Optimization of swarm robotic inspection - Micro with different - - PowerPoint PPT Presentation

• Optimization of swarm robotic inspection

Micro with different learning schemes

Vincent Cattin & Nils Raning Course project presentation Swarm Intelligence EPFL Lausanne Switzerland February the 6 2006

The blade experiment

A swarm of robots inspect a regular environment to detect problems on blade.

Our microscopic model

Search Avoid robot Avoid wall Circle inspected blade Circle virgin blade pinspected(k) pvirgin(k) probot(N0-1) pwall Trobot Twall Tblade Serve as Beacon Twait Tblade (1/Thalf)pbeacon(k)

Our finite state machine:

Our algorithms:

• PSO
• GA
• In-line search

PSO

Particle swarm optimization

• Particles start at random position with a random speed
• They are attracted by the best place they have been and the

best place found by the swarm.

Genetic Algorithm

In-line search

Results

100 200 300 400 500 600 700 800 900 1000

GA for 5 robots

Tmax Tstart Twait Steps

• T-start > nbr of steps
• T-start not usefull with

few robots

• Good optimization with

all algorithm

Discution

R = # robots, B = # blades. 1 beacon = -1 robot to explore and +1 blade « beaconned » 1 beaconned blade = avg 50% less time spend on the blade => 50%/B less exploration needed 100%/R less explorer => to win we need R > 2B => we need LOTS of robots to need beacon might be usefull in not uniform spatial distribution of robots not supported by our model