1 AMAM2000 Kimura 7 10 Studies on Neuro-Mechanics Sensory - - PDF document

SLIDE 1

AMAM2000 Kimura

1

1

Adaptive Dynamic Walking of the Quadruped on Irregular Terrain

autonomous adaptation using neural system model H.Kimura

• Univ. of Electro-Communications

Tokyo, Japan

2

Decerebrated Cat

[1939] 3

Current Staus in

• Legged Locomotion Studies in Robotics
• Neuro-Mechanics

4

What is legged locomotion?

Stabilization of Non-linear Oscillation

dynamic walking hopping

juggling 5

ZMP Based vs.

Zero Moment Point

Inverted Pendulum Based

ZMP

ZMP

Stable Limit Cycle on Phase Plane

6

Which is primitive?

Static (or Dynamic) Walking based on ZMP position control of ZMP model trajectory based Dynamic Walking based on Inverted Pendulum limit cycle on a phase plane Dynamic Walking using CPG + Reflexes no explicit model torque based No passive static walking Passive dynamic walking Acquired by learning Genetically programmed

SLIDE 2

AMAM2000 Kimura

2

7

Studies on Neuro-Mechanics

• Simulation

Taga, Ekeberg, Ijspeert, …..

• Manipulator

– Miyakoshi [TITECH] juggling – Williamson [MIT] crank, sawing, .. – Kotosaka [ERATO/JST] drumming

• Legged Robot

– Lewis [USC] salamander type – Kimura [UEC] quadruped – Ilg [Karlsruhe, Jena] quadruped – Miyakosi [ETL, U. Tokyo] biped – …..

Why Matsuoka’s Neural Oscillators?

Dynamic Coupling between Neural Controller and Musculoskeleton

8

What’s the output of CPG?

• Torque

– Directly output to actuators – Easily combined with reflexes

• Joint Angle/Velocities

– Inverse dynamics is necessary to calcuate

• utput torque

9

Dual System vs. Single System

control traj. planning

x(t)

u

x(t) : jo int trajecto ry u : joint torque sensor information

u

Rob ot

N _Tr > 0 desir ed a n g l e e q. (4) 4 Y e s N o fle xor e xten so r b o d y angle N _T r Y e s N o

CPG

ves tibul e f | | x > t hresho ld ? eq. (9) eq. (8)

• eq. (6)
• eq. (5)
• eq. (5)

f | | z > thresho ld ?

10

Sensory Feedback to CPGs

• Peripheral sensory input:

– Somatic sensation (joint angle, torque, contact, …) – Vestibular sensation

• Directive signal from upper level:

– Vision

Essential for Adaptation to Irregular Terrain

11

Dynamic Walking on Irregular Terrain

Conventional Method trajectory planning control

precise model

Autonomous Adaptation Problem

?

variety of irregularity

12

Why Neural System Model?

Animals show marvelous ability of autonomous dynamic adaptation. In spite of difference in sensors and actuators, there exist same principles as a physical phenomenon between animals and robots.

SLIDE 3

AMAM2000 Kimura

3

13

A Quadruped Robot

Length:36cm, Width:24cm Height:33cm, Weight:4.8Kg Hip&Knee joint: active DC motor:23W Gear Ratio:40 Ankle joint: passive rate gyro force/torque sensor stereo camera encoder 14

Control Mechanism in Animals

spinal cord ex tenso r muscle flexsor muscle vestibu le

vestibular n u c l e i purkinje n e ur o n s γ

muscle muscle spindle Golgi tendon

• rgan

skin

α

motor neuro ns

vestibula r sensation somatic sensation stretch reflex vestibulospinal reflex flexor reflex brain stem cer ebellu m

α

vison

motor cortex v i s i o n a n d a s s o ciation cort i c e s

cer ebrum

s p i n a l cord

extens or muscle flex sor mus cle

γ

mus cle mus cle spindle Golgi tendon

• rgan

skin

α

motor neurons

somatic sensation

stretch reflex flexor reflex

α

Lower Control Mechanism in Animals

Central Pattern Generator 15

y = m a x (u , 0 )

e e

+

• N_T r

β τ τ ,

Feede

β τ τ ,

N e u ral Osc illator (a) u e u f v e v f

y = m a x (u , 0 )

f f

u0

Σ

y

j

w

ij

u 0 Extensor Neuron Flex

• r Neuron

w

fe

Σ

y

j

wij Feedf

• 1
• 1
• 1
• 1

u 0 u 0

• 1 : Inhibitory Connection

1 : E xcita tory C

• nnection

(b ) Neural Osc illator Network for Trot

L F L H R F R H : left fore leg : l e f t hi nd le g : right fore leg : right hind leg

Neural Oscillator As a Model of CPG

Matsuoka[87], Taga[91]

16

musculoskeletal system

CPG

torque

Control Model

not good for irregular terrain

muscle length contact with floor s o m a t i c sensation tonic s t r e t c h reflex

Fe ed

Proposed by Taga[91]

