I would personally recommend the tubing used on bicycles for their gearing and brake cables. already designed to do exactly what you need, and a good way to solve cable management if you have simply clip into the carpal and metacarpal regions. they come in spools you can cut to length and little metal caps so you just need to slot them in place endwise for them to maintain tension exactly where you need it. glad to see you back on this, looking forward to more RnD
quite bulky though, wouldnt it be? im also not sure bikes use bowden tubes with the same inner diameter as his cables, might be difficult to source, no?
Ehh...They might be a 'bit' bulky. (3/16 in. OD ) But I can confidently say they would hold those cables without issue. Brake cables on a bike are at least twice that thick. (1/16 in.)
Unfortunately, by the time you stack 5 or 10 (or more) of these together for all the required tendons, it becomes impossible to bend. There may be some specialty mini cables used in model aircraft that are similar but they're difficult to source. I would love a good solution to this problem.
You can use something a little less slippery than PTFE but stronger. Many years ago I remember playing around with acetal and uhmw, but springs with inner ptfe tubes worked the best. The frictional force is modeled by the capstan equation.
skyentific tested that on his channel, and it turned out to be impractical as others here point out. He ended up using a system of pulleys and coils for a finite but sufficient range of travel in both directions. Backdriveable, zero backlash, but tricky to get right it seemed.
I remember watching one of your videos when i started studying Biomedical Engineering, now i've graduated and still wish to make a proyect like yours Glad to see you working on this :)
I think it is a wonderful idea. These arms are useful in dealing with dangerous materials inside a closed room and controlled remotely, such as explosive materials or toxic and acidic materials.
Great job, nice to see your progress! You could maybe use tubes made to guide brake cables on bicycles, they are made to work in compression. They might be slightly larger than your PTFE tubes however.
you could use small diameter ptfe tubes, ptfe even has self-lubrication, which is perfect. you just have to make ports on the ends, that you caN mount.
i did the hand connection to an xt ibm back 90's ! used window plastic and stainless 1/8 " wire for pins! your fingers are exact as i did and i put a plastic plate on the top of the knuckl to prevent the fingers from going backwards! the hand griped a small 380 9mm dual ammo baritz and pulled the trigger without any problems !
I still can't believe that massive government funded projects on bionics that have been around since the late 90s to early 2000s still haven't managed nearly as much accurate range of motion as this project made by a student. Looking forward to seeing this develop further
Was watching you in 2018, glad to see you have returned. The project is really great, especially in terms of freedom of motion. By the way, have you read the research of using sewing thread and make it deform under some heat application, the thread should be spinned multiple hundreds times (depends on length) and if you will apply cooper wire for heating, maybe you could use them instead of servos. Thank you!
Great work! I am doing this with PAM/ mckibben muscles. Compliant and more robust than servos. Currently i am musing on how to make control valves that regulate pressure under servo control.
I support putting the servo drivers on the main PCB, partly because it's cleaner and partly because I like watching the servo's components get increasingly distributed through the hand.
I'm so happy to see more! I started building the previous design, am struggling to get the coronas into the joints. I want to use this hand with my Inmoov android arms. Been having a good bit of fun with it
I have seen many people using other kinds of motion capture technology, I think Manus gloves right now are the most used in the robotics field. i believe even Tesla is using these gloves for Optimus. I have tested them during an event, so far they are the most accurate motion capture gloves i found. even tho theyre a bit pricy, but i didnt find a better solution. Thank you for your videos, very interesting and informative!
This is great work. Springs with a smaller ptfe tube work pretty well! You can also route the cables primarily in straight lines on pulleys where you mantain constant loop length easily and then switch over to bowden tubes just where routing gets difficult. This can help increase stiffness and decrease friction. Out of curiosity, what is your day job?
I love your work. I've been waiting for this for so long. I have a request if you could build a terminator version of the arm. If it isn't a trouble for you.
