8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

1/16

A Distributed Neural Controller for Locomotionin Linear Modular Robotic Configurations

Avinash Ranganath, Juan Gonzlez-Gmez,Luis Moreno Lorente

Robotics LabUniversidad Carlos III de Madrid

30/May/2011

9 Workshop - Robots colabativos e interaccinHumano-Robot - Robocity 2030-II

CAR, E.T.S. de Ingenieros Industriales, Madrid.

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

2/16

2

Index

A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

1. Introduction

2. Robotic platform

3. Proposed model4. Evolution

5. System 1

6. System 2

7. Fault tolerance

8. Conclusion

Index

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

3/16

3A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

1. Introduction

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

4/16

4

Sinusoidal Oscillator

A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

1. Introduction

it = A i sin

2

T t

iO

i

Sinusoidal Oscillator:

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

5/16

5

Objective

A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

1. Distributed

2. Homogeneous

3. Adaptive

4. Fault tolerant

1. Introduction

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

6/16

6

Y1 in OpenRAVE

A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

2. Robotic platform

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

7/16

7

Neural Controller

A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

3. Proposed model

Input layer : 1-2: Connector information 3: Feedback from self

actuator 4-5: Feedback from

neighbouring module'sactuator

Hidden layer : One hidden layer with five

hidden neurons Output layer :

One output neuronconnected to the actuator

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

8/16

8

Evolution

A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

4. Evolution

Genetic Algorithm Roulette Wheel selection method Intermediate Recombination method Fitness: Displacement measured as euclidean distance from origin Evaluation time: 50 seconds.

Parameters Values

Population size 50

Generations 100

Crossover percentage 50%

Elite population retained 12.5%

Mutation rate 1/Size of genome

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

9/16

9

Frequency controlled method

A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

5. System 1

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

10/16

10

Frequency controlled method

A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

5. System 1

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

11/16

11

Frequency adaptive method

A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

6. System 2

Where,

: Output from the ANN

: Angle of the actuator attime ti

and x are constants.

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

12/16

12A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

6. System 1

Frequency adaptive method

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

13/16

13A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

6. System 1

Frequency adaptive method

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

14/16

14A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

7. Fault tolerance

Limit cycle behaviour

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

15/16

15

Conclusion

8. Conclusion

Proposed model validated for locomotory oscillations. Fault tolerance with limit cycle behaviour. Fast convergence to optimal oscillatory pattern.

Future work

Validate this model on the real Y1 modular robots. Adaptive behaviour for locomotion on different surfaces

Evolve the topology to both complexify and to find the minimal neuralarchitecture.

A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

8/6/2019 Slides: A Distributed Neural Controller for Locomotion in Linear Modular Robotic Configurations

16/16

A Distributed Neural Controller for Locomotionin Linear Modular Robotic Configurations

Avinash Ranganath, Juan Gonzlez-Gmez,Luis Moreno Lorente

Robotics LabUniversidad Carlos III de Madrid

30/May/2011

9 Workshop - Robots colabativos e interaccinHumano-Robot - Robocity 2030-II

CAR, E.T.S. de Ingenieros Industriales, Madrid.