Machine Learning Prosthetic Arm Concept

Рет қаралды 208,204

Күн бұрын

I've been wanting to try this project for some time. I'm attempting to use machine learning to control a prosthetic arm from my other body motions by training it through various postures while wearing a motion capture suit.
With thanks to:
Ogma Corp: ogma.ai/
Cool Components: coolcomponents.co.uk/
ROBOTIS: bit.ly/3Az11Wr
DYNAMIXEL XM540-W270-T: bit.ly/2TF97Mj
DYNAMIXEL Shield for Arduino MKR series: bit.ly/3yp0tQW
Filament from: www.3dfuel.com/
You can support me on Patreon or buy my Merchandise:
***************************
Patreon: / xrobots
Merchandise: teespring.com/stores/james-br...
***************************
Affiliate links - I will get some money of you use them to sign up or buy something:
***************************
Matterhackers 3D printing supplies: www.matterhackers.com?aff=7500
Music for your KZfaq videos: share.epidemicsound.com/xrobots
***************************
Other socials:
***************************
Instagram: / xrobotsuk
Facebook: / xrobotsuk
Twitter: / xrobotsuk
***************************
CAD and Code for my projects: github.com/XRobots
Huge thanks to my Patrons, without whom my standard of living would drastically decline. Like, inside out-Farm Foods bag decline. Plus a very special shoutout to Lulzbot, Inc who keep me in LulzBot 3D printers and support me via Patreon.
HARDWARE/SOFTWARE
Below you can also find a lot of the typical tools, equipment and supplies used in my projects:
Filament from: www.3dfuel.com/
Lulzbot 3D Printers: bit.ly/2Sj6nil
Lincoln Electric Welder: bit.ly/2Rqhqos
CNC Router: bit.ly/2QdsNjt
Ryobi Tools: bit.ly/2RhArcD
Axminster Micro Lathe: bit.ly/2Sj6eeN
3D Printer Filament: bit.ly/2PdcdUu
Soldering Iron: bit.ly/2DrNWDR
Vectric CNC Software: bit.ly/2zxpZqv
Why not join my community, who are mostly made up of actual geniuses. There’s a Facebook group and everything: / 287089964833488
XROBOTS
Former toy designer, current KZfaq maker and general robotics, electrical and mechanical engineer, I’m a fan of doing it yourself and innovation by trial and error. My channel is where I share some of my useful and not-so-useful inventions, designs and maker advice. Iron Man is my go-to cosplay, and 3D printing can solve most issues - broken bolts, missing parts, world hunger, you name it.
XRobots is the community around my content where you can get in touch, share tips and advice, and more build FAQs, schematics and designs are also available.

Пікірлер: 443

@RonaldFinger 2 жыл бұрын

Honestly this would be great for a Grievous or Goro cosplay!

@RobReynolds 2 жыл бұрын

Or a transformer based on a Fiero.......

@Matty.Hill_87 2 жыл бұрын

If someone managed to use this successfully in a goro cos play that would be insane

@Avetho 2 жыл бұрын

_"A fine addition to my collection!"_

@fryreartechnology7611 2 жыл бұрын

It has to have the spinning hands to make it just right lol.

@aleisterlavey9716 2 жыл бұрын

I prefer something more Doctor Octopus style XD! But are you gonna stuff that arm with sensors, tools and gadgets? Please stuff it with Gadgets....Laserpointer, mini vacuum cleaner, Walkman, nerf gun, abacus, ...

@jdmeesey 2 жыл бұрын

This is exactly the kind of thing I’ve wanted to do since I mistakenly choose to focus on computer engineering in Jr. High. 🙏 Thank you so much for letting me live the science vicariously through you 😂

@experimentalcyborg 2 жыл бұрын

Changing careers is hard but not impossible. Make sure you do hobby projects in your desired field and document them properly, it'll help you get a job later!

@ashwin372 2 жыл бұрын

@@experimentalcyborg can you please explain more about the documentation part?

@experimentalcyborg 2 жыл бұрын

@@ashwin372 If you document your private projects with the same quality as you are expected to document a professional project, you can use them to impress employers. "Holy shit if this is how he documents is hobby code, his work must be amazing"

@kayakMike1000 Жыл бұрын

Junior high has computer engineering focus? What?