this is impressive, but one tip that I learnt the hard way: you only have two tendons per finger; the finger can flex so much because it's one tendon anchored at several points and it can flex independantly because tendons are malleable. you therefore only need 5 servos per finger, at least one for splay, and about two for the thumb: one for positioning along a curve, the other for flexing it like any other finger the CMC is basically an overdeveloped splay-muscle for a finger that's at 45* and in fact you can line your thumb with your finger for this exact reason. this alone means that the total servo count shouldn't exceed 8 for the hand, 2 for the wrist for something that is completely biomimetic. you can afford to simplify the main palm into just two halves, which should save on mechanical complexity, but a 3-part palm is closer to the real thing. for said tendons, bike cables would work well, and you can anchor the tube along the 1st and 2nd joints of each finger, with the cable itself being anchored on the finger. this way you have a compliant, durable and strong system. another thing to note, human palms reach farther than the knuckles due to them being flesh, which from the underside gives the illusion of having an extra set of joints in the fingers, also also, the shape of the palm is largely immutable save for the outer metacarpals of the ring and little finger which ever so slightly curve, though that is mostly caused by the compliant nature of having a frankly unnecessary amount of bones in the palm actually working on something similar myself, albeit drastically simplified and less "humanised" (it's just angular because I don't have fusion360 :p) either way uh... hope that I can help simplify things and my apologies if my knowledge of the series has gone a tad rusty, it has been a few years... EDIT: to clarify for understanding; by anchor, I mean like a thread passing through the eye of the needle: as the tendon is pulled, its shortening forces it against the structure, dragging the rest of the structure with it.
Excellent progress. I was wondering if you could route the tendon cables through parts of the hand structure? This would shorten the tubes and reduce the kinking. Or where several cables run parallel, 3D print a structure to encase the cables, reducing the length of tubes needed as well as keeping the cables tidy. Keep up the amazing work.
Use flexor and extensor electrostatic motors to get rid of the tubes. And just use guide loops to keep things tidy. (forearm anatomy but in tech form) Think of a z-folded capacitor where the dielectric is rubber. Apply voltage and it contracts. No volts and it relaxes. Use one for the flexor another for the extensor.
Need stiffer guide tubes for the cable tendons! Squishy tubes don't lend themselves well to transmitting much power. This is why bicycle brake cables have a metal coil that's then sealed in plastic to hold it together. Basically a long thin spring should work, you can wrap something around it or probably just find cable housing/cover designed for the application. There's all kinds!
Any kind of stiffer cable housing would be great! Bicycle cable housing is meant to be very incompressible (especially for brake cables) but they might be too thick for your design. Good to checkout though, and maybe there are some engineering bowden cables that might be the right thing! Bicycles also sometimes have springs embedded into their brake assembly so you can't 'lock up' the wheel. Might provide some of that grip strength. Love this project, keep it up!
Note that you'll have to deal with cable stretch over time. Also, if you haven't encountered this yet, if you keep all the tendons tight to the skeleton, you'll soon find that when you make it make a fist, then bend the wrist, you'll have to further adjust the tendons in the fingers to give more or less cable length as you articulate the wrist. You'll need to account for that in your cable length adjustment calculations.. unless if your just using a closed loop system to just ensure the right amount of finger articulation while moving the wrist
Nice! I'm glad to see you ditched that bulky glove! Though, I feel like you're missing an opportunity with 3D printing... You could route out internal cavities for the wire management, and even ditch the PTFE tubing by running the cables internally in their own dedicated cavities as well. I understand that the length of each section would not be constant like it would with PTFE, but it would be predictable and calculatable for any given pose, and the arduino has plenty of capacity to calculate and interpolate the non constant cable distance.
Hi Will, I have written my master thesis on robotic hands and built a prototype using continuum robotics. Have a look at the DLR Awiwi hand. It is probably the most advanced hand out there. If you need any literature, let me know! Cheers
Dremel would've made short work of cutting that tab off, using a metal cutting disk you could cut it right next to the gear itself with minor filing for cleanup after...
Obviously building robotic hands at some point you want to program them to do amazing things. But the whole robot humanoid is another awesome goal. Then there's just so much to do. Think of the huge teams and resources that are going into creating essentially a robotic humanoid.
Have you considered using electromyography sensors on the arm to control the hand? You could record the electric response from the muscles in your forearm, combined with the motion capture data to train an AI model to predict the hand position from the EMG signal alone.
Your ptfe tubing looks really soft, almost like silicone tubing. The ptfe tubing in the bowden tube on my 3d printer is must more rigid. I think once you find better ptfe tubing which is more rigid and tighter around the cable then you'll see smoother movement. I know some sections need to be flexible so you'll probably have to transition between the two tubes in some places but for the straighest run from the forearm to the wrist it can be the more rigid one.
