A Novel AAN Controller for Locomotion Training in the fMRI

Sensory-Motor System Lab

Prof. Dr. Robert Riener

A Novel Assist-as-Needed
Controller for Locomotion
Training in the fMRI



Fall Term 2011











Supervised by Author:

Dr. Laura Marchal Crespo

Isaac Gomez-Lor Lopez
Dr. Christoph Hollnagel

---

Contents
List of Figures ........................................................................................................................................... 3
List of Tables ............................................................................................................................................ 4
Abstract ................................................................................................................................................... 5
Acknowledgements ................................................................................................................................. 6
1. Introduction ......................................................................................................................................... 7
1.2. Neurorehabilitation ...................................................................................................................... 7
1.3. Locomotor Rehabilitation ............................................................................................................. 7
1.3.1. Epidemiology ......................................................................................................................... 8
1.3.2. Effectiveness .......................................................................................................................... 8
1.3.3. Conventional Physiotherapy ................................................................................................. 8
1.4. Rehabilitation Robotics ................................................................................................................ 9
1.4.1. Rationale for Robotic Rehabilitation ................................................................................... 10
1.5. Robotic Research in Human Motor Control and Learning ......................................................... 10
1.6. Locomotor Rehabilitation Modeled as an Optimization Problem of Error and Effort ............... 11
1.7. Nervous System Recovery .......................................................................................................... 12
1.8. Neuroimaging in Neurorehabilitaion Research .......................................................................... 13
1.9. Survey of Controls for Rehabilitation Robotics .......................................................................... 13
1.9.1. Active-Assistive .................................................................................................................... 14
1.9.2. Active-Resistive ................................................................................................................... 15
1.9.3. Summary.............................................................................................................................. 15
1.10.Guidelines in Patient-Cooperative Strategies ........................................................................... 15
1.11.Assist-as-Needed Algorithm ...................................................................................................... 16
2. MARCOS Robot .................................................................................................................................. 18
2.1. fMRI-Compatible Rehabilitation Robots .................................................................................... 18
2.2. MARCOS General Description .................................................................................................... 18
2.3. Mechanics and Electronics of MARCOS...................................................................................... 19
2.4. Previous MARCOS' Controllers ................................................................................................... 20
2.4.1.Patient Passive. Position controller ...................................................................................... 20
2.4.2. Patient Active. Zero-Force Controller .................................................................................. 21
3. Methods ............................................................................................................................................ 23
3.1. Passive Assistance Faded AAN ................................................................................................... 23
3.1.1. Implementation ................................................................................................................... 24
3.1.2. Simulation ............................................................................................................................ 26
3.1.3. Real Behavior ....................................................................................................................... 29
1

---

3.2. Active Force Modulation AAN .................................................................................................... 30
3.2.1. Continuous-Error Based (CEB) ............................................................................................. 30
3.2.2. Single-Error Based (SEB) ...................................................................................................... 31
3.3. Proof of Concept......................................................................................................................... 32
3.3.1. Overall Behavior .................................................................................................................. 32
3.4. Single-Error Based vs. Continuous-Error Based.......................................................................... 34
3.4.1. Non Systematic Error Performance ..................................................................................... 34
3.4.2. Systematic Error Performance ............................................................................................ 35
3.5. An AAN Fine Controller Based on Active Modulation ................................................................ 39
4. Results ............................................................................................................................................... 41
4.1. Experiment Protocol ................................................................................................................... 41
4.2. General Results ........................................................................................................................... 42
4.3. EMG Results................................................................................................................................ 44
4.4. fMRI Results ................................................................................................................................ 45
4.5. Evaluation ................................................................................................................................... 47
5. Discussion .......................................................................................................................................... 48
6. Conclusion ......................................................................................................................................... 50
6.1. Summary..................................................................................................................................... 50
6.2. Future Work ............................................................................................................................... 50
Annexes ................................................................................................................................................. 51
A Contents of Enclosed CD ................................................................................................................ 51
B Project Timing ................................................................................................................................ 52
C Bibliography .................................................................................................................................... 53









