Design av en rörlig gummihand för användning i miljöer med magnetisk resonans

(1)

Design of an actuated rubber hand for use in magnetic resonance imaging environments

SOFIA CRUZ-FERREIRA FRÖMAN

Master of Science Thesis

(2)

(3)

Design of an actuated rubber hand for use in magnetic resonance imaging environments

Sofia Cruz-Ferreira Fröman

Master of Science Thesis MMK 2011:55 MDA408 KTH Industrial Engineering and Management

Machine Design SE-100 44 STOCKHOLM

(4)

(5)

Master of Science Thesis MMK 2011:55 MDA408

Design of an actuated rubber hand for use in magnetic resonance imaging environments

Sofia C. Fröman

Approved

2011-06-14

Examiner

Mats Hanson

Supervisor

Jan Wikander

Commissioner

KI Neuroscience

Contact person

Alexander Skoglund

Abstract

For this thesis, an initial prototype for an actuated rubber hand with the index finger actuated, was designed and manufactured. A rubber hand is an artificial hand used for studies of body image and ownership, an integral part of research on consciousness in human beings. With an artificial hand, researchers can create an illusion in which a subject can believe that the artificial hand is part of their body. To increase the immersion into the illusion, a cosmetic prosthetic glove can be used for added realism. Another integral part of body image is the sense of agency, in which a subject believes to be the source of an event in the external world. To study agency, while separating agency from ownership, a moving rubber hand is a useful tool. The idea is that a subject is to control the moving rubber hand remotely with their own hand. To be able to study the changes in brain activity during the onset of the illusion, the subject needs to be in an MRI scanner and undergo a so called functional MRI analysis. Therefore, all the materials and components in the actuated rubber hand needed to be MRI compatible.

(6)

(7)

Examensarbete MMK 2011:55 MDA408

Design av en rörlig gummihand för användning i miljöer med magnetisk resonans

Sofia C. Fröman

Godkänt

2011-06-14

Examinator

Mats Hanson

Handledare

Jan Wikander

Uppdragsgivare

KI Neurovetenskap

Kontaktperson

Alexander Skoglund

Sammanfattning

För denna avhandling har en första prototyp av en rörlig gummihand tillverkats, där endast pekfingret rör sig. En gummihand (rubber hand på engelska) är en konstgjord hand som används inom studier av kroppsuppfattning och ägande, vilka utgör ett viktigt område inom studier av människans medvetande. Forskare kan skapa hos testdeltagare illusionen av att en konstgjord hand är en del av en deras kropp. För att förstärka illusionen så kan en kosmetisk proteshandske användas. Ytterligare en del av kroppsbilden är känslan av att orsaka en händelse i den yttre världen. En rörlig gummihand är ett värdefullt verktyg för att studera känslan av både orsakande och ägande, som kan hållas isär ändå. Tanken är att en testdeltagare ska kunna fjärrstyra den rörliga gummihanden. För att studera ändringar i hjärnans aktivitet under illusionens gång så måste deltagaren befinna sig i en MR-skanner och undergå en så kallad funktionell MR undersökning. Därför behövde alla delar och komponenter i den rörliga gummihanden vara MR- kompatibla.

(8)

(9)

List of Figures

1 The names of the bones in the hand. Source: Heikenwaelder Hugo on Wikimedia commons, with alterations. . . . ix 1.1 Classical and moving rubber hand illusion. Source for both: An-

dreas Kalckert, BBS Lab. . . . . 2 1.2 The general system as planned. Source: Sofia C. Fr¨oman, BBS

Lab. . . . 5 2.1 Ways of performing the moving rubber hand illusion. Source:

(a). [2] and (b). [14]. . . . . 8 2.2 Feedback video rubber hand illusion. Source: [15]. . . . 9 2.3 A selection of robotic hands. Source: (a). [18], (b). Sofia C.

Fr¨oman, BBS Lab, (c). [19] and (d). [20]. . . . 10 2.4 (a). The FLUIDHAND. (b). FLUIDHAND bellows. (c). The

ultralight hand. Source: (a). [28], (b). [28] and (c). [29]. . . . . 12 2.5 The FLUIDHAND with a prosthetic glove. Source: [28]. . . . 13 4.1 Design schedule in weeks. Source: Sofia C. Fr¨oman, BBS Lab. . 22 4.2 Operation of a PAM. Source for both: [18]. . . . . 25 5.1 Left-handed prosthetic glove. Source: Sofia C. Fr¨oman, BBS Lab. 29 5.2 Different configurations of finger prototype 0.1. Source for all:

Sofia C. Fr¨oman, BBS Lab. . . . 30 5.3 Close up of ¨Orebro hand joint in different positions. The drawn

on outlines highlight the geometric shape. Source for all: Sofia C. Fr¨oman, BBS Lab. . . . 32 5.4 Close-up of iCub knuckles. Source: [19]. . . . 33 5.5 (a). Distal tip. (b). Parallel tendon holes. Source for both: Sofia

C. Fr¨oman, BBS Lab. . . . 34 5.6 Finished finger prototype 1.0, also demonstrating the difference

in glove deformation with and without finger prototype 1.0, and one design issue. Source for all: Sofia C. Fröman, BBS Lab. . . 35 5.7 Static PAM model. Source: Sofia C. Fröman, BBS Lab. . . . 36 5.8 Valve configuration for one PAM. Source: Sofia C. Fröman, BBS

Lab. . . . 38

(14)

5.9 Tendon tightener and rig. Source for both: Sofia C. Fr¨oman, BBS Lab. . . . 39 5.10 Flow chart for the bang-bang controller. Source: Sofia C. Fr¨oman,

