Soft robotic spine

Maecer · Post by **Maecer** » Tue Sep 04, 2018 8:58 am

There's been discussion on the forum about soft robotics, movement and programming. While looking through Soft Robotics Toolkit, I came across this:

https://softroboticstoolkit.com/fetch/t ... r-platform

I find this to be very promising for possibly constructing a flexible spine for future dolls. They show that multiple units can be linked together, and I think replacing the fishing line servos with small hobby pneumatic shocks would provide good feedback and dampening. While the actuation video is rather stiff, it is possible to program for smooth motion, and using these actuators would provide a more natural motion; whether it be leaning the doll for standing and walking, or achieving a humping motion for sex. Even whilst the doll is laying down, a flexible actuated spine could allow for more natural positioning during any position for sex. I would move from the 3 actuators to 4 for more stable positioning on a spine. There is software given on the website for modeling soft actuators.

I'm really excited about this because it a) removes the need for complex mechanisms to achieve doll motions, b) air can be replaced with a fluid to make it self contained and c) the equations used in the middling can be programmed when actuation and motion are needed for the doll.

Thoughts, comments, feelings????

LDF · Post by **LDF** » Tue Sep 04, 2018 9:36 am

I'm a proponent of soft robotics for love dolls/androids.

I have been exploring, studying, and tooling up for my own pneumatic, 3D printed, 2 axis, soft robotic joint.

IMHO:

1) Keep it pneumatic for safety reasons. Air is safe while hydraulic fluids can cause many issues.

2) .3 bars is a good low power setting for starters. That's less than 5 psi.

3) Control and power are easier to achieve in a pneumatic, soft robotic platform. I've had to begin designing my own pneumatic reed valves, fluidic logic controller, etc, but it's do-able . A carbon fiber 4500 PSI paintball tank can hold enough air to do a useful amount of work. Weight is paramount for safety.

4) Once a pneumatic robot is developed, love dolls/love androids will be close behind.

However, I'm still at the point where I'm studying Python, taking an adult ed. course in CAD starting next Friday, trying to rearrange some space to set up for 3D printing w/ a work bench, and gathering materials while solving problems, etc. .

The design for the 2 axis pneumatic joint is evolving to something that's simpler and more modular. I'm also designing it so that it's easier to use for injection molding (should I succeed).

I'm also basing my 'skeleton' on existing sex doll skeletons to make things easier.

Those are my thoughts and comments for now.

Maecer · Post by **Maecer** » Tue Sep 04, 2018 10:09 am

LDF wrote:I'm a proponent of soft robotics for love dolls/androids.

I have been exploring, studying, and tooling up for my own pneumatic, 3D printed, 2 axis, soft robotic joint.

IMHO:

1) Keep it pneumatic for safety reasons. Air is safe while hydraulic fluids can cause many issues.

2) .3 bars is a good low power setting for starters. That's less than 5 psi.

3) Control and power are easier to achieve in a pneumatic, soft robotic platform. I've had to begin designing my own pneumatic reed valves, fluidic logic controller, etc, but it's do-able . A carbon fiber 4500 PSI paintball tank can hold enough air to do a useful amount of work. Weight is paramount for safety.

4) Once a pneumatic robot is developed, love dolls/love androids will be close behind.

However, I'm still at the point where I'm studying Python, taking an adult ed. course in CAD starting next Friday, trying to rearrange some space to set up for 3D printing w/ a work bench, and gathering materials while solving problems, etc. .

The design for the 2 axis pneumatic joint is evolving to something that's simpler and more modular. I'm also designing it so that it's easier to use for injection molding (should I succeed).

I'm also basing my 'skeleton' on existing sex doll skeletons to make things easier.

Those are my thoughts and comments for now.

I've got a programming background and extensive AutoCAD skills. Finite element analysis is very useful for both, as the results will be used for programming functions and limits, and modeling in 3D will give projected movements and forces.

I'm a strong proponent for Arduino and RaspberryPI based projects. The PI supports numerous coding platforms, and has the backbone needed for soft robotic calculations. Arduinos would execute based on what the RaspberryPI calculates for movements.