17

CPG alone

Walking on flat terrain

18

CPG alone

Walking on simple irregular terrain

SLIDE 4

AMAM2000 Kimura

4

19

musculoskeletal system

CPG

muscle le ngth contact with floor somatic sensation tonic stretch reflex

torque Feed

Control Model

v e s t i b u l e p h a s i c s t r e t c h r e f l e x v e s t i b u l o s p i n a l r e f l e x f l e x o r r e f l e x

Reflexes Independent of CPG

20

CPG + Reflexes Independent of CPG

Walking over an obstacle by using flexor reflex

21

CPG + Reflexes Independent of CPG

Walking up a slope by using vestibulospinal reflex

22

vestibule

musculoskeletal system

CPG

muscle le ngth contact wit h floor somatic sensation tonic stretch reflex

torque Feed

Control Model

Biology

• ex. Grillner[83], Cohen[99]

v e s t i b u l a r s e n s a t i o n

phase

v e s t i b u l o s p i n a l r e f l e x t e n d o n r e flex e x t e n s o r / f l e x o r r e f l e x

tendon force & obstacle

Reflexes via CPG

New

23

Neural Oscillator

(1) 24

Sensor Feedback to CPG

vestibuolospinal reflex: tendon reflex: extensor reflex: flexor reflex:

extensor: flexor:

SLIDE 5

AMAM2000 Kimura

5

25

Actual Control Diagram

v s r v s r

N_Tr > 0

d e s ir e d a n g l e e q . ( 4 )

4

Yes No Feed e t s r v s r

flexor extensor body angle

Feed e tr Feed f tsr vsr N_Tr

N_Tr > 0

Yes No Feed f fr

C PG

vesti bule

Feed e e r

f | |

x > threshold ?

e q . ( 9 ) e q . ( 8 ) e q . ( 6 ) e q . ( 5 ) e q . ( 5 ) v s r v s r v s r

f | |

z > threshold ?

rate gyro force sensor

26

Walking on Irregular Terrain with Fixed Parameters

28 cm 30cm 2 cm 3cm 3 cm

44 cm 66cm 3 c m 7 c m 12 5 cm 12

Ability of Autonomous Adaptation

27

CPG + Reflexes via CPG

Vestibulospinal, Tendon and Flexor Reflexes

28

CPG + Reflexes via CPG

Vestibulospinal, Tendon and Flexor Reflexes slow motion

29

CPG + Reflexes via CPG

Vestibulospinal, Tendon and Flexor Reflexes

30

CPG + Reflexes via CPG

Vestibulospinal, Tendon and Flexor Reflexes

SLIDE 6

AMAM2000 Kimura

6

31

1 2 3 4 5 time [sec]

• 2
• 1

1 2 3 CPG torque N_Tr [Nm]

Walking on terrain undulation

flexor reflex tendon reflex Programmed Adaptation

RF LF

Autonomous Adaptation of CPG

flexor extensor 32

Adjustment of External Input to CPG based on Vision

directive signal autonomous motion generation

spinal cord

extensor muscle flexsor musc le vestibule

vestib ular nuclei purkinje neuron s γ

m u sc l e m u sc l e sp i n d l e G

• l

g i te n d

• n
• r

g a n

skin

α

m

• t
• r

n e u r

• n

s C P G

vestibular sen sation somatic sensation

str etch ref lex ves tibulospinal reflex flexor reflex

brain stem cerebellum

α

vison

motor cortex vision and asso ciation cortices

cer ebrum

33

Visual Adaptation

By increasing external input to CPGs

34

Visual Adaptation

Visual Processing Walking over an obstacle

35

Walking over an obstacle

(sec)

RF LF LH RH

3 3 3 3

vision input

(Nm)

u0

N_Tr

CPG torque: external input to CPG:

u

10 10 10 10

u0

adjustment of landing point

36

Adjustment of CPG

• Period
• Amplitude
• Phase

Adjusted by reflexes Adjusted autonomously

• n a CPG network

Adjusted by an external input

SLIDE 7

AMAM2000 Kimura

7

37

What is Dynamic Walking?

Dynamic walking is generated by

external force : gravity internal force : CPG torque

Passive Dynamic Walking (PDW) Neural System Model (NSM)

Mechanism itself has ability of dynamic walking.

38

PDW vs. NSM

1 2 3 4 time [sec]

• 4
• 3
• 2
• 1

Fz

CPG Torque 2 1

Additional Torque by G ravity in PDW

Externla/Internal Torque [Nm] Fz [N]

supporting swinging flexor extensor 39

Mechanical Design & Coupling with a Neural System

• Small gear ratio & Large torque motor

– Backdrivability of a joint for passive adaptation – Quick motion

• Dynamics of mechanical system is encoded

into parameters of neural system

– Relation between the leg length and the frequency of CPG 40

Conclusion & Future Work

• Autonomous adaptive dynamic walking on

terrain of medium degree of irregularity by using reflexes via CPG

• Effectiveness
• Well coordinated system by centering CPG
• 3D dynamic walking on 3D irregular terrain

41

END