2

---

List of Figures

1.3.3. Body Weight Supported Treadmill training used in conventional physiotherapy........................9
1.4.
Robotic gait orthosis Lokomat (Hocoma AG, Switzerland)...............................................................9
1.6.
Human motor adaption scheme........................................................................................................12
1.9.
Difficulty in doing the task for the different control strategies.....................................................14
1.11. Conceptual overview to human-robot for an assist-as-needed gait training strategy..............17
2.2.
MARCOS device....................................................................................................................................18
2.3.
MARCOS' mechatronics. ....................................................................................................................19
2.3.
Layout of MARCOS into the fMRI scanner. .....................................................................................20
2.4.1. Evolution of the actual performance with ILC and without ILC, compared to the reference
desired one. .........................................................................................................................................21
2.4.1. MARCOS' passive mode controls. Position controller....................................................................21
2.4.2. MARCOS' active controls. Zero-force controller. ...........................................................................22
3.1.
Fading of the P-controller and remaining effect of the ILC compensation. ...............................23
3.1.1. MARCOS' passive assistance faded AAN. .........................................................................................24
3.1.1.2. Change of frequency without data normalization..........................................................................25
3.1.1.2. Change of frequency with data normalization................................................................................25
3.1.1.2. Data normalization according to current frequency......................................................................26
3.1.2.1 Robot model for the passive assistance faded AAN simulation.....................................................26
3.1.2.2. Human model for the passive assistance faded AAN simulation..................................................27
3.1.2.3. Model for the passive assistance faded AAN simulation...............................................................28
3.1.2.3. Simulation results. ..............................................................................................................................29
3.2.
Active force modulation concept for an AAN. ................................................................................30
3.2.2 CEB measurement of the error..........................................................................................................31
3.2.2 SEB measurement of the error...........................................................................................................31
3.3.1. Overall behavior during learning for the active force modulation AAN......................................32
3.3.1. Overall behavior after learning for the active force modulation AAN.........................................33
3.3.1. Force pattern after the passive learning for the active force modulation AAN. ........................33
3.3.1. Overall performance after learning for the active force modulation AAN with foot ground
effect .....................................................................................................................................................34
3.4.1. Comparison between CEB and SEB with voluntary good performance. .....................................35
3.4.2 Visual feedback provided during the experiments. .......................................................................35
3.4.2. Systematic voluntary deviation for error compensation test........................................................37
3.4.2. Error comparison between CEB and SEB with voluntary wrong performance...........................37
3.4.2. Performance comparison between CEB and SEB with voluntary wrong performance.............38
3.4.2. Systematic error compensation comparison for SEB and CEB. ....................................................39
3.5.
SEB + CEB in cascade AAN. .................................................................................................................40
4.1.
Experiment's protocol flow chart. ....................................................................................................42
4.2.
Assistance evolution for the 6 subjects. ..........................................................................................43
4.2.
Average mean absolute tracking error evolution for the 6 subjects. ..........................................43
4.3.
Set of muscles for the EMG experiment...........................................................................................44
4.3.
Correlation between mean Bicep Femoris EMG activity and mean assistive force (RMS).......45
4.3.
Mean Bicep Femoris EMG evolution during the experiment........................................................45
4.4.
Main brain regions involved in motor planning and coordination...............................................46
3

---

4.4.
AAN and active-mode brain activity contrast .................................................................................46
4.5.
Response of a point of the primary motor cortex as a function of the assistance.....................47
List of Tables

1.4.1. Manually assisted therapy versus robot aided therapy.................................................................10
1.10. Guidelines for Patient-Cooperative Strategies................................................................................16
2.2.
MARCOS to real gait comparison.......................................................................................................19
4.3.
EMG analyzed muscles........................................................................................................................44






















4

---

Abstract

In this thesis it is explained the work done in order to develop a patient-cooperative control that
assists only as it is needed for the MR-compatible stepper (MARCOS) developed at the Sensory
Motor System Lab of ETH Zurich.
The so called assist-as-needed (AAN) control algorithm adapts the robotic assistance to optimize the
patient's effort and the error generated while performing a task. This algorithm has already been
applied to some rehabilitation devices. In this work an implementation of such strategy for MARCOS
is described. It is shown that in MARCOS an AAN has to be based on a force controller due to some
limitations related to its pneumatic actuation.
Preliminary results of the new controller's performance are provided. A first experiment based on
EMG-measured muscular activity reported a correlation between assistance level and Bicep Femoris
activity, evolving from a patient-passive to a patient-active situation. A second experiment aimed to
observe the brain activity related to the assistance. It also showed a negative dependence between
primary motor cortex activity and the degree of assistance. Furthermore, an active-like activation of
the brain was observed during the AAN. The results obtained showed the adequacy of the new
controller to help the patient and, at the same time, stimulate him/her to participate in the walking
task.
Finally, the controller is compared to those found in literature. It is discussed the advances that such
AAN controller may bring for future research projects, as its force controller-based implementation
brings a proper scenario to a natural evolution from assistive rehabilitation strategies to resistive
ones.












5

---

Acknowledgements

To Robert Riener for offering me the opportunity of taking part on his team.
To Laura Marchal and Christoph Hollnagel for suggesting me this very interesting project and
supporting me all the time.
To Lukas Jager for helping me on the fMRI experiments.
To Jasmin Schneider for being so kind and help me with the EMG experiments.
To Marina , my family and God for being always by my side.



















6

---

1. Introduction

In the last twenty years there has been intense research effort to develop robots that can perform
rehabilitation exercises to aid patients with locomotion impairments originated from Neural System's
injuries. The purpose of these robots is to help physiotherapists on their labors, enlarging session's
times and automating the rehabilitation process. It has been hypothesized that the use of
rehabilitation robotics may enhance the recovery process. A very promising field inside rehabilitation
robotics is Magnetic Resonance Imaging (MRI)-compatible robotics, robots designed to be
compatible with the high magnetic fields operating inside of an MRI scanner so that brain activity can
be recorded during the training tasks. Obtaining direct information of the brain activity could make
possible to understand better neuro-recovery and to point to the most effective rehabilitation
approaches.
The purpose of the project described in this thesis is to develop an "assist-as-needed" controller for
the MRI-compatible robot MARCOS developed in the Sensory Motor System Lab of the ETH Zurich.
The "assist-as-needed" control strategy aims to promote human active participation in performing a
task through a customized robotic assistance that provides the assistance needed to perform the
task only if the person cannot do it on his/her own. This is achieved by progressively fading the robot
assistance applying a "forgetting factor". This stimulates the person to compensate with effort for
the lack of assistance, and thus preventing the patient to rely on the robot assistance.
The behavior of such a controller is based on physiotherapists' common practice of guiding the
impaired limb physically to perform a task with the intention of teaching instead of driving so the
patient becomes more autonomous with time.