Unfortunately for me, I have what I call 'single, new dad money ' at this point and cannot facilitate the prototyping I would like to pursue. I'm doing all the research I can on how I would integrate all system components, sensors and actuators though.

For your muscles, check this out:

https://softroboticstoolkit.com/book/pn ... al-muscles

The documentation shows relations to amount of force exerted, time required for full force and percent contraction based on silicone elasticity. They also show how the supporting lattice structure from the mesh corresponds to force and contraction as well.

For hands, I'm planning on using every resource on that site to develop a fully segmented hand. There's articles under documents that show their processes. Found here:

https://softroboticstoolkit.com/documents

All of the examples and documents I've seen so far have the 3d printer files included. There's examples of printing circuits and sensors into soft hands/components.

LDF · Post by **LDF** » Tue Sep 04, 2018 11:21 am

I am familiar with the soft robotics toolkit.

I'm using an actuator that's more like the ones in that 'spine' YouTube video.

Except, I've designed my own origami pneumatic bellows/piston.

Currently, my piston is a soft robotic TPU pneumatic actuator, but the joints will be PLA or nylon. The rest of the robot skeleton will include PVC pipes to keep costs down.

I would love to be able to fabricate a completely soft robotics 3D printed robot that both keeps its shape and moves via pneumatic pressure.

3D printers are still too 'primitive' and their low speed limit them to printing 'prototypes'.
So, I'm constrained to start w/ 3D printed TPU actuators in PLA/nylon joints.

I've worked w/ micocontrollers in the past, and I do have an uno, and a bunch of other controllers around.

I plan on using adafruit controllers, etc., since they're (adafruit) local, and they have the kinds of controllers I like.
Besides, I can always just hop on the subway and visit them if need be. Power consumption is an issue, and I'd prefer to hand off more demanding CPU/GPU/TPU operations wirelessly.

I'm learning python because of the AI required to make this thing work properly.

I'm not going to 'program' the resulting 3D printed, 2 axis, robotic joints.

I'm going to help the 2 axis joints to learn to function properly.

So, while I'm updating my own AutoCAD skills, and setting up shop, things are beginning to go from the exploratory and design stage, into the prototype stage.

: WFX_6532.jpg (23.99 KiB) Viewed 3266 times

I like the above type of skeleton because with the type of 2 axis joints I've designed, I can get a good range of movement. (I can do without shrugging shoulders though)

Maecer · Post by **Maecer** » Tue Sep 04, 2018 11:37 am

Best of luck with learning to use neutral networks. I have been looking into those as well, and have found there are several Arduino projects out on the web that give good tips and advice for constructing those networks.

As far as I'm aware, you have multiple end states that you would have the robot learn to reach; one could be like a state called 'straight spine.' I would have an IMU for each soft joint to help each your alignment; the end states could be reach based on an angle of N-degree of difference from top to bottom. 'Straight spine' would then feature +/-2degrees of difference from the top IMU to the bottom.

Would love it if we could collaborate some.

Maecer · Post by **Maecer** » Tue Sep 04, 2018 12:15 pm

See my sketch

: Sketch of hypothetical soft robotic spine with attached 9- D.o.F. I.M.U.s; 20180904_120753~2.jpg (1.13 MiB) Viewed 3259 times

IMUs would provide sensor feedback to calculated positions and angles based on expansion from soft actuators. These values would be compared, and the AI would adjust the soft actuators until the IMUs' position is true to the desired end state.

LDF · Post by **LDF** » Tue Sep 04, 2018 12:29 pm

As for positioning, I'm examining pose estimation modelling using a video camera.

The idea being w/ a neural network, video a person moving w/ a pose estimation app, then get the robot to learn to imitate it w/ a video camera and the same pose estimation app -training it via the neural network.

I would rather train the robot in a 3D unity setting using a method like the one above. But, that's a bridge too far for me.

I've also got other sensors that can be trained along with the pose estimation app data, but that's down the road.

First things first. I need to finalize the design on AutoCAD, 3D print a 2 axis joint, then get it to work properly.