BBS Lab. . . . 40 5.11 Diagram of the hardware system. Source: Sofia C. Fr¨oman, BBS

Lab. . . . 41 6.1 Results of a sinus wave angle reference. Source: Sofia C. Fr¨oman,

BBS Lab. . . . 44 6.2 Results of a step wave angle reference. Source: Sofia C. Fr¨oman,

BBS Lab. . . . 45 6.3 Results of a step wave pressure reference. Source: Sofia C. Fr¨oman,

BBS Lab. . . . 47 6.4 Results of a step wave pressure reference for a system with longer

tubing. Source: Sofia C. Fröman, BBS Lab. . . . 48 B.1 The Bill of Materials. Source: Sofia C. Fröman, BBS Lab. . . . D F.1 Kinematic study of finger. Source for both: Sofia C. Fröman,

BBS Lab. . . . I

(15)

List of abbreviations

BBS - Brain, Body, Self DOF - Degree of Freedom

fMRI - Functional Magnetic Resonance Imaging MRI - Magnetic Resonance Imaging

PAM - Pneumatic Air Muscle PCU - Pneumatic Control Unit ROS - Robot Operative System

SPCU - Shadow Pneumatic Control Unit

(16)

(17)

Notations

The anatomical names of the bones in the hand will be used throughout the thesis even in reference to the artificial hand. Figure 1 presents the names of the bones of the hand.

Figure 1: The names of the bones in the hand. Source: Heikenwaelder Hugo on Wikimedia commons, with alterations.

Increasing a joint angle is referred to as extension of the joint; decreasing a joint angle is referred to as flexion of the joint.

(18)

(19)

Chapter 1

Introduction

1.1 Background

The goal of this project was to initiate the first prototype for an actuated rubber hand. An actuated rubber hand is a tool that can be used by brain scientists for investigating the relationship between the body, mind and self by experimenting on human subjects. While investigation into the higher functions of the brain is still in the experimental stages, the conclusions drawn from such research can in the future be used to help those with deviations in normal cognitive function. The project was proposed by the Body, Brain and Self (BBS) Lab, led by Henrik Ehrsson, at the Department of Neuroscience at the Karolinska Institute in Stockholm.

1.1.1 Body ownership and the rubber hand illusion

The sense of ownership can be described as the “sense that I am the one who is undergoing an experience [. . .] that my body is moving regardless of whether the movement is voluntary or involuntary” [1].

The BBS Lab focuses on identifying the multi-sensory mechanisms that dic- tate how the brain separates the body from the external environment, and so forms the self. Illusions that distort the body image are useful for researching this ability that the brain possesses. One such illusion is the rubber hand illusion, in which a subject is made to believe that an artificial hand is part of their body. Figure 1.1a demonstrates how the illusion can be induced.

In the rubber hand illusion, a subject’s hand is hidden from view, and replaced with an artificial hand in front of said subject. Figure 1.1a shows the subject’s hand hidden behind the screen on the left. The artificial hand and the subject’s hidden hand are then stroked with brushes simultaneously. The brain can then try to resolve the conflicting sensory inputs from the different sensory modalities. That is, the difference in hand location according to the visual input compared to the inputs from touch and proprioception. The conflict may then

(20)

(a) (b)

Figure 1.1: Classical and moving rubber hand illusion. Source for both: Andreas Kalckert, BBS Lab.

be resolved by incorporating the artificial hand as part of the body image and abandoning the real hand.

A change in body image is often of a subjective quality, but there are quantitative measures for the degree in which the artificial hand is a part of the subject’s body image. One quantitative measurement concerns the subject’s change of skin conductance. A stress response includes an increase in the amount of sweating, which in turn increases the skin conductance. A stress response can be induced by threatening the subject, perhaps by placing a knife close to or on their real hand. If doing the same on the artificial hand also results in an increase of skin conductance, the artificial hand is then part of the body image of the subject.

Another quantitative measurement of the incorporation of the artificial hand into the subject’s body image is given by measuring the perceived shift in hand location. That is to say, where the subject believes their hand is located, and how far that is from the actual location.

The sense of body ownership and body image is entwined with the sense of agency. The sense of agency concerns the events in the external world that is percieved as being caused by oneself.

(21)

1.1.2 Determining the source of agency

Agency can be described as “the sense that I am the one who is causing or generating an action. For example, the sense that I am the one who is causing something to move, or that I am the one who is generating a certain thought in my stream of consciousness” [1].

The combination of proprioceptive sensory feedback and the motor signals sent by the brain create a “Sense of Agency” [1], and can be added to other information from the external world from other senses, such as vision [2]. It has been suggested that the brain creates an internal model that can be used in controlling the actions of the body and for recognizing the source of these actions. The brain does this by comparing the predictions created by the model with sensory feedback [3]. It is further hypothesized that the sense of agency is influenced depending on if the action is active (with intention) or passive [4].

According to the study in [3], two important parameters for action, that will influence the sense of agency, are time and direction. Time can be manipulated by introducing latency from when an action was predicted to take place by the subject, to when the action actually takes place. Direction can be manipulated by introducing spatial discordance between where the action was predicted to take place by the subject, and the location where the action actually takes place.

It is theorized that spatial discordance plays a greater role in influencing the brain’s function in determining the agent. By introducing such discordances in subjects’ actions, the ability of the brain to determine an agent can be studied.

1.1.3 Separating ownership and agency

Body ownership is closely related to agency, and often the two phenomena occur at the simultaneously. In order to separate the two, it was proposed to have a rubber hand illusion with a moving hand, controlled remotely by the subject.