@techmouse. 2 жыл бұрын

**thump** **thump** "No, bend more in this direction." **whirring noises** **thump** **thump** "Stop jerking so much." **thump** **thump** **whirring noises** Imagine what his neighbors think when they hear that coming from his attic.

@AshesWorkshop 2 жыл бұрын

We’re one step closer to getting a real Battle Droid. Roger roger.

@Zeithri 2 жыл бұрын

Roger Roger

@demonite2473 2 жыл бұрын

Roger Roger

@blizzard4025 2 жыл бұрын

Roger roger

@fryreartechnology7611 2 жыл бұрын

Roger Roger

@irnoble6832 2 жыл бұрын

Roger Roger

@Scott_C 2 жыл бұрын

Very cool! Using EEG (brain electric signals) combined with an AI is more or less what is being done by Neuralink. EMG (muscle electric signals) is also being used for moving prosthetics. Seeing these topics explored here is exciting! I can't wait for what's next.

@jamesbruton 2 жыл бұрын

Still planning my next move. I want to see what other suggestions there are on the video and have some time to think about what's practical.

@CLIFTON-qf3ng 2 жыл бұрын

@@jamesbruton Can this project be made into a much bigger project,(like the hacksmith industries' alien loader) with higher budget, crew and longer time. I would love to see something like this as opensource, cause we as humans are not that far.

@CLIFTON-qf3ng 2 жыл бұрын

This wouldn't just benefit people with disability, but also people working with heavy machinery. As they could use a much stronger and buffed version of this arm for lifting heavy stuff or working in un-survivable environments(such as deep sea and space).

@Taygetea 2 жыл бұрын

I get what you're saying, but I want to nitpick: generally EEG as a term is restricted to noninvasive, through-the-skull electrical signal detection, not the more direct route Neuralink is taking.

@Scott_C 2 жыл бұрын

@@Taygetea yeah that's why I tempered my statement with. "more or less" I just wanted brevity as this is just a KZfaq comment.

@theexo-engineer9603 2 жыл бұрын

You’ve done it mate 1Mil Awesome job. Iv been watching you for years and you are continuously improving your content and builds. Keep it up. Love your work

@tboicelli 2 жыл бұрын

Using Quaternions and spatial rotation will address your gimbal lock (where the axis rotations get confused when aligned with another axis when moved)

@paul_vantieghem 2 жыл бұрын

Rotation matrices can also avoid the numerical singularities which we experience as gimbal lock, and they are much easier to understand so maybe he could start with that. The only downside is that rotation matrices carry 9 numbers instead of 4 for quaternions.

@daveabittner 2 жыл бұрын

The problem is, his arm is quite literally a physical gimbal. It's a limitation of the anatomy to an extent, and as a result (if you want something that is attached to your arms) you will have to duplicate sensors I believe in different orientations.

@robertmorby3694 2 жыл бұрын

kzfaq.info/get/bejne/ns50l8mat6m1g2g.html you might like this

@robertmorby3694 2 жыл бұрын

@@paul_vantieghem kzfaq.info/get/bejne/ns50l8mat6m1g2g.html

@JinKee 2 жыл бұрын

Geometric Algebra has a concept of bivector rotations which is also useful

@deadlokIV 2 жыл бұрын

Thank you _so much_ for this - been working on a conceptually identical project for a few years now (but on the tensorflow/cuda level and non-public) and this honestly was a huge asset to those works. Really appreciate the upload - you may have just supercharged a bunch of efforts to make the concept closer to a deployable reality thanks to a lot of the ideas you have introduced here. All the best!

@richardson100489 2 жыл бұрын

I really hope you continue this project! It could end up helping a lot of people.

@mvwouden 2 жыл бұрын

You could consider Deep Reinforcement Learning with added imitation learning in the reward function. PPO and DDPG are some solid options if you are considering. For my thesis I'm working on the control of an active transfemoral prosthesis during normal walking using DRL.

@ElevenElevenAus 2 жыл бұрын

Great video again, please keep up the work as it's so interesting to watch the progression while you piece all the components together. Thank you.

@iAmTheSquidThing 2 жыл бұрын

This is pretty impressive for a few minutes' training on a microcontroller with limited sensors. I wonder what it could do with something like an Xsens mocap suit worn for a few days and trained on a powerful GPU.

