Download Evolved Flocking

yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Ethology wikipedia , lookup

Cooperative hunting wikipedia , lookup

Dietary biology of the Eurasian eagle-owl wikipedia , lookup

Seismic communication wikipedia , lookup

Robotics wikipedia , lookup

Deception in animals wikipedia , lookup

Animal communication wikipedia , lookup

Ambush predator wikipedia , lookup

Anti-predator adaptation wikipedia , lookup

Evolving Flocking
Simulation and Robotics
Dan Sayers
Flocking in Nature
• Flocking and other group behaviours in animals and humans
have natural beauty and have long been a source of
• Group behaviour in animals gives rise to the phenomena of
“swarm cognition” or “group intelligence” in which whole
groups respond and act for the collective interest of the group
• Flocking in animals is usually observed in prey species –
studies have demonstrated a link between predation and
flocking behaviours
Flocking Behaviours
Flocking in Simulation
• Flocking and other group behaviours have been studied in
simulation since the Eighties, using simple rules for individual
flock members, from which flocking emerges as a group
• This furthers our understanding of group behaviours in
various species (including humans)
• Also, it given us a method to interact with the beauty of
flocking in artistic ways
Simulated Flocking
Craig Reynolds – 3 rules:
• Collision avoidance (separation)
• Velocity matching (alignment)
• Flock centering (cohesion)
These three (locally applicable) rules are observed to be
sufficient for realistic flocking in 2D and 3D simulations
Each rule influences a steering force on each agent. These agents
were named ‘Boids’ by Reynolds
Evolved Flocking Under Predation
In my simulation, flocking is shown to evolve when under the
influence of predators. The evolution takes place through a
gradual improvement of the connections between sensors
and steering forces.
The control system for the prey members is very simple – a
vector sum of the sensor vectors gives the steering force
F = c1S1 + c2S2 + …
The amounts (c1, c2 etc.) that the different sensors affect the
steering force is what is evolved. Prey and predator
populations are kept constant.
Flocking “Sensors”
Each flock member is capable of
responding to fellow
members within a defined
radius, and within a defined
vision angle
Prey members also respond to
predators within their field of
The prey have sensors for:
• Average nearby prey position (close)
• Average nearby prey position (far)
• Average nearby prey velocity
• Average nearby obstacle location
• Average nearby predator location
These vectors are added together (multiplied by genetically
determined weights) to give the prey’s steering force at any
given time
From Simulation to Robotics
This simulation contains a number of idealisations:
• The sensors are ideal in that the information about
neighbouring vehicles and obstacles is communicated directly
rather than being sensed by real-world, “noisy” sensors
• The steering vector is also idealised
In the Real World …
• Robots need to able to survive collisions
• Real, noisy sensors to deal with
Consideration of how to sense nearby boids:
• Vision
• Ultrasound
• Sonar / Radar
• Wifi / Broadband
Existing Robots for Flocking
Festo Penguins
Existing Robots for Flocking
Khepera Robots
Existing Swarmbots
Formica mini swarm bots
• Cost as low as £15 each to make
• IR communications / sensors
• LED ‘mood’ indicators
Web Links
Craig Reynolds’ Boids
My evolved flocking pages
Khepera robots
Formica PCB robots