1.2. Neurorehabilitation
Neurorehabilitation aims recovery from nervous system injuries, helping the patient to increase his
or her quality of life and achieving more independency. There are several injuries that affect the
brain (encephalopathy) or the spinal cord (myelopathy), and lower physical or psychological
functions like mobility or cognitive functions. These nervous system's injuries include, among others:
stroke, cerebral palsy, Parkinson's disease, multiple sclerosis and spinal cord injury (SCI).
Neurorehabilitation tools are a set of therapies that aim to recover patients with a nervous system
injury. These therapies include occupational therapy, psychological therapy, speech and language
therapy, physical therapy, etc.

1.3. Locomotor Rehabilitation
One of the most common consequences of a nervous system disease is the total or partial loss of
mobility functions and abilities. The goal of locomotor physiotherapy (physical therapy) is to teach or
re-train locomotor abilities. This includes: balance retraining, neuromuscular retraining, etc.
7

---

The physical therapy to recover from mobility loss intends to achieve impairment recovery: strength
in muscles, range of motion, balance, co-ordination, naturalistic trajectories [18]; and functional
performance recovery: task-specific training like activities of daily living (ADLs) [12].
1.3.1. Epidemiology
The majority of studies related to locomotor neurorehabilitation refer to stroke or spinal cord injury
(SCI) as the causes of the impairment, as they are common nervous system injuries that are normally
accompanied by locomotion malfunctions.
Stroke is the third mortality cause worldwide, being one of the greatest health public problems in
industrialized countries [3]. It is the second cause of death in Western world, over cancer [4]. Each
year there are 920.000 new stroke cases reported and around 80% of the survivors (there is a 22% 30
day mortality) need rehabilitation interventions [13].
SCI epidemiology data is more diffuse and there are not comparable worldwide registers. However,
some studies have tried to unify the available data. An SCI incidence between 10.4 and 83 cases per
million inhabitants and year has been reported [23]. The principle causes of SCI are traffic accidents,
falls, violence and sports injuries.
More than one half of both stroke and SCI survivors, show incomplete lesions and are thus
theoretically able to recover independent and functional mobility after an intense rehabilitation
process [1].
1.3.2. Effectiveness
While the mechanisms of brain plasticity underlying neurorehabilitation are still not well known,
physical therapy is commonly used in order to enhance motor skills of patients that suffer from
nervous system injuries, both degenerative and traumatic [18], and its effectiveness is strongly
supported [1].
However, several factors determine the effectiveness of the intervention. Some factors depend on
the severity of the injury (site, etiology and chronicity) and some others on the treatment (type,
duration and specificity) [1]. The efforts in locomotion neurorehabilitation respond to the fact that
the treatment factors are controllable and so, susceptible of being optimized.
Ideally, knowing the severity of the injury a specific activity and timing planning for the recovery
intervention could be provided. Unluckily, nowadays it has not been yet possible [12].
1.3.3. Conventional Physiotherapy
The conventional intervention consists in a set of exercises and techniques supervised by
physiotherapists that strengthens the muscles, maintain the range of motion and look for balance,
coordination and appropriate motion trajectories.
In the recent decades, new therapies for locomotion recovery have been proposed. They include
body weight support (BWS) exercises, functional electrical stimulation (FES) and pharmacological
solutions. These more recent techniques (specially the BWS-aided or a combination of them) have
shown greater effectiveness compared to the previous therapies [1]. For example, the manually
assisted BWS Treadmill (BWST) training has been widely successful used in hemiparetic patients and
has shown positive therapeutic effects in other neuropathologic conditions [19].
8

---

Figure 1. Body Weight Supported Treadmill training used in conventional
physiotherapy. [Source: Journal of Rehabilitation Research and Developement.

2000. 37(4)]

1.4. Rehabilitation Robotics
Over the past twenty years there has been an increasing interest in developing robotic devices for
assist in neurorehabilitation, both for the upper and the lower limbs [10]. During the past ten years
robotic-aided neurorehabilitation research has been more intense [14]. Although also non-contact
coaching robots have been developed [14], these devices are generally robots that interact with the
human limb along a desired trajectory (predefined or patient defined). They are equipped with
sensors that record data from the patient performance (velocity, position, force, joint torques, etc.)
and actuators (electric motors, valves, pneumatic cylinders, etc.) that move the limb following a
desired strategy (algorithm).

Figure 2. Robotic gait orthosis Lokomat (Hocoma AG, Switzerland).
9

[Source: Medimex commercial brochure]

---