Lab Blog RSS

Sarcos Parts

ornstein.mike@gmail.com — 2010-09-29T11:31:47-04:00

There was a significant failure on the Sarcos Humanoid robot.

I redesigned the part in SolidWorks and had it sent out to be CNCed at ZeroHourParts. Here are some comparison shots.

New Semester

ornstein.mike@gmail.com — 2010-08-30T15:05:56-04:00

Things Chris Said:

Meet with Steven Collins, Mech E prof. http://www.andrew.cmu.edu/user/shc17/
Monkey robot: swing from tightropes...
eventually robots that use escapement (swing/letgo)
robots that run and glide
robots that fly!
Spec out Maxon motors
Spec out dynamixel servos- email links to order to Chris directly
Find strain gage kit

Things Sid said:
Bon Voyage!

PWM and Position Control with Pneumatics

ornstein.mike@gmail.com — 2010-07-26T23:43:54-04:00

In an attempt to gain finer control of pneumatic actuation, I attempted to employ PWM to a single solenoid valve. The modulation would occur between fill and vent. For the pneumatic muscles, the maximum response rate from the solenoid valves is inadequate to gain position control. The valves are limited to 20ms average cycle time, but oscillating at that rate yields considerable vibrations. In fact, the current pneumatic muscles fully contract and relax in that cycle time.

After realizing that a modulated signal does not work well with the pneumatic set up I went to a two stage solenoid valve arrangement. In this configuration one solenoid valve is responsible for filling the actuator and the second is responsible for venting. The four possible states of these two binary valves follow:

fill closed/vent closed: All flow is stopped; the actuator maintains current state.
fill closed/vent open: Flow out of the actuator; the actuator relaxes.
fill open/vent closed: Flow into the actuator; the actuator contracts.
fill open/vent open: Useless, the actuator maintains some equilibrium state.

By linking a pair of sensors to the control of the fill and vent solenoid valves, it is possible to drive an actuator to a certain position/force and then maintain that position/force without bleeding off excess air.

One limitation of the dual solenoid setup is the necessity for an activation pressure in the vent solenoid. This requires that there be a pressure above ~3psi in whatever vessel is meant to be vented.

Inflatable Tasks and Control

ornstein.mike@gmail.com — 2010-07-16T17:54:58-04:00

At Friday’s lab meeting I demonstrated the inflatable grasper performing the two tasks I have video of: shaving and toothbrushing. The major feedback I received about the videos was that the model should appear more enthused and in the shaving task lather should be applied. I will act on this feedback. Additional tasks were also discussed, designed to determine the versatility and weakness of the current gripper design. Some of the ones that stuck include:
using a washcloth
applying makeup- eyeliner, mascara, lipstick etc.
wielding electric razors and scissors

I also received feedback on my sensor choices. It was agreed upon that the most ideal measurement to work from would be internal pressure of the cuff. The issue with this method of feedback was the possibility for plastic deformation which causes pressure values to become unlinked from the exerted force/volume of the tube. With creative link and cuff construction, the plastic deformation can be avoided and thus pressure can be a useful means for feedback.

My original solution to this problem involved making a tactile skin. I will still construct this skin by embedding forces sensors in the polyurethane film. According to Chris, feedback from these particular sensors has proved unreliable in the past, and I should not expect accurate feedback from them.

Pneumatic Joint Actuation

ornstein.mike@gmail.com — 2010-07-12T01:05:07-04:00

I have been experimenting with various methods of joint actuation. I have so far investigated three possible routes:
one-way link with springs
two-way bellow
pneumatic muscles

The one-way link with springs consists of a single airbag located at the joint and springs (silicone tubing in this case) the run between the two jointed segments. The airbag extends the joint and the springs return the joint to its contracted position. This particular configuration is best suited to actuating a joint between ~45º and 180º. The 45º lower bound has only to do with the interfering geometries of the two jointed segments. The primary limitation of the design lies in the fact that the maximum force applied in the return direction is limited by the spring. The extension force is much greater by comparison.
The two-way bellow actuator is constructed of a 3” diameter polyurethane bellow. I trimmed this bellow so there were 3 convolutions, enabling a maximum inflated extension of approximately 8”. The bellow can be actuated using both positive and negative pressure (vacuum). This enables a joint that is pneumatically actuated in both directions using a soft membrane. This joint is best suited for actuating between ~45º and 180º. Again, the ~45º lower bound has to do with link geometry, not the actuator. The main drawback of this actuation type is the various ways in which the bellow can collapse. Instead of consistently returning to its flat configuration under vacuum, the bellow collapses on itself in place. Utilizing a more stable/rigid bellow could solve this problem.
Pneumatic muscles have not yet been integrated onto an inflatable link. However, the method for actuating a joint can be applied such that a joint was an increased range of motion (passed 180º) and also additional degrees of freedom in a single joint. The maximum displacement of pneumatic muscles is their limiting factor; a 6” muscle will contract to 4.5” at best yield only a 1.5” displacement. The force output from each muscle is very high compared to the previous forms of actuation.