A simple but effective set-up was designed and built by Andreas Kalckert of the BBS Lab, seen in Figure 1.1b.

The set-up is made up of a mannequin hand clad with a glove, and a vertical axle that connects the index finger of the subject to the index finger of the mannequin hand. The mannequin hand’s index finger therefore moves in a similar way to the subject’s index finger in the vertical plane. The subject’s hand is hidden in the box underneath the mannequin hand.

The regular mode of operation is when the subject moves the mannequin hand’s index finger with their own, and both the mannequin hand and subject’s hand are facing the same way (as in Figure 1.1b). Other modes of operation are that the researcher uses the axle to move the subject’s finger, which moves the mannequin hand’s index finger, which is passive movement from the perspective of the subject. The mannequin hand can also be placed backwards while the index finger is still controlled by the subject, introducing spatial discordance.

These different configurations allow researchers to study the effects of passive and active control, and of spatial discordance, on the process of the brain in creating body image.

(22)

However, the set-up has several limitations, as only one movement is possible (the up and down movement of the index finger), and the mannequin hand has to be placed on top of the box. Furthermore, introducing time discordance is difficult. An automated solution would allow for the control of latency between the subject’s movement and the artificial hand’s movement.

Though there are quantitative measurements for determining the effect of a non-moving rubber hand illusion on the body image of the subject, there is as yet no such analogous measurement for the sense of agency. Determining what was going on in the subject’s brain during the experiment can be done using questionnaires after the experiment, but questionnaires are notoriously subjective. Therefore, it can be useful to be able to monitor changes in blood flow within the brain during these experiments, which gives an indication of areas being activated [5]. This can be done with Functional Magnetic Resonance Imaging (fMRI) in a Magnetic Resonance Imaging (MRI) csanner.

A MRI scanner produces a powerful magnetic field (about 1.5-3.0 Tesla [6, 7, 8] and up to 7 Tesla for research purposes [7]), wherein hydrogen atoms within the body align themselves in the direction of the magnetic field. The scanner then applies a radio frequency pulse, so that some of the atoms resonate at a given frequency. The resonating atoms can be picked up by the scanner and an image of the internal anatomy of the body can be recorded.

1.1.4 Functional Magnetic Resonance Imaging

fMRI distinguishes itself from regular MRI in that the brain is scanned several times over a time span, and the images are of a lower resolution. The higher the frequency of the scans, the lower the resolutions are of the images produced, and vice versa. The loss of resolution and the time span leads to fMRI being more sensitive to artifacts than regular MRI [7].

The powerful magnetic fields produced by the MRI scanners raise a challenge for the researchers who wish to use equipment in the scanner room. There are two major criteria for equipment allowed in the scanner room during a scan.

Firstly, equipment in the scanner room must not be affected by the scanner, as this poses a safety risk. Usual safety concerns are, for example, ferromagnetic materials that are attracted to the strong static magnetic field, otherwise called the missile effect [9]. Another example is eddy currents that may occur, which can cause mechanical and thermal effects on materials, which may cause skin burns on the subject inside the scanner while in close proximity to said materials.

Other concerns stem from malfunction of the equipment within the scanner room, which is a further safety risk and may also make any results useless, depending on what equipment malfunctioned [9].

Secondly, the equipment must not interfere with the scanner, as interfer- ence may introduce artifacts into the images produced or otherwise affect their quality [10, 11], making the images useless for analysis.

(23)

1.2 The MRI compatible actuated rubber hand

Figure 1.1b presents the basic project idea, and this project’s goal was to bring the concept one step further. It was desireable that the artificial hand be more independent of the positioning of the subject’s hand, so that new experiments can be designed in the future for further research into the areas of body ownership and agency. Remote control by the subject meant that sensors were needed to sense the movement of the subject and then transmit it to the artificial hand.

The actuated rubber hand system needed to be MRI compatible to allow for fMRI studies. Figure 1.2 gives the system overview, as was planned from the beginning, for the actuation of a single finger.

Figure 1.2: The general system as planned. Source: Sofia C. Fr¨oman, BBS Lab.

1.3 Applications

The main application for the actuated rubber hand is for the BBS Lab to use it directly in experiments. The research performed by the BBS Lab will in turn yield a better understanding of the brain’s functions, which can then be applied in many areas, such as prosthetics and telerobotics.

In the case of prosthetics, the artificial hand in this project will involve the same problematic areas as those in actuated prosthetics, such as movement, material constraints and so on. The sense of agency is also important in creating a unified body image that includes the prosthetic hand [12]. That is also a core component within telerobotics, as the sense of immersion is important for feeling that a remote reality is being manipulated by the self. The problem of

(24)

immersion shares the same difficulties in equipment as those used for studying agency, such as lag in sensory feedback or asynchrony between the intention of a movement and actual movement (that is, unintended time and position discordances). In the case of this project, the master and slave mechanisms have similar kinematics [13], in which the master is the equipment transferring signals from the subject’s movement, and the slave is the equipment receiving the signals and moving in accordance to the signals.

Studying body ownership is also vital for further research in dysfunctions of body image. Dysfunctions in body image can make a sufferer have difficulty in participating in society in a meaningful way. There is speculation that a disturbance in body ownership may be an underlying cause for schizophrenia [1]. A similar tangent of this research is investigation for sufferers of phantom limb syndrome, in which an amputee can feel that their missing limb is painfully paralyzed.

