm to try to learn a set of stimulating configurations that would generate coherent structures. We used a fitness function inspired by the notion of local checkability and a crossover function that switched stimulating configurations to try to arrive at a set of stimulating configurations that were good. Then, we also developed a GA-inspired algorithm (that was not like a typical GA) to try to generate more coherent structures more quickly. We managed not only to generate some interesting structures with this new algorithm, but we also found that it could be used as a framework for answering the question ``Do most sets of random rules generate coherent structures?'', which is a question Bonabeau asked in his paper and attempted to answer, but did not have any data to back it up. (pdf)

Title: Dynamic Shape Formation
Zain Khalid and Vaidya Rajagopalan

In this paper we are investigating dynamic shape formation, with the aim of realistically simulating dynamic maneuvers commonly exhibited in nature. These simulations were built as an extension of the existing flocking model provided by Reynolds, consisting of a number of simple identically programmed agents that respond to each other when within a certain radius. Our approach involves developing his model to support some form of arbitrary leader, who then leads the rest of the group of agents through the path described by the dynamic shape. Our approach is shown to be both robust and relatively stable, if not completely realistic. Using such simulations, we hope to get a better understanding of the mechanics of fish schooling and bird flocking in nature. (pdf)

ACO Routing in Wireless Sensor Networks
Jason Waterman and Atanu Chowdhury

The recent popularity of applications based on wireless sensor networks provides a strong motivation for pushing its technological limits. We feel that sensor networks can benefit from some of nature's robust solutions. In this work we evaluate the applicability of pheromone based foraging techniques, found in many ants, to route data packets in sensor networks. We find that such a solution is very robust and can easily recover from significant outages within the network. However high overheads do not allow it to replace traditional routing techniques in its current form. Our work is implemented in the ns2 simulator (pdf)

Understanding Godsib
Rohan Murty

This paper examines the study of gossip (godsib being the classical term) in human communities. In particular, we compare and contrast the study of gossip in the humanities (and social sciences) and compare it with the development of gossip techniques as communication primitives in computer science. Distributed computer systems designers have developed various gossip based techniques for efficient communication and state maintainence in large unstructured networks. On the other hand, academics in the humanities have examined how gossip occurs naturally in various human communities, its spread, and effects. Our goal in this paper is to survey the literature in the two broad fields and compare them to see if academics on either side can learn from the other. (pdf)

Lazy Calibration for Wireless Sensor Networks
Billy Lau and Michael Lyons

Sensor networks are designed to monitor an area but their sensors no- toriously suffer from poor accuracy. Although these sensors could be manually calibrated or replaced with precise alternatives, the time, man- power, and monetary costs may be prohibitive. Further, environmental changes may require regular re-calibration, adding to the expenses pre- viously mentioned. We propose a method for lazily calibrating mote sensors to reduce sensor network deployment costs. Our approach uses the entire sensor network to calibrate motes dynamically and does not require explicit routing schemes or maintained links. As a result, our approach does not depend on any specific nodes and packet collisions, wireless networking irregularities and mote failures will cause minimal impact in the sensor network. In this paper, we demonstrate up to 4x reductions in sensor error by using a computationally inexpensive al- gorithm for low power mote processors while requiring only a small fraction of pre-calibrated motes. (pdf)