Transcript 下載/瀏覽

Real-Time Obstacle Avoidance Method for Mobile
Robots Based on a Modified Particle Swarm
Optimization
Yuxin Zhao
College of Automation
Harbin Engineering University
Wei Zu
Institute of Automation
Chinese Academy of Sciences
2009 International Joint Conference on Computational Sciences and Optimization
指導教授:王啟州
學生:陳俊旭
Outline




Abstract
Introduction
Traditional PSO algorithm
Modified PSO algorithm



Simulations



PSO diversity measure and diversification method
The flow of modified PSO algorithm
Performance test of MPSO algorithm
Experiments of global path planning
Conclusion
Abstract

This paper. Based on the analysis of visual modeling,
the reason of premature convergence and diversity
loss in PSO is Explained.

Two parameters of particle-distribution-degree and
particle-dimension-distance.

Simulation results show that it has better ability of
finding global optimum, and still is more efficient.
Introduction(1/2)

Behavior of organisms such as bird flocking.
Empirical evidence has been accumulated to
be a very effective.

Many researches have indicated that the PSO
often converges significantly faster.
Introduction(2/2)

Visually modeling method of individual
particle for the purpose of exploring the
behavior of the individual particle movement
in the search space.
Traditional PSO algorithm

Each particle adjusts its trajectory toward its
own pBest and the gBest attained by the
whole swarm.
Modified PSO algorithm(1/3)

Adopt a parameter particle-dimension–
distance to measure the distance between
different particles.
Modified PSO algorithm(2/3)

If particles distribute equally in problem scope, the
value dis(s) will be zero.

If particles dimension cluster in the same
separationarea, dis(s) will satisfy Eq. (4).
Modified PSO algorithm(3/3)


The first one can measure the distances
between particles and the proposed parameter
can measure the swarm clustering degree.
The second way is that the particle real-time
update velocity is not adopted as a whole
individual.
Simulations(1/5)

We have used four functions to test the
modified PSO.
Simulations(2/5)

The generations was set to 10000 for
Rosenrock and Rastrigrin problems, while
2000 for Sphere and Griewank problem. C1
and C2 were set to 2.0. The size of particle
population is 20.
Simulations(3/5)
Simulations(4/5)

Design an environment including nine
rectangle obstacle and test performances of
three methods to solve path planning problem.
Simulations(5/5)
Conclusion

In this paper, a novel particle modeling
method is presented which can illuminate the
reason of Premature convergence in the
optimization process.
Thanks for your attention!