As brain research is a specialized area, there are no commercial actuated rubber hands as of the writing of this thesis. A succesful prototype, that can lead to a reliable product for use in experiments, can be of interest for other neuroscience labs. A market for small-scale manufacturing of customizable actuated rubber hands is concievable, and so a standardized production model for the manufacture of actuated rubber hands can be an interesting development for this project. Solutions that have been used in experiments in place of a physical actuated rubber hand are presented in the State of the Art.

(25)

Chapter 2

State of the art

This section is organized into three parts. The first part discusses current solutions for studying body image using variations of the rubber hand illusion.

The second part discusses the components that are available for constructing an actuated rubber hand. The third part discusses MRI compatible systems and MRI compatible components that currently exist.

2.1 Current solutions for moving rubber hands

The set-up designed by the researcher Andreas Kalckert of the BBS Lab is presented in section 1.1.3. It can be argued that an artificial hand belonging to the real world may create a stronger sense of ownership than an artificial hand in a virtual world. However, alternative solutions have been used in previous experiments, and are presented here.

2.1.1 A real hand

A simple solution for a moving rubber hand is for the researcher to use their own hand, and mimic the subject’s hand. Figure 2.1a shows an example of an experiment using the real hand approach. The researcher can control whether the subject S sees their own hand or the researcher’s hand by changing the configuration of a series of mirrors.

The task in this particular example was for the subject to draw a straight line. The subject would either see their own hand or the researcher’s hand, so when the subject was seeing the researcher’s hand, the researcher could make the subject see a movement that did not match the actual movement of the subject’s hand. The subject would then try to compensate for the deviation, convinced that the moving hand they were seeing was their own [2].

(26)

(a) (b)

Figure 2.1: Ways of performing the moving rubber hand illusion. Source: (a).

[2] and (b). [14].

2.1.2 Virtual reality

Virtual reality opens a span of possibilities, as theoretically any object can be designed within the virtual world. A set-up to test this was done in the study in [14]. Figure 2.1b shows how the illusion is performed using a ’rubber hand’

projected on a screen [14].

The subject’s real arm was hidden behind a screen, as with the original rubber hand illusion. The projected rubber hand was adjusted so that the subject felt that the placement was ’correct’, that is, projecting out of the subject’s shoulder. A virtual ball would then stroke the arm simultaneously as the subject’s real arm was stroked. The subject was then told to close their eyes and place a piece of blue-tack where they felt their right hand was, measuring proprioceptive drift. The conclusion of this study was that the virtual hand illusion produced similar results to the rubber hand illusion, and so both can be used interchangeably. Although the experiment did not study agency, in the virtual world a hand can be made to move.

2.1.3 Feedback video

With the feedback video study, a video of the subject’s real hand was recorded, and the recorded video clip then projected back to them with modifications. The lag was as little as 20ms [15], so the subject did not notice the time discrepancy, as that is well within the maximum lag allowed of 300ms as specified by the study in [16]. Figure 2.2 shows a sample of what is possible with feedback video.

(27)

Figure 2.2: Feedback video rubber hand illusion. Source: [15].

2.2 Actuated rubber hands

To create a system that used a physical artificial hand as a rubber hand, several areas, including robotics and prosthetics, can be used to put the actuated rubber hand system together.

2.2.1 Hands within robotics

Anthropomorphic robotic hands are essentially actuated hands with human morphology. However, to date there are no commercially available MRI-compatible anthropomorphic robotic hands, and they are generally expensive. Additionally, robotic hands are often designed with the goal to mimic human grasping and handling of objects. In contrast, the actuated rubber hand can have a greater error tolerance, as the movement needs to be accurate enough for the sense of agency to manifest, but not for grasping and manipulating objects. The difference in goals also means that designers of humanoid hands may take liberties in the design in terms of human morphology, as can be seen in the examples presented in this section.

The Shadow hand

The C5 model of the shadow hand uses 40 pneumatic actuators configured in antagonistic pairs. Figure 2.3a is an image of the shadow hand. The designers of the hand, Shadow Robotics, have put effort into mimicking several grasping techniques of the human hand, but the shadow hand does not mimic the morphology of a human hand completely, as all the fingers are of the same length.

The price is in the order of magnitude of 100k EUR [17].

(28)

(a) (b)

(c) (d)

Figure 2.3: A selection of robotic hands. Source: (a). [18], (b). Sofia C.

Fr¨oman, BBS Lab, (c). [19] and (d). [20].

The ¨Orebro hand

The Center for Applied Autonomous Sensor Systems at ¨Orebro university [21]

has a modified version of the Sheffield hand [22]. The custom-made robotic hand was manufactured by Elumotion Ltd [23]. The hand has twelve Degrees of Freedom (DOF), and incorporates different types of actuators and actuation methods. There are also passive elastic elements for guiding the movement of the fingers. As it was not in use at ¨Orebro university during the course of this project, it was borrowed and brought to the BBS Lab for study. Figure 2.3b is an image of the ¨Orebro hand.

(29)

iCub robot

The iCub robot is a European collaborative project [19]. The idea behind the iCub project is to study cognition and learning, based on the idea that “ ...

many of [the] basic skills are not ready-made at birth, but developed during ontogenesis”. And so “RobotCub aims at testing and developing this paradigm through the creation of a child-like humanoid robot” [24]. The entirety of iCub is open source, including all CAD designs, although the 3D models were created by Pro-Engineer [25], and can only be opened with Pro-Engineer viewers. As there has been much effort put into the similarity of the movement of the iCub to the movement of a child, there was potentially a lot of material that could have been used in this project. The price of the hand and forearm is about 75,000 EUR. Figure 2.3c is an image of the iCub grasping a ball.

KTHand

The KTHand has been developed at the Royal Institute of Technology (KTH, Stockholm), which aims to develop autonomous grasping techniques for robots.

