collective perception in a robot swarm: biologically inspired
TRANSCRIPT
Beyond
robotics
Beyond
robotics
I-SWARM project507006
Collective perception in a robot swarm
Biologically inspired strategies for swarm robotics
Thomas Schmickl, Christoph Möslinger & Karl CrailsheimSAB, Rome, Italy, 2006
Beyond
robotics
Beyond
robotics
I-SWARM project507006
Hardware
• I-SWARM Robots
• JASMINE(www.swarmrobot.org)
Beyond
robotics
Beyond
robotics
I-SWARM project507006
The goal
• Robots should aggregate at a target area.
• When there are more targets, the swarm should split.
• If target sizes are of different size, the robot swarm should split proportionally to the target sizes.
Beyond
robotics
Beyond
robotics
I-SWARM project507006
Target 2 (small)
Target 1 (big)
Task:
Beyond
robotics
Beyond
robotics
I-SWARM project507006
The constraints
• Robots can communicate with horrizontalhorrizontal LED lights and phototransistors– Communication-radius = ~ 12-15cm
• Robots can detect targets onlyonly beneath theirbase plate– target-detection-radius = ~1.5cm
• We applied random error (10% / 15°) to communication and navigation.
Beyond
robotics
Beyond
robotics
I-SWARM project507006
The constraints
•• No global information!No global information!•• Purely autonomous robots!Purely autonomous robots!
Robots should collectively Robots should collectively generate a generate a map of the arenamap of the arenaand share this map.and share this map.
= = COLLECTIVE PERCEPTIONCOLLECTIVE PERCEPTION
Beyond
robotics
Beyond
robotics
I-SWARM project507006
Technics
• Communication with horizontal LED light-cones and photodiodes
• Also obstacle avoidance is made this way
• Motion: two wheelsdifference to
I-SWARM
Beyond
robotics
Beyond
robotics
I-SWARM project507006
We developed
• Two control strategies:1. Hop Count – based
(simple & straight-forward)2. Trophallaxis-derived strategy
(more complex & bio-inspired)1 trophallaxis = mouth-to-mouth feedings
performed by honeybees
Beyond
robotics
Beyond
robotics
I-SWARM project507006
(hc=15)(hc=8)
Target
Hop-Count strategy
I found thetarget!(hc=0)
(hc=1) (hc=2)(hc=3)
(hc=6)(hc=2)(hc=1)
(hc=0)(hc=3)
Beyond
robotics
Beyond
robotics
I-SWARM project507006
The idea: Trophallaxis
transfer of nectar
consumption of nectar consumption of nectar
Beyond
robotics
Beyond
robotics
I-SWARM project507006
bee hive
comb
comb
comb
The idea: Trophallaxis
gradie
nt
Beyond
robotics
Beyond
robotics
I-SWARM project507006
bee hive
comb
comb
comb
lousy weather !
The idea: Trophallaxis
gradient
Beyond
robotics
Beyond
robotics
I-SWARM project507006
V=0 V=0
Target
Trophallaxis strategy
I found thetarget!
V=1000
V=250 V=75V=10
I found thetarget!
V=1750
V=220I found thetarget!V=2550
V-- V-- V--
V++
V++
V→V→ V→
Beyond
robotics
Beyond
robotics
I-SWARM project507006
The trophallaxis-derivedstrategy at runtime
Beyond
robotics
Beyond
robotics
I-SWARM project507006
Recruiting proportionally to differently sized targets
random
Beyond
robotics
Beyond
robotics
I-SWARM project507006
Why?
Beyond
robotics
Beyond
robotics
I-SWARM project507006
Density of robots
0 0.1 0.2 0.30
50
100
150
200
250
density of robots
num
ber
ofag
greg
ated
robo
ts
robots leftrobots rightno aggregation
Beyond
robotics
Beyond
robotics
I-SWARM project507006
The proportionality can beadjusted by a threshold
-300 -200 -100 0 100 200 3000
50
100
150
200
threshold (thagg)
num
ber
ofag
greg
ated
robo
ts
robots leftrobots rightno aggregation
Beyond
robotics
Beyond
robotics
I-SWARM project507006
Dynamic environments
After 500 timesteps, we changed the environment to this:
Start:
final picture?consumption-
rate?
Beyond
robotics
Beyond
robotics
I-SWARM project507006
Dynamic environments
Beyond
robotics
Beyond
robotics
I-SWARM project507006
Conclusion
• The trophallaxis-derived strategy allows– Gradient navigation– Proportional recruitment– Works without any global information
• It is robust against random errors and perturbation.
• It is flexible, because out-dated information leaves the system over time.
Beyond
robotics
Beyond
robotics
I-SWARM project507006
Thank you!