Solenoid Control

ornstein.mike@gmail.com — 2010-07-13T08:34:56-04:00

I have successfully been able to control the actuation of a pneumatic solenoid over USB. Using a laptop, an Arduino, an external power supply, compressed air and a pneumatic solenoid I am able to inflate and deflate links on command. By switching open and closed the exhaust port of a Venturi vacuum generator with a solenoid it is possible to switch between positive and negative pressure (fill and vacuum). With a form of feedback either at the joint or in an end effector, actuation of any vessel can be safely automated.

Control Circuitry:

solenoid_driver.gif

Code:

//specify the digital ports to which the solenoid circuits are connected
int solenoid1 = 2;
int solenoid2 = 3;
//set the threshold analog value to switch, 0-1023
int threshold = 650;

void setup() {
// initialize the serial communication:
Serial.begin(9600);
//set up digital i/o pins
pinMode(solenoid1, OUTPUT);
pinMode(solenoid2, OUTPUT);
}

void loop() {
// send the value of analog input 0:
Serial.println(analogRead(0));
int sensorValue = (int)(analogRead(0));
delay(5); //update interval in ms
if (sensorValue > threshold){
digitalWrite(solenoid1, HIGH);
digitalWrite(solenoid2, LOW);
}
else{
digitalWrite(solenoid1, LOW);
digitalWrite(solenoid2, HIGH);
}
}

Arduino

ornstein.mike@gmail.com — 2010-07-07T00:10:20-04:00

The Arduino has arrived! I chose the Arduino development platform due to its ease of interfacing with the components necessary for control of the arm and its massive user support base. The Arduino is based on an Atmel ATmega328 microcontroller, has a USB interface for programming, can control servos and integrate with sensors.

The current setup can take in analog values from our bend sensor and display the input as a graph.

Here is a photo of the current setup:

The bend sensor is a variable resistor. I have an 18kΩ resistor in series with it to create a voltage divider. The analog reading comes from the node connecting the bend sensor to the resistor (orange wire). The bend sensor has a resistance that varies between 8kΩ and 50kΩ.

Inflatable Shaving and Toothbrushing

ornstein.mike@gmail.com — 2010-07-03T00:09:00-04:00

The two following videos demonstrate some of the capabilities of the inflatable arm.

Inflatable Shaving:

Inflatable Tootbrushing:

Link on a Stick

ornstein.mike@gmail.com — 2010-07-02T17:23:09-04:00

In response to previous suggestions posed in the System Review post, I have made progress in integrating a solid attachment point onto typical inflatable links.

The link is attached to a malleable steal rod.

A detail of the insertion point of the rod.

This method is highly successful in constraining the link translationally, but is weak in resisting torsion. To solve this problem I could employ a double prong stick. Doing so would maintain the translational rigidity and add rotation rigidity. Following is a short video demonstrating the effectiveness of this approach to link interfacing:

Wrist Update

ornstein.mike@gmail.com — 2010-06-30T13:49:20-04:00

Puppeting/Control
Relocated tendons

In the original wrist and arm the top two tendons ran parallel to the longitudinal axis of the arm. To gain better control of the wrist I crossed the two upper tendons.
In addition to crossing the two upper tendons

Manual movements

I currently have a single knot that links all three tendons.
By pulling the knot in any direction I got a correlated response at the wrist.

Manual pressure control

To achieve a grasping hold on an object inside the hand, I manually inflate the cuff.
I do this with a bicycle pump and vent the cuff by disconnecting the air tubing from the check valve.

Meeting
I noted a desire to see this particular arm perform more tasks as is. I suggested making another version that was prettier, computer controllable or both, but it was suggested that I focus on the current setup.

Video
A video of the current state of the wrist follows.

