Robot Walking with Genetic Algorithms Bente Reichardt 14. December - - PowerPoint PPT Presentation

▶
robot walking with genetic algorithms
SMART_READER_LITE
LIVE PREVIEW

Robot Walking with Genetic Algorithms Bente Reichardt 14. December - - PowerPoint PPT Presentation

Apr 13, 2023 35 likes •563 views

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation Robot Walking with Genetic Algorithms Bente Reichardt 14. December 2015 Bente Reichardt 1/52 Introduction Genetic Algorithms Quadruped Robot Hexaped

slide-1
SLIDE 1

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Robot Walking with Genetic Algorithms

Bente Reichardt

  • 14. December 2015

Bente Reichardt 1/52

slide-2
SLIDE 2

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Outline

Introduction Genetic algorithms Quadruped Robot Hexapod Robot Biped Robot Evaluation

Bente Reichardt 2/52

slide-3
SLIDE 3

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Why walking?

Mobile robots Similar to humans and animals Can move forward on different surfaces Can climb

Bente Reichardt 3/52

slide-4
SLIDE 4

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Why walking?

Mobile robots Similar to humans and animals Can move forward on different surfaces Can climb Problem: Walking is complex

Bente Reichardt 4/52

slide-5
SLIDE 5

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Evolutionary Algorithms

http://www.deviantart.com/art/Fish- Fishapod-Tetrapod-Tree-186821546

Biological inspired Survival of the fittest Reproduction and Mutation

Bente Reichardt 5/52

slide-6
SLIDE 6

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Genetic Algorithm

Population based

Bente Reichardt 6/52

slide-7
SLIDE 7

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Genetic Algorithm

Population based Solution as chromosomes

Bente Reichardt 7/52

slide-8
SLIDE 8

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Genetic Algorithm

Population based Solution as chromosomes Fitness function

Bente Reichardt 8/52

slide-9
SLIDE 9

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Genetic Algorithm

Population based Solution as chromosomes Fitness function Generating a new generation

Bente Reichardt 9/52

slide-10
SLIDE 10

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Example: Finding Orange

Bente Reichardt 10/52

slide-11
SLIDE 11

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Example: Finding Orange

Bente Reichardt 11/52

slide-12
SLIDE 12

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Example: Finding Orange

Bente Reichardt 12/52

slide-13
SLIDE 13

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Example: Finding Orange

Bente Reichardt 13/52

slide-14
SLIDE 14

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Quadruped Robot

http://www.roboticstoday.com/robots/aibo-ers-7

4 legs 2 DOF per leg

Bente Reichardt 14/52

slide-15
SLIDE 15

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Setup

Code for the walking exists

Bente Reichardt 15/52

slide-16
SLIDE 16

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Setup

Code for the walking exists 235 mm/s with human testing

Bente Reichardt 16/52

slide-17
SLIDE 17

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Setup

Code for the walking exists 235 mm/s with human testing 4 Robots to test parallel and autonomous

Bente Reichardt 17/52

slide-18
SLIDE 18

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Setup

Code for the walking exists 235 mm/s with human testing 4 Robots to test parallel and autonomous Some hours to learn

Bente Reichardt 18/52

slide-19
SLIDE 19

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Genetic Algorithm

Population: 30 Chromosomes: Parameters of the walking algorithm Fitness function: Distance moved forward

Bente Reichardt 19/52

slide-20
SLIDE 20

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Result

Distance measurement sometimes failed

Bente Reichardt 20/52

slide-21
SLIDE 21

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Result

Distance measurement sometimes failed Time was enough

Bente Reichardt 21/52

slide-22
SLIDE 22

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Result

Distance measurement sometimes failed Time was enough 290 mm/s with GA optimization

Bente Reichardt 22/52

slide-23
SLIDE 23

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Result

Distance measurement sometimes failed Time was enough 290 mm/s with GA optimization The robot got faster

Bente Reichardt 23/52

slide-24
SLIDE 24

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Hexapod Robot

http://robot-kingdom.com/hexapod-robot-tutorials- where-everything-begins/

6 legs 2 DOF per leg

Bente Reichardt 24/52

slide-25
SLIDE 25

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Setup

Evolution of a neural network for walking control

Bente Reichardt 25/52

slide-26
SLIDE 26

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Setup

Evolution of a neural network for walking control Separated in two steps

Bente Reichardt 26/52

slide-27
SLIDE 27

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Setup

Evolution of a neural network for walking control Separated in two steps Step one: move a leg

Bente Reichardt 27/52

slide-28
SLIDE 28

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Setup

Evolution of a neural network for walking control Separated in two steps Step one: move a leg Step two: walking with 6 legs

Bente Reichardt 28/52

slide-29
SLIDE 29

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Genetic Algorithm 1

Chromosomes: Neural Network for one leg

Bente Reichardt 29/52

slide-30
SLIDE 30

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Genetic Algorithm 1

Chromosomes: Neural Network for one leg Fitness function: Range of the oscillation

