Reconfigurable CPGs for the Implementation of Adaptive ...jhbarronz...Reconﬁgurable CPGs for an Adaptive Locomotion Problem Description Problem Description Legged locomotion concerns

Reconfigurable CPGs for an Adaptive Locomotion

Reconfigurable CPGs for the Implementation of

Adaptive Rhythmic Patterns of Locomotion.

José Hugo Barrón Zambrano.

Advisor: César Torres Huitzil.September, 2013.

José Hugo Barrón Zambrano. Reconfigurable CPGs for an Adaptive Locomotion 1


Outline

1 Motivation

2 Problem Description

3 Objectives and Hypothesis

4 Related Work

5 Methodology

6 Conclusions



Motivation

Motivation

A mobile robot depends on a locomotion mechanisms thatenable it to move throughout its environment.

Robot locomotion must perform in a controlled and reliablemanner through an unknown environment without the aid of ahuman operator.

To achieve an autonomous robot locomotion there are a largevariety of possible mechanisms to move.

Experiments and research work have addressed the feasibilityof the design of locomotion control systems based onlocomotion mechanisms present in humans and animals .



Problem Description

Problem Description

Legged locomotion concerns in how to determine the bestsequence, locomotion pattern, for lifting off and placing thefeet.

Legged locomotion requires multi-dimensional coordinatedrhythmic patterns that need to be correctly tuned to satisfymultiple constraints.



Problem Description

Locomotion Approaches

Approach Mathematical model-

based

Biologically inspired ap-

proach

Main featu-

re

Joint angles are calculatedin advance

Central Pattern Generator

Design met-

hodology

Trial-and-error or recordingdata

Trial-and-error

Robot

knowledge

Exact robot model It does not need a robotmodel

Perturbation

problems

Require an additional mo-dule

Robust against perturba-tions

Control

scheme

Centralized Distributed



Problem Description

Central Pattern Generator (CPG)

The CPG is a neural circuit capable of producing coordinatedpatterns of rhythmic activity in open loop.

CPG can adapt to various environments by changing theperiodic rhythmic patterns through simple input signals.

The biological mechanisms underlying locomotion havetherefore been extensively studied by neurobiologists.



Problem Description

CPG model

CPGs are often modeled as nonlinearoscillators through second order differentialequations that have mutually coupledexcitatory and inhibitory connections.

Oscillator model

xi = F (xi , xi , pxi , xai , Sfeed )(1)

Coupling contribution

xai = xi +Σwij ∗ xj (2)

For i =1, 2, 3, . . . , n , where xi is the output signal from oscillator i , n isthe number of oscillators in the CPG, pxi = [p0, p1, ..., pn] is the oscillatorparameters, Sfeed is the feedback signal,and wij is the coupling weight.



Problem Description

CPG: Open Topics



Problem Description

CPG Design

To build a CPG able togenerate different basiclocomotion patterns.

Oscillator tuning.CPG structure.Transition between differentpatterns.

Given an oscillatorxi = F (xi , xi , pxi , xai), to find the valuesof vector pxi , feedback signal Sfeed , andthe coupling weights wij to generate aspecific locomotion pattern.



Problem Description

Environment Feedback Integration

CPGs need environmentinformation to modulate andreact according to a sensedsituation.

The integration of feedbackinformation in the CPGmodulation process is notdirect.

To find the function, F , that mapsinformation sensors , Si , tolocomotion control parameters, Lc .

SiF→ Lc . (3)José Hugo Barrón Zambrano. Reconfigurable CPGs for an Adaptive Locomotion 10


Problem Description

Hardware Implementation

General purpose processor

Providing high accuracy andflexibility.Good performance onaverage.Tasks share the processortime.Real-time response cannot beguaranteed.

Analog implementation:

Computation and powerefficient.Lack flexibility to be reused.Large design cycles.