Modular Links

ornstein.mike@gmail.com — 2010-06-29T14:34:01-04:00

Modular links are a bit of an offshoot from the path I’ve taken so far in my research. This concept involves having many identical inflatable segments with two fill ports, one at either end of the link, for modular filling.

The links can be daisy-chained together so every link can have access to a single valve for inflation/deflation. The links are secured end to end in combination to form regular polygons. In my video example I created a square pyramid. Future shapes could include more circular polygons. In the long run, and interesting concept would be to inflate and deflate sections of structure to motivate locomotion or manipulation.

I have explained the concept and its possibilities further in the following video.

Wrist Action

ornstein.mike@gmail.com — 2010-06-25T11:21:18-04:00

I’ve made progress in creating an articulated wrist that connects any typical inflated link to a cuff type grasper.

The main concept simulates a human fist and lower arm. The fist can open and close to grasp an item. Its air supply is routed through the forearm. The articulation at the wrist is done with three tendons whose method of actuation is yet to be determined. Pneumatic Muscles might be the solution. Other methods of actuation are hobby servos or pneumatic airbags.

Here is a photo of the wrist assembly to date:

Pneumatic Muscles

ornstein.mike@gmail.com — 2010-06-24T14:44:11-04:00

Linear actuation is of major interest to me and the inflatable projects I’m dealing with. Having a flexible, integral component of an arm or grasper is very attractive. I think it will be possible to use pneumatic muscles in the inflatable structures.

I have purchased materials to make some small muscles. Those materials make up the following McMaster Order:Confirming your order from 6:24:2010

Pressure Components

ornstein.mike@gmail.com — 2010-06-24T13:32:43-04:00

Pressure Transducers are very expensive.

Combine a pressure transducer (analog sensor) with a proportional valve (analog output) and you have an electronically controllable pressure regulating machine. Beyond those two pricey components is the need for a microcontroller, and a DAC (Digital to Analog Converter). There’s also a significant programming overhead to tell the microcontroller how to map the signal it’s getting from the transducer into something the DAC will send to the valve.

This is all doable, but is it worth it?

In industry, this problem is solved using position controllers. They effectively bundle up all those components into a tidy box that you can plug into a computer to do the control with. These controllers cost hundreds of dollars.

I contacted Initek again, and I’ll be meeting with Joe to play with a PPC (Proportional Pneumatic Controller) valve from Mac Valves. I think this is going to work well for many purposes.

Update 6/25: I met with Joe and he was kind enough to leave me an older Mac Valve to test out. It operates at 24V and takes in a DC voltage between 0-10V (4-20mA) to control to output pressure.

The applications for all this work are many: Regulating pressure in links, (0-5psi range), regulating pressure in bellow/expansion links (0-5psi), controlling pneumatic muscles (0-60psi)...

Vacuum Testing

ornstein.mike@gmail.com — 2010-06-20T21:43:43-04:00

With my shiny new camera in hand I have been snapping pictures and video like there’s no tomorrow. Vacuum parts arrived Thursday and I spent the day cataloging and familiarizing myself with the new stuff. Overall the vacuum generator is absolutely satisfactory. I have found a few clever usage scenarios for it including the planned vacuum cup attachment as well as possible application in variable pressure links.

First are photos of the components I received:

The Venturi Vacuum Generator with fittings

The Clippard Proportional Valve: This item is much smaller than I anticipated and its airflow is lower than I was expecting. Its usefulness might be limited.

Two of the three types of vacuum cups. VC-356 (left) is made of soft, thin silicone. VC-359 (right) is made of a sturdier, but still flexible vinyl. Each have a lip diameter of about 1”.

The third type of vacuum cup: VC-B3. Each of these is made of vinyl and are about .6” in diameter.

I made a video that qualitatively demonstrates the differences between each vacuum cup when attached to a spoon.

In addition to testing with the spoon I did some minimum pressure tests to quantitively differentiate between the three cups I have on hand. In my test I lifted a 568g (1.25lbs) rubber hunk with each cup, and slowly reduced the input pressure to the venturi until the hunk fell from lack of suction. The results of the test are as follows:

Cup	Fail Input psi
VC-356	50 psi
VC-359	15 psi
VC-B3	41 psi

What I can say after looking at these numbers in addition to the qualitative testing is that the VC-359 has by far the best lifting potential granted it is able to get an initial grasp. The VC-356 cup is very soft, and despite being the same size as the VC-359 has significantly worse lifting and holding potential. The small VC-B3 lacks the flexibility of the VC-359 cup (one convolution vs. two) so is more subject to disengaging when a moment is applied to the cup. It also has worse lifting/holding performance than the VC-359 due to its smaller diameter.