Bente Reichardt 30/52

slide-31
SLIDE 31

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Genetic Algorithm 1

Chromosomes: Neural Network for one leg Fitness function: Range of the oscillation Stops when oscillation reaches the full length

Bente Reichardt 31/52

slide-32
SLIDE 32

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Genetic Algorithm 2

Population: 10; 2 stay

Bente Reichardt 32/52

slide-33
SLIDE 33

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Genetic Algorithm 2

Population: 10; 2 stay Chromosomes: Neural network for all legs

Bente Reichardt 33/52

slide-34
SLIDE 34

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Genetic Algorithm 2

Population: 10; 2 stay Chromosomes: Neural network for all legs Fitness function: distance from starting point in the body axis

Bente Reichardt 34/52

slide-35
SLIDE 35

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Genetic Algorithm 2

Population: 10; 2 stay Chromosomes: Neural network for all legs Fitness function: distance from starting point in the body axis Crossover rate 0.1

Bente Reichardt 35/52

slide-36
SLIDE 36

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Genetic Algorithm 2

Population: 10; 2 stay Chromosomes: Neural network for all legs Fitness function: distance from starting point in the body axis Crossover rate 0.1 Mutation rate: 0.04

Bente Reichardt 36/52

slide-37
SLIDE 37

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Result

Learned to move their legs

Bente Reichardt 37/52

slide-38
SLIDE 38

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Result

Learned to move their legs Robots walked backwards

Bente Reichardt 38/52

slide-39
SLIDE 39

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Result

Learned to move their legs Robots walked backwards Similar to living insects

Bente Reichardt 39/52

slide-40
SLIDE 40

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Result

Learned to move their legs Robots walked backwards Similar to living insects Wave-gait vs tripod gait

Bente Reichardt 40/52

slide-41
SLIDE 41

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Biped Robot

http://www.engadget.com/2006/11/28/kondo-adds- pivot-to-khr-1hv-biped-robot-kit/

2 legs 6 DOF per leg

Bente Reichardt 41/52

slide-42
SLIDE 42

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Experiment

Truncated Fourier Series and polynomial equations used for motor control

Evaluating GA and PSO Evolutionary Algorithms for Humanoid Walk Pattern Planning Bente Reichardt 42/52

slide-43
SLIDE 43

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Experiment

Truncated Fourier Series and polynomial equations used for motor control Sinusoid equation the swing leg

Evaluating GA and PSO Evolutionary Algorithms for Humanoid Walk Pattern Planning Bente Reichardt 43/52

slide-44
SLIDE 44

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Experiment

Truncated Fourier Series and polynomial equations used for motor control Sinusoid equation the swing leg partly polynomial equations for the support leg

Evaluating GA and PSO Evolutionary Algorithms for Humanoid Walk Pattern Planning Bente Reichardt 44/52

slide-45
SLIDE 45

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Experiment

Truncated Fourier Series and polynomial equations used for motor control Sinusoid equation the swing leg partly polynomial equations for the support leg Learning of the equations for every motor

Evaluating GA and PSO Evolutionary Algorithms for Humanoid Walk Pattern Planning Bente Reichardt 45/52

slide-46
SLIDE 46

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Genetic Algorithm

Population: 20 Chromosomes: Coefficients of the equations Fitness function: Walking without falling

Evaluating GA and PSO Evolutionary Algorithms for Humanoid Walk Pattern Planning Bente Reichardt 46/52

slide-47
SLIDE 47

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Result

Coefficients are learned Each motor has specific equation robots learned to walk

Evaluating GA and PSO Evolutionary Algorithms for Humanoid Walk Pattern Planning Bente Reichardt 47/52

slide-48
SLIDE 48

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Evaluation

Optimization algorithm

Bente Reichardt 48/52

slide-49
SLIDE 49

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Evaluation

Optimization algorithm Finds a good solution

Bente Reichardt 49/52

slide-50
SLIDE 50

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Evaluation

Optimization algorithm Finds a good solution PSO (particle swarm optimization) usually faster

Bente Reichardt 50/52

slide-51
SLIDE 51

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

Thank you for your attention!

Bente Reichardt 51/52

slide-52
SLIDE 52

Introduction Genetic Algorithms Quadruped Robot Hexaped Robot Biped Robot Evaluation

References

Chernova, Sonia, and Manuela Veloso. ”An evolutionary approach to gait learning for four-legged robots.” Intelligent Robots and Systems, 2004.(IROS 2004). Proceedings. 2004 IEEE/RSJ International Conference on. Vol. 3. IEEE, 2004. Lewis, M. Anthony, Andrew H. Fagg, and Alan Solidum. ”Genetic programming approach to the construction of a neural network for control of a walking robot.” Robotics and Automation, 1992. Proceedings., 1992 IEEE International Conference on. IEEE, 1992. Azarkaman, Mostafa, Mohammad Aghaabbasloo, and Mostafa E.

  • Salehi. ”Evaluating GA and PSO evolutionary algorithms for

humanoid walk pattern planning.” Electrical Engineering (ICEE), 2014 22nd Iranian Conference on. IEEE, 2014.

Bente Reichardt 52/52