A RUclipsr I believe named Will Cogley solved the cable lengthening problem by using a servo and having the encoder at the end joint so regardless or stretch the servo would try to maintain position.
Between that, and all the DIY/OS Robotic Arms being made on RUclips right now, i seriously think we need to meet up and share+compare designs and whatnot.
Also Skyentific NEEDS to strap that hand as an end effector one day lol. (Then put it on James Bruton's "open dog" or some bipedal robot (l o n g t e r m goofyness unlike use as an end effector) ?)
FYI, if you design the holes printed in this orientation with a "tear drop" shape, or just a small triangular wedge cut out of the top, the result on FDM printers will will become much more circular. Can save a lot of time with complex geometry that's difficult to post-process by hand. Great video!
@@TDOBrandano Yeah and some slicers basically do this on their own. That extra bite is great when you're threading screws into plastic parts but still not ideal when clearance is needed for projects like this.
@Lassi Kinnunen 81 I've also found adding a small cutout to the bottom of really small holes can help even more depending layer height. I believe Cura basically adds a bridge automatically to most horizontal holes but it would be nice to adjust how steep of an angle it starts. Changing the orientation is a good thing to note. It's just not always practical and may affect how reliable the overall dimensional accuracy and part strength is. I've been looking into creating a slicer plugin that automatically applies an optimal tear drop to holes between a set diameter. It may exist already but I haven't seen one. I'd also like to have bridging options for holes that'll be threaded and automatic sacrifical layers for recessed vertical holes .
material types make differ but kuka and others solved this problem using strong timing belts. If this worked for robot arms probably all companies would use it.
I like the wrist design. When I saw your design I had the following idea. Replace the cables with belts. A belt runs from a pulley on the wrist to a double pulley on the elbow, then from this pulley to the motor. You can even use the double pulley for gear reduction. Now you can have three motors near the shoulder. Elbow motor pulley -> belt -> elbow rotation. Wrist pitch motor -> belt -> pitch double pulley -> belt -> wrist pitch rotation. Wrist yaw motor -> belt -> yaw double pulley -> wrist yaw rotation. The double pulleys are mounted as idler on the elbow axis.
That's really great design despite the poor performance of the Bowden cables! You can see that great care was taken and a lot of thought was put in. Also the video is well made.
I've thought of things like this for years, and often referred to such things as "fly-by-wire prosthetics", using cables to control limb movements while placing the heavy and bulky motors/servos elsewhere. Using Bowden cables as you have is probably the most problematic way to do so. Instead of trying to manage the whole "they get longer when bent" thing, just keep the cables straight at all times. If the cable needs to go around the elbow, use a transfer mechanism to connect two cables, one in the forearm and one in the upper arm, that moves along the same pivot the elbow does. When the elbow bends, use software to ensure the other joints don't move by adjusting accordingly. The actuation of the cables shouldn't be done by wrapping them around a drum as this itself causes the stretching issue. Instead, actuate them with lever arms, or mounting them on the surface of a wheel/disk/circle/sandwiched disks to ensure the cable stays straight. Yes, it means you need a larger disk to actuate the cable a given length as compared to wrapping around a drum, but it means there's bending effects. This should also reduce the amount of wear that occurs on both the cable, and the actuator. The use of cables to transfer motion does work. Airplanes have used this technology (called Fly-By-Wire, hence the name I use) for almost as long as planes have been a thing. The important bit is that the distance between the actuator and the controlled surface be fixed as much as possible. This means having a structural forearm, elbow, and upper arm (which incidentally is just known as "the arm", but that's confusing...), and likely shoulder to keep everything held in place. It's also important to have the extension and retraction cable anchors to be connected to each other so that extending one retracts the other by the same amount. It's also worth mentioning that this type of device would be best suited to something such as a human hand, as it requires multiple cables per finger to give it the same or similar range of movement that a normal hand has. If you're willing to revisit this idea, It would be interesting to see how not bending the cables at all works out.
no it is okay to bend them, just have to keep them bent consistently at the same bend and not have them bending during use then unbending again. This way they aren't bending at all any more than a preset bend that is kept consistent. Also, he has more friction since he's using thick metal cable rubbing on a metal interior. He should use teflon tubing with a PE fishing line instead which will cut down friction. He also uses ZERO lubricant on the cable which also greatly would reduce friction.