My next step is to put multiple VC-B3 cups together and test their combined performance. My thought is that there is an optimal arrangement of the three cups such that they have a more desirable grasp on any of the task objects: Spoon, Fork, Toothbrush and Cup.

Inflatable Progress

ornstein.mike@gmail.com — 2010-06-18T12:45:23-04:00

I have become a master polyurethane seam welder. I wear that badge proudly. Here are some examples of my work on inflatables:

This is the flat pattern with an installed Shrader Valve off a bicycle inner tube. I will detail the installation and construction process of a similar shape in a later entry.

Under extreme inflation the airbag becomes very rigid and enspheres to minimize surface tension.

Failure mode: The seam gave out at around 5psi.

I created a very successful torus. There are many foreseeable applications for a shape like this.

Because of the valve location and the seam in the torus, the shape tends toward a triangle. Even at low pressures this shape grasper outputs a firm grip compared to a human hand.

This is my most complicated seem welding project to date. It employs three precisely cut layers of polyurethane to form discrete pockets with the advantage of being able to fill the whole apparatus via a single port.

The design called for four grasping fingers but I made a mistake when welding and ruined one of them. This is the resulting soft link based grasper.

This is a detail of the valve on the grasper. It works the same as in the torus and simple airbag above.

When held up to the ceiling lights, the translucency of each layer of plastic makes it easy to see the three distinct layers used to construct this grasper.

Making Inflatables

ornstein.mike@gmail.com — 2010-06-15T23:17:19-04:00

Monday and Tuesday included the testing of different techniques for sealing, filling and destroying inflatable structures. I was also to demonstrate a proof of concept for my big vacuum cup plans using found materials.

In my introduction to inflatables I took it upon myself to sacrifice a failed inflatable for its toy-plastic valve.

With that valve in place I achieved moderate success in creating air tight bags, using the foot pump until the bag failed. I found that I got a better seal by refreshing the teflon shield on the seam welder and reducing the temperature (to 6, the top of the red on the gauge) and performing two seal cycles on the same spot with a little cooling time in between. This is the method of sealing I have been using for all subsequent inflatable links.

By necessity I next attempted to create bladders without a valve at all because I did not want to sacrifice any other arms and had no real other options on Monday. This ended in moderate success. I was able to inflate the bladders by leaving small gaps (made in various ways) and finish the seal after the bladder was inflated.

A moderately successful example of a valveless, inflated bladder.

The deformation of the plastic due to expansion under pressure made for leaky bladders, so alternative methods will need to be utilized. In previous iterations of inflatable links, hard valves were used to try to connect to shop air. I made some attempts at expanding on this idea. One concept I am in the process of installing uses a round ‘puck’ as a back plate for a small metal valve. I think having a rounded shape might be the key to integrating on the soft material. In previous attempts a square shape was used and it consistently failed. Another fitting-based route I’m in the process of testing involves the use bicycle inner tube valves: Schrader valves. I have cut the valves from the tube and have attempted to adhere them to the plastic sheet with limited success. I require a bicycle pump to test these valves further, but having a unified fitting would be nice.

The final revision for the past two days is my simplest approach. I started this technique with a simple slit in the plastic through which I could insert the stem of the Schrader valve (about 1/4” in diameter). Failing to acquire an appropriate pump I sought to try out a simple 1/4” tube, the same sort used in the rest of the pneumatic system. I was pleasantly surprised with the ease of fabrication and the relative success of my first attempt. At first, I used a small amount of vinyl adhesive to secure the tubing to the sheet. This adhesive did not prove strong enough to keep the pressure from ejecting the tube from an inflating bladder (under relatively high pressure). I then thought to double up the fill port and try a slightly different hole-generating technique. Instead of cutting a linear slit for a round tube, I figured a circular opening would be subject to less stress. To create this round opening I utilized a hot soldering iron to melt a small hole in the two layers of plastic.

This method of construction proved to be very effective in creating a quality seal that could sustain relatively high pressures.

This image is of the bulging shape, the valve is quite effective.

The valve remained intact throughout the destructive testing.

The bladder failed along a seam instead of at the valve. That goes down as a success in my book. The bladder expanded to about double the size of the first photo before failing, reaching a gauge pressure of about 5psi. One of the many things learned in this test is that we require a much finer scale to take accurate pressure readings for this project.

