The Challenge - Stanford Computer Science...ASIMO.. not only walk, but also interact with the world! How much programming? 12 34 56. 2 Mobile Manipulation Human Guided Motion & Human-Robot

1

Humanoid Robots

The Challenge:beyond mobility

.. not only walk, but also interact with the world!a unified mobility & manipulation framework! ASIMO

.. not only walk, but also interact with the world! How much programming?

1 2

3 4

5 6

2

Mobile ManipulationHuman Guided Motion & Human-Robot Interaction

Romeo & Juliet (1993)

Human-Robot InteractionRomeo & Juliet (1993-96)

Multi-contact Manipulation

7 8

9 10

11 12

3

• whole-body control

• constraints and obstacles

• multiple contacts

• internal forces & balance

• manipulation skills

Humanoid Robot Controlbranching and under-actuated Multi-Contact Whole Body Motion

Physical Interaction, Constraints, and Task Behaviors

Balanced Supporting ContactsInternal Force Control – Virtual Linkage

Balanced Supporting ContactsInternal Force Control – Virtual Linkage

Task BehaviorPosture

Balance

Internal ConstraintsSelf CollisionLocal Obstacles

Multiple Contacts

Physical Interaction, Constraints, and Task Behaviors

13 14

15 16

17 18

4

Task consistent with {

interact with the world, cooperate, and manipulate

Unified Whole-Body Control Framework

postureconsistent with {

balance

}

internal constraintsself collisionlocal obstaclesconsistent with {

contactconsistent with {

}} }

interact with the world, cooperate, and manipulate

Constraint-Consistent Task-Space

cc space

fp F F

Control

ˆ ˆ ˆF pF

Dynamics

Unified Whole Body Control FrameworkTask, Posture, Constraints, Multiple Contacts, and Balance

Task & Posture Decomposition

Whole-body Control

Task Dynamics and Control

Task Control

T FJ

Task Dynamics

)ˆ ˆ ˆ( TaskV pF TaskVF F

x

x Fp

Task Dynamics – Branching Structures

x J

1

2

.

m

x

xx

x

5x5f

2x2f

3x3f1x 1f

6x6fx Fp

12 1

21 2

11

22

1 2

...

...

...................

...

L

L

L L LL

19 20

21 22

23 24

5

Posture Space Control

TJT

J

Task Control: TJ

Posture Control: TN

Joint-Space Dynamics

Task-relatedPosture

TN x Fp

Control Structure

task postuT Ttask task reJ NF Γ Γ

0taskx

Decomposition of torque vector

Dynamic Consistency

Posture Energy

Whole-Body Control: Dynamics

( )Posture Desired Po t rT

s u eN V Γ

ˆ ˆ ˆ[ ( ])Task a kT

T s pVJ Γ

PostureTask ΓΓΓ

Posture Desired Po uT

st reN Γ

TaskT FJΓ

Dynamics in Posture Space

25 26

27 28

29 30

6

Posture Dynamics & Control

p px J

| tp t pJ J N

Posture sub-task:

Task-consistentposture Jacobian:

| | | | |p t p t p t p t p tx p F Dynamics:

Control:

| | ||ˆ ˆ ˆ[ ]( )p t postu

Tp t p tposture rep tJ pV Γ

| ( )p t p tx J qN

Constraints

Constraints and Priorities

( )T Ttas

T Tcns task post cnst k taskn t turs ec FJJ NNF Γ ΓΓ T

taTta sksk task postureNJ F Γ

Self Collision

Obstacles

31 32

33 34

35 36

7

Multi-Contact Whole-body ControlIntegration of Whole-Body Control & Locomotion

Under-actuated Balance Reaction forces

( )T Tcontact conta

T Ttascontact task postuc ret k taskJF NN FJ Γ Γtask postu

T Ttask task reJ NF Γ Γ[ ( )]T T

contact contT Ttas

Tactact contact tk taask posturekact sJFJ F NNJ Γ Γ

Multi-Contact Whole-body ControlIntegration of Whole-Body Control & Walking

Balanced Supporting ContactsInternal Force Control – Virtual Linkage

Unified Whole Body Control FrameworkTask, Posture, Constraints, Multiple Contacts, and Balance

taskconsistent with {

interact with the world, cooperate, and manipulate

Unified Framework

postureconsistent with {

balance

}

internal constraintsself collisionlocal obstaclesconsistent with {

contactconsistent with {

}} }

37 38

39 40

41 42

8

interact with the world, cooperate, and manipulate

Constraint-Consistent Task-Space

cc space

interact with the world, cooperate, and manipulate

Constraint-Consistent Task-Space

|

| |

| | |

| | | |

c

f c

m f c

p

s

m

s

s

f c s

fp F F

Control

ˆ ˆ ˆF pF

Dynamics

( )T Tsa UN FJ

Unified Whole Body Control FrameworkTask, Posture, Constraints, Multiple Contacts, and Balance

Actuated Torques

Unified Whole Body Control FrameworkTask, Posture, Constraints, Multiple Contacts, and Balance

43 44

45 46

47 48

9

49 50

51 52

53 54

10

Human Movement

understandinghuman movment

Motion capture

Simulation 79 DOF and 136 MusclesBiometric Data & Bone Geometry

Dynamic simulation

Data from Subjects

Physiology-based Posture Field

T FJ TL m

A Task, F:

Muscle actuation:

Data from Subjects

55 56

57 58

59 60

11

In learned tasks, humansminimize muscular effort, under physical and “social” constraints

Learning from the Human

Humans grow to learn the efficiency of the body’s physio-mechanical advantage

Physio-Mechanical Advantage

Physio-Mechanical Advantage Physiology-based Posture Field

Human posture is continuously adjusted to reduce muscular effort

2E cmMuscular Energy minimization:

Function of physiology, mechanical advantage, and task

2 1( )[ ]T TT TcL N LE F F FJ J E F

2E cm

Validation - Arm Effort

2E cm

Validation - Arm Effort

61 62

63 64

65 66

12

Validation – whole-body effort

2E cm2E cm

Upper body

Lower body

Physio-Mechanical Advantage

ASIMO

SAI Neuromuscular Library

67 68

69 70

71 72

13

Human MotionAtlas

Human Motion ReconstructionInjury prevention, Pathology Evaluation, and Athletics

Motion capture

Simulation 79 DOF and 136 MusclesBiometric Data & Bone Geometry

Dynamic simulation

73 74

75 76

77 78

14

79