The GOLEM Project Automatic Fabrication Results Marcin Pilat - - PDF document

The GOLEM Project

Marcin Pilat

Contents

• Evolutionary Robotics
• Project Introduction
• Evolutionary Design
• Simulator
• Automatic Fabrication
• Results
• Further Research
• Conclusions

Evolutionary Robotics

• Robotics:
• Design/manufacture/programming of machines
• Standard Robotics:
• Human design
• Human programming
• Evolutionary Robotics:
• Full autonomy in design
• Self-programming
• Self-manufactured

GOLEM Project

• Name: Genetically Organized Lifelike Electro

Mechanics (GOLEM)

• Creators: Hod Lipson (Cornell), Jordan B.

Pollack (Brandeis)

• Goal: automatic design and manufacture of

robotic lifeforms

• Results: First instance of automatic design and

manufacture of robots

Evolutionary Design: Intro

• Building Blocks:
• Bars (with ball joints); Linear actuators
• Artificial neurons

Evolutionary Design: Structure

• Morphology (Body):
• Arbitrary rigid, flexible, and articulated structures
• Revolving, linear, and planar joints
• Control (Brain):
• Arbitrary network of

sigmoidal neurons

• Feed-forward nets,

recurrent nets, state machines, and multiple independent controllers

Evolutionary Design: Evolution

• Based on ES and EP
• Generations: 300-600
• Population: 200 individuals (steady-state)
• Initially null individuals (no bars/actuators/neurons)
• Selection: Fitness-proportionate
• Replacement: random
• Genetic Operators: mutations
• Fitness: net Euclidean distance moved over

12 cycles of neural control

Evolutionary Design: Mutations

• Mutation types (probabilities):
• removal/addition of bar (0.01)
• removal/addition of unconnected neuron (0.01)
• change in length of bar (0.1)
• change in neuron synaptic weight (0.1)
• split vertex with separating bar (0.03)
• split bar by a vertex (0.03)
• attach/detach neuron to bar (0.03)
• At least one mutation applied

Simulator

• Mechanics and neural control simulated

concurrently

• Mechanics:
• quasi-static motion (each frame statically stable)
• relaxation (minimize energy term)
• static friction and noise
• Neural Net:
• Discrete cycles
• Actuator length in increments 1cm

Automatic Fabrication: Printing

• Rapid-prototyping technology (“3D Printing”)
• Point model converted to solid model
• Built with layered thermoplastic material:
• Complex joints:
• Can be melted and recycled

Automatic Fabrication: Assembly

• Evolved neural controller

downloaded into PIC microcontroller:

• Linear actuators connected

in proper places

• Robot is ready to be used:

Results: Overview

• Different and elaborate solutions evolved
• Typically around 20 building blocks
• Some developed symmetry
• Typically 10s of generations passed before

initial movement

• Typical evolution:

Results: Evolutionary Patterns

extreme divergence intermediate divergence extreme convergence massive extinction

Results: Population

• Snake

family:

Results: Fitness

• Plots of individual fitness over generations:
• 1) progress in noticeable jumps
• 2) progress rate slows down as fitness

improves

Results: Robot Examples 1

• Arrow:

Results: Robot Examples 2

• Tetra:

Results: Robot Examples 3

• Ratchet:

Results: Robot Examples 4

• Snake:
• Biped:

Results: Robot Examples 5

• Crab:
• Balance:

Results: Robot Examples 6

• Complex1:
• Complex2:

Further Research

• Study very large populations with more CPU

power [Golem@home]

• Inclusion of more sophisticated training

environment, sensors, robot interactions

• Discover and reuse modules
• How can more complex modular structures

self-organize?

Conclusions

• Demonstrated:
• automatic manufacturing of automatically designed

electromechanical systems

• with minimum human interaction

GOLEM

• Thank You
• All images and videos from GOLEM website at:

http://demo.cs.brandeis.edu/golem/