Amazing work! You must have incredible patience to do such a project as this. Did I hear that right? You are using Arduino(s) to control this hand? If so, may I ask why you wouldn't use ESP32s for this project? Faster, Dual CPUs, more memory, built in WiFi (or peer-to-peer) communications, etc? Perhaps I didn't hear that right. Personally, I've been deving on Arduinos for years, but only a year or 2 ago I got into ESP32s and just fell in love with these. So much more robust. Anyway, lovely job there.
For the tubing, you may be able to use bike shift/break housing instead, or take a note from their design and put your existing housing thru something to give it more rigidity, like a spring with the same i.d. as the o.d. of the tubing, however, if sized wrong, the spring may also introduce increased internal friction, in which case, going with an existing product will probably be the better choice.
@@artbyrobot1 please explain you reasoning, as i know the OD of road bike shift tubing to be about 2-3mm larger than the tubes he's already using, so based upon the size of the pieces the tubing attaches to, it would make no appreciable difference to the size of the attachment points.
@@THarSul max diameter od should be 2mm tpfe with 1mm id since 130lb test woven PE fishing line fits into this and is 0.65mm OD. Anything larger than this is unnecessary weight and size and bloats everything which DESPERATELY needs minitiurization at every turn to fit the human form factor. going WAY UP in size and weight is foolish like you are saying. If you want to thicken your tpfe, you can use super glue and thin fabric to create a little cast around it as a composite structure which would prevent any warping or buckling during actuation. but this should not be needed in most cases I don't think.
@@artbyrobot1 the reason for my suggestion was that if this is a benchtop prototype, it wont be fitting into a human form factor, in which case bulk has less of an impact. With regard to using fabric and superglue to make a composite around the tubing, that would likely be too rigid to handle the forces involved, and would lead to breakage at points where flexibility is still required. A similar performance to the cable housings could be achieved with a length of tension spring with a similar ID to the tubing’s OD, which was how road bike shift cable was routed in the past, due to the low profile and lower weight than the tubing with the rubberized outer coating.
@@THarSul the problem area seems to be just in certain sections like in his case near the motor and would not be the whole length of the tubing. so the composite cast sleeve would be fine for such areas that don't need to flex anyways. very few areas need to flex. also, just because it is a prototype does not mean space isn't a huge concern: it is during prototyping you find the solutions needed for human form factor capability. so using something that will not be compatible for human form factor is bad to do. it pigeon holes you into tactics and methods that are not usable in later editions which is taking 2 steps backwards. miniaturization is EVERYTHING for this. getting it all minimum weight is also a big factor. using metal for this is adding unnecessary weight. also, thin fabric and super glue will not snap or be too rigid for most sections and you can use a glue that has a little bend to it which most do. you can use epoxy instead of super glue if you want more flex in some areas and don't mind a little more fumes and drying time.
For servo management make it a full forearm the basic hand receives nerve impulses from as far back as the elbow a way so just continue to mimic natural and give yourself more room in that model should help some. Also elastic, spring, or even a secondary servo as a return to home position on the opposite side acting as a tendon with a dampener on it can help with smoothing the movement and making it feel more real for you.
Dunno if this will be a useful suggestion, but there are some compliant finger designs on articles/research papers that might be worth a look. One example is "Mechanical Structures for Robotic Hands based on the “Compliant Mechanism” Concept" where the entire closing motion is done by a single pulling "string" attached to all three parts of the finger.
A very interesting design. Employing spring tension to the individual phalanges may resolve some of the jerkiness. But this may complicate your design, as the proper tension is a variable.
what you are doing is very cool, but it will be even cooler to attach a neural network to this hand that could read brain signals using an Electroencephalography device and convert them into signals for the movement of actuators
I wonder how well a solenoid + spring assembly inside the finger would work compared to the servos. it probably would use a lot more power, because you'd need it to always be active. another solution might be to use basically a linear stepper* (i.e. a stepper motor but instead of being in a circle, the coils are in a line), which again is complex, but might give a compact system. (and you only need like 1cm-1/2inch of travel.) * basically what you'd do is a block of magnets facing opposite sides (i.e. if you look from the top going NSNSNS, and the bottom SNSNSN) with a pair of solenoids, one on top, the other on the bottom, with teeth on each pole, spaced the width of 2 magnets, and the poles being offset by one magnet. that makes it so to move one side you go {[+0],[0-],[-0],[0+],[+0]...} and repeat, and to move the other way, you do the opposite {[+0],[0+],[-0],[0-],[+0]...}. it should be relatively easy to cast some abrasive with epoxy in the shape of the needed file, and file down a bar of ferrite with a holder going back and forth and a vacuum, or the other option is to get some laser/waterjet cut magnetic steel sheets, or use a small desktop cnc (e.g. a rigid enough 3d printer with a motor attached to it) with an 0.8mm drill to cut the steel. for magnets you can use 1x10x5mm bar magnets. using microstepping (i.e. interpolate the values with a sinusoid) you should be able to get some fairly smooth motion. also you would probably be able to use a dedicated driver, but something like an uln 2003 would probably do the trick. though it's probably overkill, but it could be an interesting way to go about it and you don't have to worry about the transmission of power. something that could be interesting regardless is to use springs from clothespins for the spring grip, which means you only need tension in your transmission, and fixes the issue of stiffness in the sleeve.