@H34... 2 жыл бұрын

Awesome! loving this project already. Also congrats on 1 Mil

@magnusjagerstrom4388 10 ай бұрын

This is really cool, can’t wait to watch what you are doing next.

@lnz5660 2 жыл бұрын

Dam he works quick I've been working on this exact same project for 2 years. Definitely gonna take some stuff from this video.

@mikegruber7131 2 жыл бұрын

WOW. simply amazing. Keep up the great work

@Nobe_Oddy 2 жыл бұрын

HAPPY 1 MILLION SUBS JAMES!!!! AND WHAT A GREAT PROJECT TO START HEADED FOR 5 MILLION!!!!

@robotcantina8957 2 жыл бұрын

Another cool project, you have me hooked. Congratulations on the 1M !!!!!

@czateu1391 2 жыл бұрын

Just noticed you are already 1million subscribers, congratulations James!!!!

@inventorbrothers7053 2 жыл бұрын

Very cool. I'm really impressed with the results you got! You are a super skilled individual.

@rohithm6280 2 жыл бұрын

Congrats on 1 million James, plz make video on your 3d modeling

@CletoFrost 2 жыл бұрын

phenomenal, hoping to see more!

@4louisMC 2 жыл бұрын

Congrats on one mil!

@lindonwatson5402 Жыл бұрын

much excitement for this series

@ClickItYT 2 жыл бұрын

It's amazing to see that 3D printing protheses at home becomes consistently easier.

@amirhm64 2 жыл бұрын

awesome stuff as always James

@Will_Huff 2 жыл бұрын

Excellent demonstration with your prototype!

@Waitwhat469 2 жыл бұрын

Stuff like this is why subscribing and being a Patreon of you is such an easy choice.

@hesalovelybuoy5944 2 жыл бұрын

Cool! Carry on with the useful robot project too

@Mechabjornis 2 жыл бұрын

Functionality aside, this is the best looking stuff you have designed, imo. Sleeker and seems even more refined. Great work and presentation, as allways !

@Isambardify 2 жыл бұрын

IMUs are so fiddly to get working if you want anything more complicated than "did the sensor tip." Have you looked at the 9 axis versions with magnetometers? You might be able to filter out the roll/pitch confusion. I'd love to see a whole series making a full IMU digital motion capture suit if you wanted to make one.

@alvaro4186 2 жыл бұрын

1 Million subs! I've been watching you since your The RPF days. Congratulations!

@muneeb-khan 2 жыл бұрын

What a wild project. This is so cool!

@maywah1276 11 ай бұрын

Thank you so much James, may god bless you and your family, and youtube.

@shpluk 2 жыл бұрын

as always amazing!

@eunoia921 2 жыл бұрын

Congratulations for 1 million subscribers. You DESERVE more. 👏👏

@LLJW4 2 жыл бұрын

This is awesome! Great start!

@thescienceysworkshop5368 2 жыл бұрын

congratulations for 1 million subs , mr james! i hope your channel grows even more

@DavidSpencer22 2 жыл бұрын

Would love to see an open source build for an assisted walking device, exoskeleton. Possibly for older people with walking troubles. Thanks for your great content!

@Skipperrrrrrrrrrrrrrrr 2 жыл бұрын

CONGRATS ON 1 MILLION JAMES 🎉🎉🎉

@Saxandviolins 2 жыл бұрын

So amazing again, keep up the good work!

@fireinnature9445 2 жыл бұрын

Wow Just Amazing!

@CreepyChappy 2 жыл бұрын

You’re my favorite kind of nerd !!!!!!! Love your stuff bubba

@isthissiddh 2 жыл бұрын

CONGRATS ON THE 1M!!!

@OsoPolarClone 2 жыл бұрын

Awesome project!!!!

@masterpig5s 2 жыл бұрын

Very cool to see it gone into

@jamesbruton 2 жыл бұрын

Thanks!

@sanoopsiby8296 2 жыл бұрын

You never sieze to amaze me

@fightme5543 2 жыл бұрын

This plus the muscle monitors could be amazingly unique. Not only would you have accurate motor controls, but unique situation awareness relating strength of signal and force feedback. This would allow someone the ability to differentiate between crushing a brick and shaking a hand whereas currently gripping is much more of an on off signal.