Maecer · Post by **Maecer** » Tue Sep 04, 2018 12:36 pm

PixyCam is a user friendly camera system that can be trained to detect objects and their orientation. Two can be mounted together for binocular vision. OpenCV is the programming language for video processing using your own camera system.

Ive seen sensors and programming out there that are out there for training your robot/ hand to act like you do. Several use the Xbox Kinect. All fall under either voodoo processing or sensors on a suit/ glove/ exoskeleton.

Maecer · Post by **Maecer** » Tue Sep 04, 2018 12:46 pm

I knew there was something out there. Here's an example of what I found

http://metamotion.com/gypsy/gypsy-motio ... system.htm

Maecer · Post by **Maecer** » Tue Sep 04, 2018 12:50 pm

Aaannnddd this that I found for those who don't want to spend lord of money

https://hackaday.io/project/9266-motiosuit

Searching for motion capture Arduino is how I found it

LDF · Post by **LDF** » Tue Sep 04, 2018 1:01 pm

Maecer wrote:See my sketch

20180904_120753~2.jpg
IMUs would provide sensor feedback to calculated positions and angles based on expansion from soft actuators. These values would be compared, and the AI would adjust the soft actuators until the IMUs' position is true to the desired end state.

Well, you can see from the DS doll skeleton that it's 'spine' is basically just a 2 axis joint in the upper abdomen.

Remember that you need to wire and power each of those sensors. While I will be using a something like the following to deal with general positioning/orientation...

https://www.adafruit.com/product/3387

I'm also taking a page from AI speech recognition and embedding 3D printed, ultrasonic micro whistles in to the joints, valves, etc., to see if it's possible to determine joint position w/o adding extra sensors/wires. I should take one or two well positioned ultrasonic pickups to read all the joint ultrasonic wave forms for AI processing and comparison with the other data streams ( pose estimation, 9DOF gyro/mag/accel). Different micro whistles will have their own center frequency and even envelopes. I just need to be sure the micro whistles are somewhere between 2 and 4 mm deep (they're just holes in the sidewall of an air channel) to stay between 20 to 60 kHz.

So, I would avoid costly over-designed elements. A spine, a bunch of wired sensors w/ a Raspberry pi on top of that is more than is needed to start IMHO.

As far as sensors go, I would love to see far infrared smart dust that could form a sensor net, etc..
But, it hasn't happened yet.

mandrewanley · Post by **mandrewanley** » Fri Sep 07, 2018 7:39 pm

I wonder how heavy the robot could potentially be if you used an anatomically correct skeleton then added pneumatic actuators

Maecer · Post by **Maecer** » Fri Sep 07, 2018 9:47 pm

I think it would depend on the materials

LDF · Post by **LDF** » Fri Sep 07, 2018 10:59 pm

mandrewanley wrote:I wonder how heavy the robot could potentially be if you used an anatomically correct skeleton then added pneumatic actuators

A better question to ask is: "How much would a doll built with pneumatic structural elements (inflated 'bones') and pneumatic actuators weigh?

When I say inflated bones, I mean elements made rigid via air pressure.

It's one of the reason why I'm currently studying CAD and bought a 3D printer.

In a way, it's a throw back to the inflatable love doll.

LDF · Post by **LDF** » Sat Sep 08, 2018 8:13 am

To answer the question:

Weight of the :

(9+) 2 axis joints, +

pneumatic tubing (there's a lot of it) +

PVC limbs (8), spine (2) shoulders/neck (1), +

feet and hands (2 each)

Pneumatic control manifold, +

Air tank (paint ball) and regulators, +

Micro-controller boards w/ batteries, +

smart phone/tablet (its head/face!), +

misc. connectors, fasteners.

I'm hopeful that the assembled robot 'skeleton' comes in at under 25 lbs.

The main idea is to keep it light, low power, and safe.

Also, it should be something that once proven to be safe and has enough trained models insatalled (thus the smart phone/tablet), it can be mobile (walk), communicate (talk), and be available for the market.

IMHO:

The animatronic dolls that are being developed are a start, but I feel that since they use servos, etc., they will be too heavy and unreliable.

Soft robotics is the future, and I'm hoping my pneumatic, 2 axis robotic joint will pave the way for the newer soft robots that will follow.