The hand has been designed to appear humanoid, and mimics the grasping mechanics of a human hand. The project goal was to create a robot hand that was autonomous but at a low budget. It has only two fingers and a thumb [26]. The fingers are underactuated, and also make use of leaf springs to bring the fingers back to a neutral position. This minimizes the number of actuators [26, 20, 27]. Figure 2.3d is an image of the KTHand grasping an apple.

2.2.2 Actuated prosthetics

Prosthetics that are discussed here are exclusively upper-limb prosthesis. Pros- theses come in two variants, cosmetic prostheses that do not move, and actuated prostheses. The former are often used as rubber hands at the BBS Lab, as cosmetic prosthetic hands are designed to look like human hands. The actuated prosthetics sometimes have an appearance similar to a human hand, and use myoelectric sensors (that sense the change in muscle electrical potential when a muscle contracts) so that the user can send signals to the prosthetic device.

It is often important for the patient to have a prosthetic hand that is close to the human hand’s morphology [28]. Discussions with the prosthetics engineers at the Ortopedteknik (OT) Center revealed that there is a trade-off between aesthetics and functionality, as a cosmetic prosthesis is often not as functional as one that is designed for functionality [Ortopedteknik AB, personal commu- nication]. The lack of functionality has very much to do with the prosthetic glove, the outer silicone layer that gives the prosthetic it’s realistic appearance.

Skin, as it turns out, is a difficult material to replace. Synthetic materials’ lack of flexibility impair the movement of actuated prosthesis.

Actuated prostheses have similarities with robotic hands, except that prosthetic devices employ much simpler actuation and control mechanisms and are therefore more reliable and robust. For example, most commercial prostheses

(30)

only employ the pincer grasp with one DOF [28, 29]. Gears are used to transfer the movement of the motors to the fingers. Furthermore, the number of possible input signals from the myoelectric sensors is limited, as the possibility for amputees to generate signals on their stump varies between amputees, depending on the nature of the stump and amputation in general.

The FLUIDHAND

The 3rd incarnation of the FLUIDHAND was actuated using water [28]. Figure 2.4a displays the FLUIDHAND.

(a) (b) (c)

Figure 2.4: (a). The FLUIDHAND. (b). FLUIDHAND bellows. (c). The ultralight hand. Source: (a). [28], (b). [28] and (c). [29].

Pressure was generated using a pump located in the metacarpus, with ade- quate flow rate, high pressure, low weight and size. Custom-made valves were produced to maximize flow rate while maintaining a small size and weight. The FLUIDHAND used bellows to actuate the joints directly. Figure 2.4b shows the two different sizes that were available for the bellows. Figure 2.5 shows some of the configurations possible with the FLUIDHAND, which looks very natural [28].

The ultralight hand

The aim of the ultralight hand was to create a hand that was very light compared to current commercial prosthetics [29]. Figure 2.4c shows the ultralight hand.

Like the FLUIDHAND, the ultralight hand used fluid for actuation. In each joint there was an air pocket, and the joint was actuated by the expansion of the compressed air within the pocket when pressure was applied. Each joint also used passive elastomeric spring-elements to extend the joint, thus only one actuator was needed per joint. The distal-intermediate and intermediate- proximal joints were coupled, similar in operation to human hands.

(31)

Figure 2.5: The FLUIDHAND with a prosthetic glove. Source: [28].

2.3 MRI compatibility

The rubber hand actuation system needed to have MRI compatible transducers. The criterion of MRI compatibility raised additional budget concerns, as the market for MRI compatible actuators and sensors is very specialized, typically making the components expensive. This is because transducers in mechatronic applications often use electrical or magnetic fields to actuate or to transmit signals. Furthermore, they often contain ferro-magnetic materials. Both these issues mean that mechatronical transducers will interfere with the MRI scanner and vice versa [11]. These problems can be solved in several ways, and solutions are discussed in this section. Mechatronical components with MRI compatibility in mind are presented, followed by methods for testing MRI compatibility.

Finally, a study that underwent the full process of designing a device to be used in a study involving brain scans, using fMRI, is presented.

2.3.1 MRI compatible systems

MRI compatible solutions

Actuators that can be used in MRI compatible environments include piezoceramic motors [11]. Piezoceramic motors are compact and can produce forces up to 10N , reaching speeds of 12.5mm/s. Ultrasonic motors (based on piezoceramic materials) have also be used in mechatronical systems where MRI compatibility was an issue [10, 7, 8].

(32)

Fluid actuators, such as pneumatic and hydraulic actuators, have many ad- vantages for use in MRI scanner rooms. Fluids such as air and water are MRI compatible. With pneumatic actuation, flexible tubes can be used that allow for free placement of components within the scanner room. Air leakage is not problematic, as no return path is needed for the air [9]. However, pneumatic actuators are recommended for use in “bang-bang” type of movements, that is to say movements between discrete positions, as fluids are difficult to control [7]. There is also some latency from reference signal to actuation, due to the nature of fluid actuation. Further, the actuator casings need to be made of MRI compatible materials.

For position feedback, MRI compatible encoders have been developed of non-ferromagnetic materials [11], using optic fibers instead of wires for data transmission to outside the scanner room. Other options are the use of optic encoders for measuring angles [6]. Potentiometers made of non-magnetic materials are available and can be used for position feedback [8].

Shielding MRI non-compatible objects

If MRI non-compatible actuators must be used, the actuators may need to be located outside of the MRI scanner room. The actuators then need to have some mechanism that translates to movement to within the scanner room [7].