@reececrook7021 2 жыл бұрын

james i love you content, i'm 14 and i love build things rn i'm trying to learn how to code for aurdino and i hope that i can be as good as you one day

@Matty.Hill_87 2 жыл бұрын

Stick with it and you'll get there dude

@aykutcoskun6092 2 жыл бұрын

Go on champ! You will be even better than him one day

@endthefighting 2 жыл бұрын

You can, and you will, mate :)

@linconwalker8329 2 жыл бұрын

Thank you so very much for sharing your process and results. I have been thinking of working on a similar concept, but utilising fpv quad flight controller boards, as they work on all axis' in a very small form factor. They could be used to record the data of one limb, and also to output the data to manually control a prosthetic. The flight controllers are small enough to attach just below each point of articulation on the "learning" limb. Combined with your outlined machine learning, they should then be able to output motion commands to result in a more natural movement style.

@iiinsaiii 2 жыл бұрын

Congratulations on the 1 million subscribers 👏👏👏

@EdgeImpulse 2 жыл бұрын

Awesome work, James!

@rlizardo57 2 жыл бұрын

Great experiment.

@JarydGiesen 2 жыл бұрын

This is just damn cool

@StevenIngram 2 жыл бұрын

After thinking about this for a while, I realized that you had already designed a pretty good arm concept for what you're trying to do. Remember the Performance Robots? They had shoulders with a good range of motion, elbows, forearms that rotate and grippers. Anyway, it just seems like those arms would have the range of motion to adhere more consistently to the trained behavior.

@balkiprasanna1984 2 жыл бұрын

You're just amazing. How come this video didn't get a million views is just beyond my comprehension?

@stefanguiton 2 жыл бұрын

Great stuff!

@easyBob100 2 жыл бұрын

This could really update your 2 leg walking robots. Train the AI on walking/standing still motion + remote control input. Cool stuff man!

@gusobrien8491 Жыл бұрын

I have no idea how this man has this many ideas and consistently puts out these insane videos

@micah4932 2 жыл бұрын

This was an insane project and it would be amazing to see this used for actual disabled people in the future

@NicksStuff 2 жыл бұрын

Complex yet simple

@stevesloan6775 2 жыл бұрын

Awesome upload

@lsav1085 2 жыл бұрын

Cool idea

@bonzogamer6966 2 жыл бұрын

Good start.

@imxheartz7846 2 жыл бұрын

this is really cool

@thewatcherofpaintdry9288 2 жыл бұрын

Keep up the good work

@codingvio7383 2 жыл бұрын

James: *making an exoskeleton* CIA: "Interesting"

@TheMagneticDude 2 жыл бұрын

Congrats on 1mil!

@MIGuy 2 жыл бұрын

Excellent!

@r.iyushofficial5318 2 жыл бұрын

Awesome 👌

@dumbtex6107 2 жыл бұрын

you deserve so much more

@davidlong2691 2 жыл бұрын

Always wanted some Doc Ock arms; this is the way to do it :) Nice work!

@SonicUnderground316 2 жыл бұрын

This is fascinating! I'm sure Ogma can open a lot of possibilities for future projects but I wonder if you could use it to improve the efficiency of past projects like OpenDog and/or one of your Walking Robots (which I always like to see)!

@chinmayjategaonkar5806 2 жыл бұрын

James, you could give the IMU BNO080 a try. It has internal processing that gives quaternions which are highly useful to find global orientation. That will reduce the processing cost and allow you to sample data at a higher rate, also solving your vertical orientation problem. Amazing video!

@ironman12328 2 жыл бұрын

I see absolute no one so far, sooo: Congratulations for reaching finally the 1 Million Mark you deserve it sooo much and I hope the Next Years are going to be Interesting AF and also Awesomer, Better and much more Complicated haha again Congratulation for the 1 Million

@jamesbruton 2 жыл бұрын

Thank you so much 😀

@Abishek_Muthian 2 жыл бұрын

Great project as always! I was wonder if pose detection would have been easier than manually encoding the position of the arms, legs for training and inference.

