Evolution and RobotsHow to Create Artificial Brains for Machines
Who am I?
• Robotics & AI PhD Student @ ISCTE-IUL
• Geek, Hacker, all the good stuff
• Sci-fi, Tech and Robotics enthusiast
• Metalhead and Petrolhead
Overview
• Why Evolutionary Robotics (ER)?
• How Evolutionary Algorithms (EA) Work
• Challenges
• Cool projects
Archytas’ pigeon (~400 BC)
da Vinci’s Mechanical Knight
(1495)
Good Old Fashion Artificial
Intelligence (and Robotics)
Behavior Based
Robotics
Evolutionary Robotics
•Self-organized behavior
•Adaptable controllers
Evolutionary Algorithms
Evolutionary Algorithms
Fitness
Example: finding the best color for camouflage
Evolutionary Algorithms
Mutation (and/or Recombination)
Artificial Neural Networks
Neuron
Synapse
OutputInput
0.42.0
Neuron activation function
-1.2
Connection/Synaptic weight
-0.7
Sensors
Actuators
Inputs Outputs
Left Light Sensor
Right Light Sensor
Right Motor
Left Motor
Left Light Sensor
Right Light Sensor
Right Motor
Left Motor
Simulating Evolution in Robotics
• Robot Model
• Task
• Evaluation/Fitness Function
The Bootstrapping Problem
The Reality Gap
My Work
• (trying to) solve the Bootstrapping Problem and crossing the Reality Gap
• Hierarchical approach to the evolution of behaviors
Wrapping It Up
• Pros
• Self-organization of behavior
• ANNs are tolerant to noise
• Adaptable controllers
• Cons
• Currently only works “in the lab”
• Simple robots & simple behaviors
• Controllers might be... unpredictable
Questions/Discussion
@miguelduarte42