There is a reason why Bowden cables are used on bikes and not multi-axis robotic mechanisms. "Yes, it seems like Bowden cables sucks" pretty much sums it up @ 15:50. Keep up the good work sir!
I just started thinking about this yesterday. Thanks for the physical proof. My research on bowden cables came to the same conclusion without making anything.
when i was doing this about 12yrs ago now,i used car window winding mechanisms , the amount of flex in cable was my advantage as it gave very smooth and lifelike movements to the neck and head of a 4 meter long robot butterfly
Wow! This is what I've been waiting for! Classic skyentific video with another cool ingenious design to be tested! Thanx a lot for inspiration to complete my own robotix projectz
I experimented with spectra fishing line running through ptfe tubes, it could only pull of course so they had to be used paired, but it is so slippery, you can run both down the same tube and they slide past each other.
@@freakinccdevilleiv380 I used a spool from my gus Simpson build to actuate the spectra line with a few dozen wraps around it couldn't go forever in one direction but could go several rotations in each direction which was more than enough to wiggle an arm. but no technical reason you wouldn't be able to do a loop for full continuous rotation forever I think. You would need to add a guide to keep the thread from walking off the end of the rollers.... Or actually, if you just loop around a couple times that may not be an issue. Worth experimenting with!
yes this is a much better materials choice. I prefer PE braided fishing line over spectra. It comes in 300lb test down to 10lb test and everything in between. Might even go up to like 400lb test and up a bit more 500lb or 600lb I think is around the stopping point. But yes TPFE teflon tubing. This is what I'm using in my robot.
Great project idea, well done. Probably the best is a hydraulic fluid system or maybe an air vacuum pump like used in the Mercedes central door locking system, very little friction and waterproof. For example 2 syringes connected with a small bore plastic pipe. giving you instant push/pull linear motion.Keep up the good work.
You could fix the extension problem in software. When its attached to an arm its linearly proportional to the bend of the arm, so should be quite easy to calculate. The friction problem is solved by using pulleys and bearings, instead of tubing. (there's a lot of friction present in those tubes.)
you know i cant help but think that bowden cables and the concept of tensegrity should be able to work really well together to make some pretty strong but flexible robots
I totally agree with you. I thinking in this direction for the long time, but have not find any good solutions yet. But I agree that this should make a great design
That spring is a great idea.. The problem with the flex of the block can probably solved by a dovetail pattern on the base and having the cable retainer slide.. Or have a small linear rail holding them.
some ideas i had that might alleviate some issues -springs on both ends of the tubes so that all the travel isn't just on one end of the system should help a little -spring in the middle instead [or in addition to the others], more specifically, at the hypothetical elbow, we basically just wanna give the tube as much room to expand and contract as possible and spread the expansion and contraction out, kind of like a real muscle -concentrating the springs in places where the cable does a lot of flexing should also keep the tube from needing to compress as much, as the cable has more room to find a better path
Should be able to do precise positioning if you put encoders on the wrist joints themselves, so the cable length change can be compensated by moving the steppers. Friction will still be a problem, but at least to start with, simply using larger steppers would solve it. In the long run you may have trouble with cables/tubes wearing out. Bicycles only apply significant tension to the cables after the brakes have engaged, and then there's not much more movement of the cables until you release it. But on a robot you'll be continually sliding the cables in the tubes while under tension.
A couple thoughts here. 1. if you move the motor half way - at the elbow - you still reduce its influence on moment of inertia at the shoulder by a factor of two which is significant. 2. this way transmission from "elbow" to "wrist" does not need to negotiate bends so simple cable pulleys aka capstans or belts are fine, which means is less friction, less flexing and option of having a lightweight reduction gear at the wrist with motor (and first stage reduction) at the elbow, if that feels appropriate.