Another solution is to simply shield the non-compatible devices within the scanner room. Additionally, the scanner is most sensitive to magnetic distur- bances within the bore, meaning that electronics and such can be placed within the scanner room, but far away from the bore. The chosen solution for any project depends on many factors that need to be taken into account.

2.3.2 Testing for MRI compatibility

Due to the niched market for MRI compatible systems, some components may be MRI compatible without being specified as such. The components can then be tested for MRI compatibility. Testing can be more or less rigorous, and often depends on the size of the object being tested. Smaller components, such as sensors with questionable solder pads, can be tested simply by placing the object in the scanner and looking for the presence of artifacts in the images produced.

For bigger objects, or those that need substantial amounts of magnetism to work (such as DC motors), call for the use of statistical image processing to calculate the amount of distortion in images caused by the object. Typi- cal statistical measurements include signal to noise ratio (SNR) and standard deviation (SD) analysis on images, comparing the produced images with so- called phantom (“blank”) images. However, tests with phantom images may not always reveal artifacts, and images may still become distorted [7].

(33)

2.3.3 Designing an fMRI compatible device

An example for designing an fMRI compatible device in a systematic way is given in the study in [10], where a device with two DOF and position feedback was developed that was to be used in an fMRI study. SNR and SD analysis were used to see the effect of various components on the images from the scanner.

The device was designed according to these tests.

Materials tested were brass and aluminium, neither of which produced sta- tistically significant artifacts on the images. Two different actuators were also tested, a DC motor and an ultrasonic motor. The ultrasonic motor did not affect the image quality, neither when turned off or when loaded. However, the DC motor greatly affected the image quality. Two potentiometers were also tested, one that contained a carbon conductive element, and one that contained a plastic conductive element. The former proved to be MRI non-compatible, while the latter proved to be MRI compatible.

The MRI compatible potentiometer was used for position feedback of a linear mechanism, and the ultrasonic motor was used for actuation within the scanner.

Electronic cabling was heavily shielded and kept at a defined distance from the scanner, while still within the same room.

Finally, tests were performed with subjects stimulated by the device. It was concluded that there was no significant difference in the images of the brain recorded whether the device was present or not.

(34)

(35)

Chapter 3

Design specification

Due to the encompassing nature of the project, a design specification was made for the minimum requirements for completion of this thesis. Requirements included a minimum of one finger had to be actuated, and signals from the subject did not have to be included. The design specification in bullet-point form can be found in Appendix A.

3.1 Investigation

Previous solutions to similar projects were studied to reduce design time. Of the several actuation alternatives presented in the State of the art in Chapter 2, actuation by Pneumatic Air Muscles (PAMs), in conjunction with tendons, was chosen, see section 3.4. Materials and components used in the actuated rubber hand system were chosen so as to satisfy the criterion for MRI compatibility, with the exception of the PAMs, where MRI non-compatible versions were initially implemented for testing purposes. The position sensor chosen for the artificial hand had not beeen MRI tested and so also may not have met the MRI compatibility criterion. As pneumatics can be difficult to control ac- curately due to the compressibility of air, the PAMs’ function and control were investigated, both through previous litterature and empirical testing. Different MRI compatible sensors were also investigated.

3.2 Design of the hand morphology

The artificial hand had to have the same morphology as a human hand. Design of the details of the artificial hand, such as joints and possible tendon placement, had to be based on existing designs, such as those presented in State of the art in Chapter 2. Space and positioning of the actuators and chosen sensors also had to adhere to hand morphology. Using actuators in conjunction with tendons allowed the actuators to be located at a distance from the actuated area.

(36)

The movement of the artificial hand had to mimic a real human hand, and as such a simple controller was implemented for testing the movement. Initially, a Shadow Pneumatic Control Unit (SPCU) was to be used for testing the PAMs.

However, the SPCU at the BBS Lab was required for another project and a custom-made Pneumatic Control Unit (PCU) was built instead.

It was taken into account that movement could be impeded by the use of a cosmetic prosthetics glove that was slipped over the artificial hand’s endoskeleton. The base requirement was for only one finger to be designed and actuated.

3.3 Choice of material and manufacturing

MRI compatibility was the chief requirement behind any material choice in the system. Exceptions for the MRI compatibility requirement concerned parts that could easily be replaced with MRI compatible versions. The artificial hand was not placed under any greater strain than its own movement, and as such it did not need to be as robust as the robotic hands presented in section 2.2.1 or the prosthetic devices presented in section 2.2.2. However, the material used for manufacturing the artificial hand still needed to withstand the forces from the actuators moving it. Therefore, an endoskeleton finger was manufactured by 3D printing. Manufacturing took only two days including delivery time, but the schedule remained flexible to account for any possible delays.

3.4 Actuation choice

The actuation of choice was PAMs, as PAMs were already being used in other projects at the BBS Lab. There was also a compressor available at the BBS Lab. PAMs can be easily made to be MRI compatible and thus can fulfill the criterion for MRI compatibility. PAMs have also been used in similar projects, such as the Shadow hand, in which PAMs are best used in conjunction with tendons. Additionally, PAMs could have been implemented in different ways, such as in combination with a spring element to bring the fingers back to a straightened position, like the KTHand [26] and the ultralight hand [29]. Springs could then have been replaced by rubber bands or any other MRI compatible alternative with similar characteristics. However, the antagonistic pairs solution was implemented, allowing for more direct control over the movement of the hand. Furthermore, an unactuated PAM could act as a spring counteracting the movement of the actuated PAM. The finger that was actuated moved with an up and down movement at the knuckle, mimicking a tapping motion.