Objectives and Hypothesis

Objectives

Main Objective

To design and implement an adaptive locomotion control, based onCPG principles and visual information integration, embedded in ahardware platform capable of reacting under real-time constraints.




Objectives

Specific Objectives

To design a CPG-based locomotion able to generate multiplepatterns of locomotion for quadruped and hexapod robots andthe transitions between them.

To propose and implement an integration mechanism tomodulate the locomotion pattern generation based on visualinformation.

To design and implement efficiently a locomotion controlbased on biological organization embedded in a hardwareplatform under real-time constraints.




Hypothesis

Hypothesis

It is possible to design a locomotion control scheme based on CPGprinciples able to generate adaptable locomotion patterns throughintegration of visual feedback information embedded in a customparallel hardware platform under real-time constraints.




Contributions

Contributions

CPG able to generate three different locomotion patterns

for quadruped and hexapod morphologies.

CPG capable of switching between gaits, and controllingthe locomotion speed.

Scalable hardware architecture of CPG-based locomotion

for a quadruped and a hexapod morphologies, so as thephysical implementation in a hexapod robot.

Integration mechanism based on fuzzy logic and finite statemachine to modulate the CPG-base locomotion usingvisual perception information.



Related Work

CPG-Based Locomotion: Quadruped Robots

Work Imple-mentation

Feedback IntegrationMechanism

Behavior

Billard andIjspeert(2000)

Software Internal (legpositions)

Linear equa-tion

Three gaits

Fukuoka etal. (2005)

Software external (ob-ject distances)

Linear equa-tion

Walk and con-trol direction

Feng andWang(2008)

Software Internal (legpositions)

PD-controller Walk and con-trol direction

Santosand Matos(2012)

Software External (ob-jects and dis-tances)

Mathematicalmodel

Pursuit andavoiding

Nakada etal. (2003)

Hardware(CMOS)

None None Three gaits

Nakada etal. (2005)

Hardware(VLSI)

None P-controller Three gaits



Related Work

CPG-Based Locomotion: Hexapod Robots

Work Imple-mentation

Feedback IntegrationMechanism

Behavior

Inagaki etal. (2006)

Software(parallel)

None Hamiltonianfunction

Three differentgaits with ve-locity control

Arena etal. (2004)

Software Internal(legspeeds)

Neural Net-worsk (SOM)

Three differentgaits

Arena etal. (2005)

Hardware(VLSI)

Internal (legspeeds andhorizontalpositions)

Neural net-works (CNN)and PID-controllers

Climbing gait

Arena etal. (2006)

Software(microcon-troller)

None Neural net-works (CNN)

Climbing gaitand directioncontrol

Manoonponget al.(2008)

Software External(Obstacledistance)

Neural net-works (FFNN)

Avoiding obs-tacles



Methodology

Methodology

It was divided into three phases:

CPG-based locomotion.

Visual perception and integration mechanism.

Embedded hardware implementation.



Methodology

Phase 1: CPG-Based Locomotion


Oscillator modelselection.

Oscillator network(CPG) design.

CPG simulations.

Result analysis.



Methodology


Phase 2: Visual Perception and Integration mechanism

Visual perception design.

Integration mechanism design.

Locomotion control assembly.

Locomotion control tests.

Result analysis.



Methodology


Phase 3: Embedded Hardware Implementation

Hardware module designs.

Oscillator model.CPG-based locomotion.Integration mechanism.

Hardware module simulations.

Physical implementation.



Methodology


Timeline of activities

year 2010 2011 2012 2013Activity Q1 Q2 Q3 Q1 Q2 Q3 Q1 Q2 Q3 Q1 Q2 Q3Selection, implementation,testing of oscillator model.

X X X

Design, implementationand testing of CPG (1).

X X X

Review of visual perception(PV) models

X X X

Implementation of PV mo-del (2)

X X X X

Design and implementa-tion of integration mecha-nism (3).