You are right. By 'integrating' the cables before the sensor, you include them in the closed loop system and the control system can compensate for the non-idealities. It is of course not as good as if it would not have any non-idealities, but better than using the cables in 'feedforward'. You will always waste (precious) motor power overcoming resistance, leading to lower efficiency, not being able to reach theoretical torques/speeds and motors getting hot. Play/backlash is something a control system cannot really compensate for very well, and can even lead to control system instability.
@@tHaH4x0r To pick up some mug and to place it in another place - this is more than enough. Really good lubrication and maybe bigger diameter and more flexible cables - can improve efficiency - like in human body : D
@@humanspirit3432 If you can improve the mechanical system of the arm, why not do so? It seems like a waste of effort to design such a beautiful arm and 3D prints, to then not nicely finish it up in the end. Also, the bigger diameter tubes will only exacerbate the issues with bowden tube getting stiffer when bent, because this is mainly caused by the difference in length between inside and outside of the cable. The resistance is also not caused by friction inside the tubes, but by the cables working against each other. This means that lubrication does not help at all either. More flexible cables would maybe help, but this would come at the cost of control system bandwidth, as this elasticity is directly in series with the actuator.
@@humanspirit3432 To elaborate on my last comment, what is the point of putting the motors far away from the joint itself? It is to decrease arm inertia. Why do we want to decrease arm inertia? Improved bandwidth and increased safety. If you then have a more elastic link between motor and joint you will decrease the bandwidth, yielding probably worse performance as a result, with higher mechanical complexity (meaning more cost, lower reliability) so it makes a lot of sense to want the mechanical part to be operating without slip, resistance and other non-idealities, as otherwise you are just handicapping the arm and negating any benefits you are trying to get.
@@tHaH4x0r I mean - cables must be flexible for bending, but not flexible for stretching. Steel cables with many wires - are OK. Maybe some synthetic cables will work also good.
I have played with something of the same kind in a smaller version But instead I used braided fishing line And Servo motor But I have a problem with it moving together when it moved as you showed. I moved Potentiometer out in the Joint and it worked really well
@@kingmasterlord shaping the rollers which are guiding the wires in a way to accommodate for the position dependent torque - this would be worth a try 👍🏼 The question is if such a artificial muscle can achieve reasonable dynamics
@@EngineeringSpareTime well I was figuring on how to take this stuff www.fibreglast.com/product/24K_Carbon_Tow_2293/carbon_fiber_all basket-weaved around strips of alternating rows of these cables and inflatable artificial muscles in the same way muscle tissue and ligaments are. combine the bow risers and some gearboxes and pulley ratios and the small action of the muscles can have big results. I've got these 4 ft fiberglass rods that I'm going to make into bones by putting them in a carbon fiber sleeve with some wiring conduits, the reduction cables can go inside of it too..maybe through some 4040 extrusion in the center third
The problem that you have is that you transfer the whole 100% force through the cables. 1) make cable to drive a system of blocks of somethign like that and you won't need to tension it too hard. 2) small change in length of bent bowden is divided by the reduction rate of your gearing mechanism. 3) use teflon coated gearshifting ca bles from bikes (those are thinner, flexier and much smoother). It is suboptimal to pull the entire torque by these cables.
A tip for the holes, you can you use a tear drop shape at the top of your holes to prevent hole collapse. Saw it on maker's muse and it works quite well. I haven't done it yet on that small of a whole, because usually I dont need it for that small of a whole as its using for screws
You should try doing something similar to the mechanism of a ‘flexible shaft Dremel’. So instead of the cable moving forward or backwards like the Bowden system, rather use rotation in the cable. Have always wondered if anyone has used flexible shaft mechanisms for robotics.
Those cables are good for the Dremel application (constant twisting/spinning), but for robotics it's not going to be good because they bend/twists too much, so you'll loose precision (backlash). I'm pretty sure I saw a guy that made a thicker version of those cables that reduces the twisting... But can't remember the application... and it's much more rigid.