i have my own little project where I've also been experimenting a little with transferring motion through tubes. You should look into bike gearbox wires
When I tried making a finger actuator, I used a single "motor" to contract the fingers and simple springs/rubber bands to extract it again. It gives much less dexterity, but do we really use individual joint control? I'm not sure, perhaps such a simple design doesn't allow for what you're looking for, but its so much simpler. Basically, the tip of the finger is connected by a wire through each joint so that pulling on the wire, the finger contracts. As I'm doing test using my actual fingers, I can wiggle the tip a little bit, but the rest of the finger seems jointly controlled? Anyway, you Bionic Hand is looking very good indeed!
This was such a good thing to see you working on this huge project again! 😊 Btw where did you get the nylon shoulder washers? Thanks a lot for the plugin idea, I always looked for an idea to send from Unity to serial... I can also recommend sending it to a server and then processing your data. I am using Python and running a Flask Server.
Since you have your potentiometers in the finger joins anyway, isn't it sufficient to constrain the cable paths all the way to the motors instead of using floating guide tubes ? Sure, by having all the joins be influences by the bigger ones it caps the speed you can run the big ones at unless you manage to make the cables run tight to the pivot points to not affect the joins down the line too much.
THe ideal conduit material is continuous length extension spring - McMaster 9664K12. You could maybe stretch a bit out to gain some compliance, but that seems sketchier to me
Instead of a compliant mech, how about replacing the servo with a BLDC? The angle is checked at the joint, so if you can replace the servo with a stepper or a small brushless controlled by a smooth driver, you could have better feedback. IDK if enough torque could be applied without a gearbox tho.
This is really interesting! You can improve precision of the servos by using magnetic encoders instead of potentiometers (this requires a new control board though). Also, would using BLDC motors with FOC drivers be viable in your design? If it works, you could set virtual boundaries and virtual tension instead of using springs in the motion transmission. Great work!! Looking forward to updates!
Hello Will, I run a robotics startup developing a bio-mechanical accurate artificial muscle. We are looking for a small group of developers/people to test and provide feedback on our tech. I came across your RU-vid videos from a few years ago and felt this was right up your alley. Please let me know if you would be interested in learning more about what we are doing.
I wish I could take 2 weeks off of work... lol! Anyways, the issue that I've always found with this design is the "finger servo motors" considering how their tension would always need to be adjusted. That and how independent each contol motor is (if this were to be for real world applications). Lastly, anything that isn't LTI is just pointless (like most arduino projects are sadly).
Im hoping to make 2 hands for my neighbor (hes a double amputee) but my takeaway from this is that this isnt really close to being practically applicable yet. It just seems like theres too much that could break.. its awesome and maybe ill just use the fingers but seems just too complicated for me if i ever have to fix it
Great work 😊 Ptfe tubes … are they the same as in 3d printers? I know they work great under tension… never seen them in compression like the way you use them. Bodem tubes under compression are bike brakes and shifters, but they are metal.
*Hello! AI Image Generator here 👋🏼* Just a quick question how do I connect the 6th and 7th fingers??? Can’t we use a cartoonish hand instead with one ☝🏼 finger less you know how cartoons have only 5 fingers right??? Plus the thumb… it seems odd to have so few fingers and the thumb… don’t forget to set a random seed for more realistic levels of thumb placement instead of always placing it on the left side of each hands now you can have the natural look of real thumbs that are naturally occurring on random side of the hands not just on the left… 😅😅😅😅 thumbs seems to have an average placement that converge in equilibrium between right and left the more you do…
try to use bowden cables for the cable joints, those will do what you need quite well with very high durability. Only issue is that they will need to be custom order sized due to design.
