Download presentation
Presentation is loading. Please wait.
Published byAnne Ellis Modified over 9 years ago
1
Controlling the Behavior of Swarm Systems Zachary Kurtz CMSC 601, 5/4/2011 1
2
Background Swarm systems are composed of many simple agents, each following a set of distributed rules or behaviors Swarm systems have a number of applications Swarm Robotics Particle Swarm Optimization The are several standard rule sets used in swarm systems Boid Model developed by Reynolds [1] Physics based models 2
3
Background: Example On the left is an example of a Boid swarm There are three rules controlling the swarm Cohesion pulls the members together Separation keeps the members from colliding Alignment keeps the velocities of the members similar 3
4
Background (Cont.) Creating more complex behaviors often requires custom rules For example creating a circular formation with a swarm requires specially designed rules: For each Agent A, select the farthest agent A’ If the distance(A, A’) > R, A moves toward A’ If the distance(A, A’) < R, A moves away from A’ 4
5
Problem Creating a desired swarm behavior requires hand- crafted rules It is often easier to evaluate how well a swarm is matching a behavior Solution: Develop an automated system to select a rule set, given an evaluation function 5
6
Related Work Finding optimal parameters for a rule set has be previously explored by Miner [2] Many groups have explored methods for creating various formations: Sugihara explored methods for forming circles, lines, and polygons with distributed rules [3] Spears and Spears created hexagonal and square lattices using distributed physics based rules [4] 6
7
Approach Have as an input, an evaluation function that determines how well the swarm is matching the desired behavior Start with a large set of basis rules A rule set can be created by assigning a weight to each basis rule If a large, represented set of basis rules is used, the optimal rule set should a subset of the basis rules 7
8
Approach (Cont.) A genetic algorithm can be applied to find the best subset of rules Start with a population of random rule sets Evaluate the fitness of each rule set by creating a swarm, and applying the given evaluation function Select members for the next generation from the old population weighted by fitness Mutate and crossover Repeat until the fitness converges (or some time limit has been reached) 8
9
Challenges May be computationally expensive to find the optimal set of rules The set of possible rule sets is limited by the basis rules General representation of more complex rules, such as rules that assign different types to the members of the swarm The evaluation function output shouldn’t need to be “fine-tuned” to work with the genetic algorithm 9
10
Evaluation Pick a set of basis rules from the literature Pick a set of behaviors with known rules sets from the literature Create evaluation functions for each of these behaviors Create a swarm from that evaluation function using the detailed approach Compare the performance of the created swarm to the swarm from the literature 10
11
Conclusion Introduced swarm systems Proposed a method for generating a set of rules to create an emergent behavior Discuss the feasibility of the approach and potential challenges 11
12
References [1] - C.W. Reynolds. Flocks, herds and schools: A distributed behavioral model. In Proceedings of the 14th annual conference on Computer graphics and interactive techniques, pages 25–34. ACM, 1987. [2] - Don Miner and Marie desJardins. Predicting and controlling system-level parameters of multi- agent systems. In AAAI Fall Symposium on Complex Adaptive Systems and the Threshold Effect, 2009. [3] - K. Sugihara and I. Suzuki. Distributed motion coordination of multiple mobile robots. In 5th IEEE International Symposium on Intelligent Control, pages 138–143. IEEE, 1990. [4] - W. Spears and D. Spears. Distributed physics based control of swarm vehicles. Autonomous Robots, 17(2):137–162, 2004. 12
13
Questions? 13
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.