3.5 Simulation

Simulation could have been a useful step for making the initial designs of a control algorithm, based on the artificial hand design and dynamics of the system.

(37)

movement transmission was done through tendons. The kinematic study was done in Matlab, the basis of which can be found in Appendix F. Existing static and dynamic models of PAMs were implemented from the studies in [30], [31]

and [32]. The complexity of the simulations was weighed against usefulness, as the project had a deadline.

3.6 Software design

The Robot Operative System (ROS) platform [33] was used for controlling the artificial hand. ROS was chosen because it is an open source platform and is already being implemented by the BBS Lab in other projects. Software design included the implementation of a simple controller. It needed as little lag as possible, as the minimum time discrepancy allowed is 300ms between initiative for movement by the subject and the actual movement of the artificial finger [16]. It was not specified if the 300ms discrepancy spans over the entirety of the movement or counts only for the initiation of movement from a standstill.

3.7 Electronic design

The electronic design originally had little priority in this project. However, a PCU was built, and electronics were needed to interface it to the PC. Sensors and pneumatic valves (for actuation) and a Labjack U6-Pro A/D converter were all used. Basic electronic components for interfacing the components with each other were available for the task.

3.8 Project specifications

The project was an initiation project, and is to be developed further. Therefore, there were requirements for the documentation and the tools used. Tools that were used had to be available to the BBS Lab to allow access and manipulation of files produced. Ideally, the tools used were to be fully open source. An open source CAD program was used for the design of the finger, and Matlab was used for studying the kinematics of the finger and implementing PAM models.

Matlab is not open source, but is available to the BBS Lab. Matlab is also the de facto standard in many computational sciences, and open source clones (such as Octave) are available. Code can be ported to open source numerical solvers as Matlab-specific toolboxes were not used.

Likewise, all relevant material produced through this project was to be open source and published online. Publication was to be on github, an online resource for coordinating and publishing collaborative open source projects [34].

(38)

(39)

Chapter 4

Planning

This chapter presents and discusses the planning that was involved in this project, and gives an overview of the time schedule as planned from the start.

Further, resources that were available for this project are presented, followed by a risk analysis.

4.1 Design process

This section presents the order of the steps in the design process based on the design specifications given in Chapter 3. The development process used resem- bled the Agile process implementation, and was characterized by a modular approach. Each step was to be finished before the beginning of the next step, analogous to sprints used in the Agile model SCRUM. However, the model was fitted for the work of one person. There was much face-to-face communica- tion with researchers, in which core requirements were focused on primarily but designed in such a way as to be open for further development.

As the project had such a broad scope, simple solutions were preferred over complex ones in the different areas, so as to complete the system as a whole.

The steps for the base requirements were ordered by level of priority and logical sequence, and are listed below. As the first step had the highest priority, it was dealt with first.

1. Mechanical design of the artificial hand.

2. Investigation of PAM actuation and control.

3. Design of a control algorithm, including simulations if relevant.

4. Integrating PAMs with the manufactured hand.

5. Control signals from the subject may be calibrated according to the control algorithm.

(40)

4.2 Design schedule

This section presents the schedule that was implemented as part of the planning process, based on the design process from section 4.1. The activities are explained briefly. The schedule can be seen in Figure 4.1.

Figure 4.1: Design schedule in weeks. Source: Sofia C. Fr¨oman, BBS Lab.

Introduction: The introductory phase included the study of ROS, literature searches, investigation of open source alternatives to licensed design tools and study of the state of the art. The specifications were also established and are given in Appendix A.

CAD endoskeleton design: The CAD design of the hand was allotted eight weeks, as can be seen in Figure 4.1. The design time included the time needed to become acquainted with the CAD tool of choice (see section 4.3.2), which took about two weeks.

Study of actuators: As PAMs were chosen as actuators, they were studied during this stage. This study included implemented models of PAMs (both static and dynamic) from several studies, mostly [32, 31, 30]. The beginnings of the PCU were also developed during this stage so that the PAMs could be studied empirically.

Study of system: This stage was used for study of the kinematics of the artificial hand, which was used during the design on the endoskeleton. Although it would have been ideal if the study had incorporated the kinematics of the en- tire hand, the study was limited to one finger.

Model design/simulation: This stage was used for implementing a kinematic study of the index finger of the artificial hand in Matlab, and for studying and implementing PAM models from the literature. This stage was heavily entwined with the actuation study phase.

(41)

Design of control algorithm: In this stage, different control algorithms were investigated and one was implemented.

Assembly of areas: The assembly of areas depended largely on the completion of previous phases. The initial finger had to be manufactured first.

Software design: The software was implemented with ROS and included the chosen control algorithm. The software could only properly be developed when the hardware was assembled.

Documentation: Before the outset of the documentation phase, the project was considered finished. However, continuous documentation was carried out over the course of the twenty weeks. The documentation includes this report and material published online.

Evaluation: Evaluation periods are a logical inclusion in a step-based design process model, and allowed for planning.

4.3 Resources

This section introduces the resources that were available to the project, either prior to start or acquired during the course of the project. A good deal of the resources were already available, as there were requirements to incorporate elements of the project into other projects at the BBS Lab. Such elements would be the software, choice of A/D converter and PAMs.

4.3.1 Budget

One of the motivations behind this project was that there are no actuated rubber hands commercially available, and the closest alternative, anthropomorphic robotic hands, are expensive. The idea was to build an actuated rubber hand system at a reasonable price. An exact budget was not specified, so any eco- nomic decisions were discussed with the project supervisor.