X X X X X

Design and hardware im-plementation of the modu-les.

1 1 1 3 3-3

2-3

2-3

Module integration in therobot platform based onFPGA.

X X X

Testing and result analysis. X X XCourses. SMR IDP RCPredoctoral and thesis dis-sertation.

P T

Publications. C J C C CH J/C J J



Conclusions

Conclusions

CPG is neural circuit capable to generate locomotion patternsfeasible to be used in robot locomotion.

CPGs are able to generate complex and adaptive locomotionbehaviors from manipulation of simple periodic signals.

The presented CPG-based controller will provide flexibility togenerate different rhythmic patterns, at runtime, suitable foradaptable locomotion for quadruped and hexapod robots.

The proposed embedded locomotion architecture will exploitthe distributed processing able to carry out in FPGA throughunits working in parallel and ensuring a real time response.



Referencias

References I

P. Arena, L. Fortuna, Frasca, G. Sicurella. An adaptive, self-organizing dynamical system for hierarchical

control of bio-inspired locomotion. IEEE Transactions on Systems, Man and Cybernetics, Part B, 34(4),1823–1837 (2004)

Y.J. Lee, J. Lee, K. K. Kim, and J. Kim, Y.and Ayers. Low power cmos electronic central pattern generatordesign for a biomimetic underwater robot. Neurocomputing, 71:284–296, December 2007

A. J. Ijspeert and A. Crespi. Online trajectory generation in an amphibious snake robot using a lamprey-likecentral pattern generator model. In Proceedings of the 2007 IEEE International Conference on Robotics andAutomation (ICRA 2007), pages 262–268, 2007

P. Manoonpong. Neural Preprocessing and Control of Reactive Walking Machines: Towards Versatile

Artificial Perception-Action Systems (Cognitive Technologies). Springer-Verlag New York, Inc., Secaucus,NJ, USA, 2007.

C. P. Santos and V. Matos. CPG modulation for navigation and omnidirectional quadruped locomotion.

Robotics and Autonomous Systems, 60(6):912 – 927, 2012.

X. Li and L. Li. Efficient implementation of FPGA based central pattern generator using distributed

arithmetic. IEICE Electronics Express. 8 (21): 1848–1854, 2011.

M. A. Lewis. Perception Driven Robot Locomotion. Journal Robot Society of Japan, 20(3):51–56, 2002.

K. Nakada, T. Asai, and T. Amemiya. Analog cmos implementation of a neuromorphic oscillator withcurrent-mode low-pass filters. ISCAS 2005, pages 1923–1926, 2005.



Referencias

References II

S. Inagaki, H. Yuasa, T. Suzuki, and T. Arai. Wave CPG model for autonomous decentralized multi-legged

robot: Gait generation and walking speed control. Robotics and Autonomous Systems, 54(2):118–126,2006. Intelligent Autonomous Systems.

H. Feng and R. Wang. Construction of central pattern generator for quadruped locomotion control. 2008

IEEE/ASME International Conference on Advanced Intelligent Mechatronics, pages 979–984, 2008.

P. Arena, L. Fortuna, M. Frasca, and L. Patane. A cnn-based chip for robot locomotion control. Circuits

and Systems I: Regular Papers, IEEE Transactions on, 52(9):1862–1871, sept. 2005.

K. Matsuoka. Sustained oscillations generated by mutually inhibiting neurons with adaptation. Biological

Cybernetics, 52(6):367–376, October 1985

S. Amari. Characteristics of random nets of analog neuron-like elements. pages 55–69, 1988.

B. Van Der Pol and J. Van Der Mark. The heartbeat considered as a relaxation oscillation, and an electrical

model of the heart. 6:763Ű775, 1928.



Thanks


Documents

Reconfigurable CPGs for the Implementation of Adaptive ...jhbarronz...Reconﬁgurable CPGs for an Adaptive Locomotion Problem Description Problem Description Legged locomotion concerns