Vittorio Sanguinetivittorio.sanguineti@iit.it

Italian Institute of TechnologyVia Morego 30, 16163 Genova (IT)www.iit.it

PRESENTATION: CogSys 2010, Zurich, SUI

MOTOR SKILL LEARNING THROUGH PHYSICAL ASSISTANCECONCEPT

The HUMOUR project aims at investigating and

developing efficient robot strategies to facilitate

the acquisition of motor skills.

We address both the (human) trainee and the

(robot) trainer sides, by combining behavioural

studies on motor learning and its neural

correlates with design, implementation, and

validation of robot agents that behave as ‘optimal’

trainers, which efficiently exploit structure and

plasticity of the human sensorimotor systems.

HUman behavioral Modeling for enhancing learning by Optimizing hUman-Robot interaction (HUMOUR): concept and preliminary results1Vittorio Sanguineti, 2Dejan Popovic, 3Roberto Colombo, 4Niels Birbaumer, 5Herbert Heuer, 6Etienne Burdet 1Fondazione Istituto Italiano di Tecnologia, Genoa (IT); 2Aalborg Universitet, Aalborg (DK); 3‘Fondazione Salvatore Maugeri Clinica

del Lavoro e della Riabilitazione, Pavia (IT); 4Eberhard-Karls-Universitat,Tubingen (DE); 5Forschungesellschaft fur

Arbeitphysiologie und Arbeitsschutz e.V., Dortmund (DE); 6Imperial College of Science, Technology and Medicine, London (UK)

GENERAL OBJECTIVES

Modular robot system:2-arm planar manipulandum

3D wrist device

1D hand device

Common software platform:Device-independent

Multiple devices

Multi-OS (Windows, Linux)

Open Source, GPL (built on top of H3D, www.h3d.org by SenseGraphics)

Analytic and modeling toolsPerformance evaluation

Optimal schemes of assistance

Model of robot-assisted motor recovery after stroke

Robot-assisted motor skill learning:Redundant tasks (putting, handwriting transfer) –

Poster #127

Visuomotor rotation – Poster #114

Neural correlates of assistive force

We will focus on the cognitive and neural

mechanisms underlying the acquisition of a

variety of motor skills, by specifically aiming at

understanding the way humans physically

cooperate in acquiring a motor skill and how

physical assistance affects motor learning.

Experiments will enable us to identify

determinants and dynamics of the learning

process in representative motor tasks, and will

provide the foundations for designing efficient

schemes of assistance.

1. Robot trainers. To develop robot agents based on an advanced understanding of human neuromotor control, its development, and skill acquisition, which will enable them to mimic (and surpass) human trainers in supporting motor skill learning and neuromotorrehabilitation

2. Application to motor skill learning and rehabilitation. To validate robot training agents for a number of different motor skills, modalities of interaction and rehabilitation applications

3. Brain-computer interfaces. To extend the domain of Brain-Computer Interface (BCI) technologies to the fields of motor learning and neuromotor rehabilitation

4. Behavioural studies on physical interaction and motor learning. To understand how physical interactions affect motor learning and – on this basis - to develop a general theoretical and technological framework for more effective motor skill learning

Movement

ASSISTANCE

MODULE

ACTION

MODULE

Assistance

Movement

Subgoals

CONTROLLER BODY

INTERNAL

MODEL

TASK

MODEL

TRAINEE

ROBOT

TRAINER

Neural correlates

PERFORMANCE

METRIC

Performance

PerformanceCONTROLLER

Assistance TRAINEE

LEARNING

Estimated

Volitional Control

Desired Volitional Control

REGULATION OF

ASSISTANCE

TRAINEE

LEARNING

MODEL

RESULTS (first year)

www.humourproject.eu

Movement

ASSISTANCE

MODULE

ACTION

MODULE

Assistance

Movement

Subgoals

<Scene DEF="world">

<!— THE DEVICE (robot end effector and attached tool-->

<Inline load="true" url="default.x3d"/>

<Inline DEF="device" load="true" url="deviceinfo.x3d"/>

<IMPORT inlineDEF="device" exportedDEF="HDEV" AS="HDEV"/>

…

<!— THE TARGET-->

<Transform DEF='target' rotation='1 0 0 1.507'>

<Shape>

<Appearance>

<Material DEF='tgtColor' ambientIntensity='0.1'

diffuseColor='1 1 1'

shininess='0'/>

</Appearance>

<Sphere radius='0.01'/>

</Shape>

</Transform>

<!—A SOUND (used as cue or feedback) -->

<Sound>

<AudioClip DEF="Ask" url="ding.wav"/>

</Sound>

…

</Scene>

<automaton>

<!-- STATE LIST-->

<state id="0" name="init" type="start"/>

<state id="1" name="moving"/>

<state id="2" name="stopped"/>

<state id="3" name="feedback"/>

<state id="4" name="back"/>

<!-- STATE TRANSITIONS -->

<transition from="init" to="moving">

<gt type="double" field="velocity" value="0.05"/>

</transition>

<transition from="moving" to="stopped">

<and>

<timer delay="1"/>

<lt type="double" field="velocity" value="0.05"/>

<lt type="double" field="distance" value="0.01"/>

</and>

</transition>

<transition from="stopped" to="feedback">

<timer delay=".1"/>

</transition>

<transition from="feedback" to="back">

<timer delay="1"/>

</transition>

<transition from="back" to="init">

<and>

<timer delay="2"/>

<lt type="double"

field="distance_from_center" value="0.01"/>

</and>

</transition>

</automaton>

WORLD

MODEL

ACTION

MODEL

500 1000 15000

2

4

wi

500 1000 15000

0.2

0.4

yi

500 1000 15000

0.2

0.4

yi est

500 1000 1500-0.1

0

0.1

xi

Trial Number

200 400 600 800 1000 12000

2

4

6

8

wi

200 400 600 800 1000 12000

0.2

0.4

yi

200 400 600 800 1000 12000

0.2

0.4

yi est

200 400 600 800 1000 1200-0.2

0

0.2

xi

Trial Number

<PositionEffect r0="pREFx

pREFy

pREFz

"

K="Kp11

Kp12

Kp13

Kp21

Kp22

Kp23

Kp31

Kp32

Kp33

"/>

<VelocityEffect v0="vREFx

vREFy

vREFz

”

B="Kv11

Kv12

Kv13

Kv21

Kv22

Kv23

Kv31

Kv32

Kv33

"/>

ASSISTANCE

0

20

40

60

80

100

120

0 10 20 30 40

0

5

10

15

20

25

30

35

0 10 20 30 40

0

0.5

1

1.5

2

2.5

0 10 20 30 40

0

20

40

60

80

100

0 10 20 30 40

Efficacy Velocity

Accuracy Efficiecy

Session # Session #

Session # Session #

Performance Metrics