4.3.2 Design tools

Open source

Licensed tools that are available for students at KTH include Computer Aided Design (CAD) programs such as AutoCad, Inventor and SolidEdge. However, these CAD programs are not available to the BBS Lab and so were not chosen. A project that the U.S. military has been developing is an open source cross-platform Constructive Solid Geometry (CSG) CAD tool, called BRL- CAD, whose project started back in 1979 [35]. BRL-CAD produces its own database files, called G (for Geometry) files. The G file type is not an industry standard, but the software comes bundled with several conversion programs,

(42)

allowing conversion to industry standard file types such as the STL file format.

BRL-CAD is available for all major operating systems such as UNIX (Linux, Mac OS X, Solaris, IRIX and FreeBSD) and Windows.

Another open source tool is ROS, a meta-operating system for robots, that runs on Linux [33]. A beta version for Windows is available. Mechatronic systems can then be controlled with a PC in conjunction with an A/D converter, ideal for prototyping stages. Data can also be recorded or displayed graphically during run-time, and as ROS has a parallel architecture, it was possible to monitor the behaviour of the system without affecting its speed. It supports both C++ and Python as programming languages. The Diamondback version was used for this project.

The A/D converter of choice, the Labjack U6-Pro module (see section 4.3.7), has open source configuration software. Example programs are on the online Labjack github repository.

Non-open source

The numerical solver Matlab is licensed both to KTH students and the BBS Lab. Matlab is not open source, and using it in the project made it no longer completely open. Matlab was chosen for the reasons presented in section 3.8.

Some freeware tools that are not open source but were useful are netfabb Studio Basic and Eagle. netfabb Studio Basic was needed for viewing STL files after the conversion from the G format to STL. The tool was needed because BRL-CAD cannot open STL files, despite being able to produce them, and neither can any of the licensed CAD programs available to KTH students.

Viewing the converted STL file allowed for any errors, possibly created through the conversion process, to be scrutinized. Eagle also has a freeware edition, and is meant for designing and drawing electronical schematics. Both freeware programs are available for operating systems such as Linux, Mac OS X and Windows.

4.3.3 Shadow Pneumatic Control Unit

The SPCU [36] had been purchased by the BBS Lab for use in their projects.

It can only support four PAMs at a time, meaning that it could only actuate two fingers at a time, since the antagonistic actuators configuration was chosen. Furthermore, SPCU documentation is lacking, as the datasheet does not specify how to communicate with the module, and the control section is only a description with little detail.

4.3.4 Actuators

PAMs from Shadow robotics were purchased prior to the start of the project.

Many projects use PAMs as actuator choice, as fluid actuators have a high re- liability [29]. The cheaper variants of the PAMs were not MRI compatible, but were used in this project as the PAMs could be replaced with MRI compatible

(43)

versions. Although PAMs can easily be made by hand [37], due to time constraints PAMs were bought in instead. There were enough PAMs available for the actuation of several fingers.

The basic operation of a PAM is that it will contract when air pressure within it increases. PAMs are made up of an elastic tube (or bladder) encased in a braided sheath. A pneumatic tube is attached and continues into the elastic tube, with the ends tied tightly. When no air pressure is applied, the PAMs can be stretched to their full length, in which there will be tension. The property of tension with a stretched PAM is similar to that of a streteched biological muscle.

When the air pressure increases inside the bladder, the bladder expands and presses onto the inside of the sheath, causing the braid pattern to shift. Because the diameter of the PAM increases, the PAM becomes shorter [32, 31, 30]. Figure 4.2 demonstrates the operation of a PAM.

(a) (b)

Figure 4.2: Operation of a PAM. Source for both: [18].

The PAMs used had a braid diameter of 6mm, a stretched length of 150mm, a 4mm tube fitting, and were able to give a force of 30N at 3.5Bar [18]. The valves chosen for controlling the pressure within the PAMs were 3/2 valves (V114-5MOU-M5) from SMC. The valves were chosen as they are readily available [38] and are also the fastest that could be found for a reasonable price.

The valves take 5ms to open and close and can be run at up to 20Hz. Further specifications can be found on the datasheet, available online.

4.3.5 Tendon material and placement

The tendons are core components of any experimental robotic hand. A regular type of fishing line was originally used. However, it had some elasticity, which could have made for inaccurate movement transfer. It was also difficult to tie and it occasionally broke.

On a similar note, the iCub has had issues with tendons breaking and have had to change tendon material [39]. Discussion with those working at the Center for Applied Autonomous Sensor Systems at ¨Orebro university brought attention to the fact that tendons can break due to high friction. Tendon attachment onto the fingers was also crucial, as disadvantageous torques about a joint could cause the fingers to lock into a fixed position.

Design av en rörlig gummihand för användning i miljöer med magnetisk resonans

Design of an actuated rubber hand for use in magnetic resonance imaging environments

SOFIA CRUZ-FERREIRA FRÖMAN

Design of an actuated rubber hand for use in magnetic resonance imaging environments

Sofia Cruz-Ferreira Fröman

Contents

List of Figures

List of abbreviations

Notations

Chapter 1

Introduction

1.1 Background

1.2 The MRI compatible actuated rubber hand

1.3 Applications

Chapter 2

State of the art

2.1 Current solutions for moving rubber hands

2.2 Actuated rubber hands

2.3 MRI compatibility

Chapter 3

Design specification

3.1 Investigation

3.2 Design of the hand morphology

3.3 Choice of material and manufacturing

3.4 Actuation choice

3.5 Simulation

3.6 Software design

3.7 Electronic design

3.8 Project specifications

Chapter 4

Planning

4.1 Design process

4.2 Design schedule

4.3 Resources