Bicycle shifter cables are built differently than brake cables for exactly the reason you show. Brake cables use a tightly coiled spring and contact between the turns to support the reaction loads of the drive cable. Shift cables uses a parallel wires with a twist every 1-2cm stabilized by the plastic jacket. This isn't as strong, but it's stiffer and doesn't change length when flexed.
I'd have a try at using bicycle gear cables instead of break cables, the housings are constructed in a different way to minimise the issue of the length increase as it bends and the cables are thinner and more flexible.
How about analyzing "rotational bowden wires" ? Removing the length from the equation and only work with rotation of the wire? Guess you need higher speed on the wire and gear reduction at the hand to get some torque.
Yes, this is great idea. But with this you cannot move the gearbox. The gearbox would have to stay close to joint. (Because, such cables have some “backlash”)
@@Skyentific have seen some 3d printer s working with this kind of rotation wire. Moving the high weight from the important point. Not sure on how they solve the backlash. Worm gear connected in two stages? But yes, gearbox is needed in the wrong end of the cable
Not a fan of flex-shafts either. They tend to have very low stiffness and significant friction. Much better to snake a timing belt in with a bunch of idlers or use a shaft with a clever gear layout.
Small hydraulic cables with a cylinder attached to a wire in the end and you can draw the lines how ever you want ( like hydraulic steering in a boat )
Since the cable will not be free-floating like this in the final product, and you can calculate the cable path from other motor positions, probably you can compensate for the cable-lengthening in software.
As a former bicycle mechanic, I know that the cables stretch and housing needs to be "seated" a bit when they are brand new. After some initial stretching the cables and housing will no longer have show the gaps that you are experiencing.
I'm wondering if you use a smaller circumference at the motor, the effective gear reduction would help overcome the friction. Very nice design, too bad it doesn't work better. Thank you for trying it, a design I've been thinking of myself. That saves me some time!
I was thinking if he put a gear reduction between the cable driven shaft to the arm past the previous degree of freedom, this could help with the change of length of cable when boden tube bends.
Have you thought about using bigger bowden cables, like Teleflex cables you use on small outboard boats with steering wheels, they are made for a lot more power. Maybe there is a small version that is not too big(or expensive)
You may replace the Bowden tubes with big PTFE tubes and steel wire or dynema. They can carry less force overall but don't have the problem of the metal spring, which makes the Bowden tubes longer when bent. Or incorporate a spring system which makes both ends slightly flexible. Then you also have some kind of shock protection for the gearbox. And if you measure the movement of the springs, you can also do force control :)
Very detailed work. Love it. However, it sounds to me like solving this stuff in software is the long term optimal solution for a lot of the issues introduced by these cables.
Thanks for the informative video. It was a nice demonstration. I would like to request the CAD files of the bowden driver since I would like to try doing this experiment.
Cool experiment! Have you considered something like beaded chain cord that’s used in drapery? If they can hold up to the high tension it might work better.
install some encoders on the end effector axis and spin the motors in such manner in order to compensate for the movement of the cable inside the bowden tubes. thanks for the video!
Bowden cables are generally intended to be used as push / pull cables. IOW a single cable. You are using them in a pull / pull arrangement. With the pull / pull setup do you really need one long continuous outer tube? It seems like bearing surfaces for the cables -- think pulleys -- at various points along the arm would be fine.
On bicycles, there's a tensioning, screwable mount that goes over the bowden tube on the outside. This solves your problem of the cable lengthening/shortening. "Stelbout remkabel"
If you use 2 steppers per Bowden Pair ... I think on an arm/joint/bend, the movement could be calculated close enough that you wouldn't need the springs and could make it more rigid. This would also allow you to put springs on the effector/joint side on the end of each cable push/pull, and you could adjust compliance/rigidity in realtime by adjusting/matching the the cable lengths independently.
I like how bicycles overcome this by using a long-enough bowden for each bend that changing the bend doesn't change the bowden length much - and then for any straight long sections it relies on the compression strength of the arm to run the cable in a straight line. Brake handle > top of bike frame, one sweeping curve that doesn't change length much when you steer Top of frame> fixed point to bottom of frame with no tube, just open cable Bottom of frame > brake actuator is straight line with no changing bends. I think a lot of the problems can be overcome by relying on this kind of geometry.