Motivated by the success of the first double-ply valve, I made another bladder and shot an awful video on my cellphone so to better log my progress.

Note: I got a shiny new camera that should be getting here soon. It shoots in 1080i HD and is smaller than a deck of cards. I am excited. Feel free to check it out on Sony’s site.

System Review

ornstein.mike@gmail.com — 2010-06-11T21:17:19-04:00

Thursday consisted of finalizing the parts order for the vacuum cups. Friday was a review of the systems at Lab Meeting and a start on inflatables.

The invoice from Keystone Components follows:

Keystone6_10_10

The majority of the components on this list are from Clippard Minimatic. The Vacuum Transducer (Venturi Generator) was sourced from Air-Vac. The vacuum cups are all from Vi-Cas. Each cup is outfitted with a unified fitting to easily connect to the rest of the system. There are three different cup sizes in the order: two with lip diameters near 1” and lifting capacities near 4 pounds, and one with a lip diameter of .6”, capable of lifting much less. This smaller cup was purchased with two like cups to be used in tandem for interesting lifting arrangements. If the vacuum cup solution seems promising I will consider purchasing a dedicated vacuum pump (or compressor, depending on my satisfaction with the Venturi generator) and perhaps other vacuum cups in order to try more gripping setups.

I made the Keystone Components order with a salesman named Mike Albright. He was extremely helpful in organizing the order, was patient with my indecision and inexperience and we were able to work out what seems to be a strong system. I plan to request him specifically when next dealing with Keystone.

Friday included the usual lab meeting where I presented my system overview including a detailed explanation of the vacuum components. I got some valuable feedback that suggested a simple route that did not include using either the existing inflatable arm or continuum arm to accomplish the task I set out. The proposed baseline and a more advanced concept are detailed below:

Chris’ Depiction of Summer Goals.

This drawing followed my presentation, which can be found here:
6_11_10 Lab Meeting

In other news, I was able to solve the puzzle cube bestowed upon me by Sid. My summer work is finished.

The solved cube in all its glory.

Order Up

ornstein.mike@gmail.com — 2010-06-09T22:10:32-04:00

Tuesday and Wednesday were spent familiarizing myself with ordering the various components that are required to start testing the vacuum cup designs.

Tuesday:
Component Research: I investigated all manner of vacuum generation and came up with two viable options:

The first viable, and less expensive route is to use a Venturi type vacuum generator to create a vacuum from the compressed air present in the lab. This method of creating vacuum is somewhat inefficient, but since there is an abundance of compressed air available, I do not see efficiency being a concern at the moment.
The second viable route is to purchase a standalone vacuum unit that simply has a single port that pulls vacuum. This unit can be quite small and would just sit on a bench top plugged into an AC outlet. This option also removes the need for many of the fittings that would be necessary to connect up the first option.

I also met with Sid and presented my vacuum cup idea.

Wednesday:
I got approval to place orders directly with Lynnetta Miller.
I had phone calls with-

Clippard: I called up Clippard technical and learned about the proportional valve that I thought I would be interested in. I settled on part ET-P-10-25A0.

Intek Systems: I am in contact with an engineer or technician at Intek who has previously worked with the RI and NREC on integrated systems. I ran my vacuum system idea by him and he seemed to think it would suite the purpose of the project. He also suggested looking into a vacuum unit, which could be convenient for many reasons. He said he will be consulting his “Vacuum Guru” and be getting back to me tomorrow by email.
Keystone Components: The Clippard tech I spoke with recommended that I order through a distributor instead of online to avoid additional handling fees, and Keystone Components is the distributor listed on the Clippard site as serving the Pittsburgh region. I ended up spending a long time on the phone with them putting together a substantial order including the proportional valve, all necessary fittings, tubing and vacuum generator and cups. Once I have the quote finalized I will elaborate on its contents. I expect that to happen Thursday.
BouncerLand: At Sid’s request I also contacted BouncerLand to see if they might be capable of creating a custom inflatable arm for us. Initially they seemed very excited about the prospect of a custom order, but upon further discussion of our needs it seemed that they were not set up for the scale that we require. The company specializes in large, simple structures, but if the designer thinks the job is possible then I will receive a call back in a day or two. I am not optimistic about BouncerLand.