@Kevin-ls9zs 2 жыл бұрын

You make me wonder why I’m even in college for mechanical engineering when I could have just done this type of stuff for a living. You have the dream job I wish I could do stuff like this everyday instead I’m doing math and memorizing formulas and stuff

@SentinalSlice 2 жыл бұрын

You should definitely make a choreographed slow motion fight. Have sensors on one arm and set the robot arm to react to those sensors. Then make the outputs be motions that would block slow punches and attacks. That would look really cool.

@rebeccac4573 2 жыл бұрын

James hit 1 Million Subscribers great job. 😊

@netgeek3010 2 жыл бұрын

Longtime member congratulations on hitting a million subs.

@dynorat12 2 жыл бұрын

looks good

@christopherashford822 2 жыл бұрын

Seems if you walk around for a day or more normally training the algorithm and switch it to playback, it would probably surprise you how good it is.

@kavindulakshitha9732 2 жыл бұрын

Hey you have 1M subscribers. Congratulations 👍👍👍👍👍

@Tsukay. 2 жыл бұрын

Very interesting!

@d-channel5841 2 жыл бұрын

Very cool

@anthonyvillanueva4500 2 жыл бұрын

this one i love very cool

@swannschilling474 2 жыл бұрын

OMG this is sooooo cool!!!! 😀

@jaysonbunnell8097 2 жыл бұрын

This is so cool! can't wait to see what you do with brain signals or something.

@VideoStefan17 2 жыл бұрын

awesome!!

@panakon366 2 жыл бұрын

You should wear a camera on the head and gloves with markes to train a hand that picks up various shaped objects. You could add the same markes on the robotic arm and having it mimicing your gestures based on the object you want to grab.

@_Piers_ 2 жыл бұрын

Happy Gold Play Button James :)

@hardiknayyar7057 2 жыл бұрын

Hey, I like the videos were you give conceptual knowledge in deep it helps me a lot please do these in depth video explaining telling about each and every step in detail no matter how long is the video I like watching long videos

@vatsal1231 2 жыл бұрын

Congratulations for 1 million subscribers 😎

@kingmasterlord 2 жыл бұрын

yes james, this is what I've been talking about! you can 3D print a Dexter's Laboratory exosuit if you throw in some aluminum extrusion. just make opendog longer and you're done!

@zeekjones1 2 жыл бұрын

Now to scale it up, add the rest of the limbs, an omni treadmill, and make a mech. Can use the learning as a feedback for a 'fly by wire' controller, even if you have all your limbs. Theoretically if it can predict your movements, it would react quicker than a direct 1:1 controller, with enough training.

@CommonApathy 2 жыл бұрын

Would probably more like, assume I'm still doing what I was doing and continue it until you get a different input, rather than predict the future.

@laureven 2 жыл бұрын

Big things have small beginnings :) ....Awesome Video

@donaldburkhard7932 2 жыл бұрын

Divide up arm to forearm, elbow, upper arm, shoulder. They can then be combined for the whole movement. Same done for legs and other body parts.

@TheGreatCthvlhv 2 жыл бұрын

This tech could be used for something like a mech suit (cosplay or otherwise). Of course a direct read from the arm sticker things would probably be the better option in that particular scenario.

@charliesanders6181 2 жыл бұрын

Yeah I was thinking with a bigger motor you could strap steel to it and it wouldn’t be exhausting to wear as a suit of armor.

@joeschmo7002 2 жыл бұрын

This is an amazing video, Jame really is a genius. I dabble in electronics & robotics, so I know how difficult it is to learn and build stuff like this, especially stuff this complicated and at the rate that he releases videos. The only critique/question that I have about this one is why didn't he just use potentiometers and MPUs/gyroscopic mechanisms in the motion suit, with a machine learning algorithm receiving data from those sensors. It would also simultaneously be fed data from another pre-trained AI algorithm to recognize his arms. He could put long strips of reflective tape down his arms, as well as adding strips around their circumference, would make the arm-recognition algorithm easier to train and more reliable. Once he's trained the model for long enough, and is wearing the arm, he could add buttons to tell the algorithm whether that was a good or bad motion, using it to progressively tune the algorithm to exactly how you like it. Also, I'm not criticizing in any way, his method got the job done, I'm just probing a little to see if I might have any good ideas on improvements.