A New Direction

ornstein.mike@gmail.com — 2010-06-07T16:58:57-04:00

I’ve spent this weekend and today researching and designing simple, compliant and inherently safe grippers. My initial thought was to use a donut like shape that inflated to grip whatever is in the donut hole. I started by researching inflatable artificial sphincters used in the medical industry to treat stress incontinence.

Images of inflatable artificial sphincter

This solution seemed promising at first, but I ran into complexities in packaging and manufacture so I continued my search. I next researched another biologically inspired soft gripper: the end of an elephants trunk. Again, I ran into issues with creating a simple pneumatic version of this shape. I attempted to design a complex sequence of seams to press in the plastic sheet currently in the lab, but again the system felt too complex for a simple tool changing operation.

Instead of looking to nature for inspiration, I moved into the realm of industrial automation. In industry, particularly machine tools, have many interchangeable tools. The tools themselves can vary significantly in size and shape but each tools is outfitted with a unified grasping point: a collet.

Image depicting various tools in collets.

With this concept in mind I elaborated on simple clamping devices that would be compliantly mounted to an arm capable of easily interfacing with each tool. The main disadvantage of this approach was that it was not inherently safe. Making a soft, compliant gripper capable of grasping and articulating the tools again seems like a complex approach, although the automation achieved in CNC tool changing is something to admire.

This is a carousel type tool changer for a CNC mill.

My final thought is still on the industrial path, we’re definitely on to something. I think that the best bet at this point is to use vacuum bellow cups to compliantly grip each tool. Each tool would be outfitted with a sort of collet, made simply of a disc of smooth, flat plastic. This disc would allow a pneumatic bellow to quickly and securely grasp anything outfitted with that attachment point. These bellows are also entirely soft except for the nozzle at the top of the bellow. These devices also come in a wide variety of sizes, working capacities, materials and colors to compliment their very low cost. The pneumatic control for a bellow is very simple: vacuum on to grasp, vacuum off to release.

An assortment of vacuum bellow cups.

Now, we can go further with this as I have some preliminary ideas about how to (add complexity, and) increase the capabilities and functionality of the setup. Elaboration on that is planned for tomorrow.

Lab Meeting

ornstein.mike@gmail.com — 2010-06-04T13:14:05-04:00

At today’s lab meeting I gave a presentation to gain some insight on the direction of the project. That powerpoint can be downloaded here:

6_4_10 Lab Meeting

The overall impression I got was that it was not worth pursuing a complex hand design, but instead focus my energy on creating a ‘system’ that can accomplish precisely the list of tasks that were originally set out. It was determined that if necessary abandoning the concept of a wholly soft gripper would be okay.

Some ideas presented include:
Tool Changer
Nerf Robot
Iris Mechanism
Extremely light-weight gripper

Brainstorming

ornstein.mike@gmail.com — 2010-06-03T23:02:07-04:00

Today I met with Sid to discuss the ideas that we’ve been coming up with to achieve the goal of creating a compliant, soft gripper.
The main contending ideas are:

1) A pneumatically driven, jointed hand.
This hand would contain a ‘skeleton’ surrounded by a padded layer to reduce contact forces. Each joint would be driven by a pneumatic bellow with a spring return.

2) A tendon driven, inflatable/soft jointed hand.
Similar to the hand described above except driven by a series of tendons. This is essentially the Shadow Hand but made soft.

3) A tendon driven continuum hand.
Each finger of this hand would be a typical three or four tendon continuum ‘arm’ capable of grasping. The remote power source could potentially be pneumatic in the form of pneumatic muscles or typical pistons, or using electric actuation as in the two current arms.

4) A pneumatic chamber driven hand.
The fingers in this configuration each contain three air chambers running the length of the finger, arranged symmetrically about the center of the finger. The finger is actuated using a variable air/gas pressure in each chamber. The dynamics of this type of system are similar to a tendon based continuum arm.
The palm could be made of silicone as well and would ideally integrate all grasping fingers as a single cast silicone part. Eventually the air chamber approach could be implemented in the palm to increase the capabilities of the grasper.

Here is a snapshot of my whiteboard with the possible design paths set out.

At this point, the vote is for option 4. I am investigating pneumatic control systems and silicone injection molding manufacture techniques. The challenges ahead include fine tuning the pneumatic functionality, ensuring appropriate build quality and eventually dealing with tactile sensing. Integrating the finger into a grasper will be done with a central silicone palm at some point, but initially work will be done to test the feasibility of individual fingers.