This system simply needs a closed loop control. No need to try to address the length variation on the mechanical design side. Slap encoders on the joints. It will also tackle the friction variation by just applying more torque.
Assuming you can solve the bowden cable "stretch" problem, I think this could work although as you have pointed out, it has drawbacks. Perhaps you need independent blocks for each tube rather than a single block for both? Once that problem is solved, tracking the position of the end effector is all you need to prevent this problem from causing unwanted motion on the outputs. Perhaps hydraulics are the way to do then...
A possible extension could be to use a bowden cable loop (for example 5 turns) with a reducer at the wrist joint. This could greatly increase the torque and accuracy despite the flaws of the bowden tube. Another type of cable way (pulleys) would solve the tensionning problem. A stiff pretensionned polymer cable / rope could also work better in a loop system.
Great video! I think you are there! Maybe I'm being too optimistic, but I think that the movement of the cables when they're bent could be modelled. Then, with the rotation of the joints between the steppers and the articulation, you always would know the change of length of the cable. So each time an intermediate joint moves, you should compensate the change in length by rotating that amount in the stepper.
Can that be a feature rather than a problem. Wanted or not, it's feedback. You could perhaps determine position and force based on that cable "backlash"? Also, can single-sided help? Boden cables aren't very good transmitting a "push", but they _do_ work a bit. Might be worth an experiment: half friction at a cost. Also removes the non-parallel motion of the spring block. Think of an alligator's mouth: really weak to open, but damn good at closing. It's just a matter of finding the application that isn't concerned for the limitations. An end-effector gripper, do you care you have no power opening the grip?
I had same idea, thanks to keep me from frustrating do it by my self. I am also not convinced. But then we use it for the gripper? This sound more usable.
Perhaps a solution to the bowden backlash so to speak (where bending changes length), is to measure at the end and closed loop correct for it. The motors could still be at he base and you'd only need encoders at the end still cutting down on end effector weight.
Agree, even with the friction and the "backlash", it still carries the force pretty well, so measuring the position at the moving end rather than the motor end makes sense. I had the same issue with my canards, but fixed it with potentiometers, and generalised the principle to all my ailerons/elevons, even using chain belts and tensioners from my servos.
The friction issue could potentially be fixed by some kind of lubrication system. It's not an ideal solution, but it would work in industrial applications where industrial machinery is already lubricated on a scheduled basis. Alternatively, different materials could be used, such as a jacket with a higher material lubricity, or a cable with a higher material lubricity. The actuation issue that occurs when the tube and cable are coupled should be a non-issue, as the tubes should only be there as a guide, not coupled to the cable. If you're rigidly mounting the tubes to a frame, you should add a spring, or a bellows if you need the cable to still be enclosed, this would allow for stretch and compression in the tube, which would allow the cable to not actuate when the tube bends, due to decoupling of the tube and cable. Your spring solution was close to working, but wasn't implemented properly, you want springs on the tubes themselves, so that the tubes essentially 'float' between the two ends of the robot's frame. What needs to be tensioned is the cable itself, not the tube. Think about water running through a pipe, the pipe is rigid while the water is not, the water is free-flowing within the pipe; now inverse this logic, where the interior material is the rigid component, the conduit around the rigid material needs to be flexible in comparison. This is exactly the situation here, the Bowden tube needs to be flexible in comparison to the rigid Bowden cable, and to achieve that you need the Bowden tube itself to be held in a sprung suspension when it's coupled to the frame of the robot; as it currently stands in the design, even with the spring tensioner, you are making both the tube and cable rigid components, when the tensioned cable should be the only rigid component. The Bowden actuator will work, it just needs design refinements and proper implementation of the components. I hope you read this, are able to understand this (I'm not the best at explaining things), and implement the latter suggestion in a version 2 of this robot arm. I know the former suggestion, the one concerning lubricity, is much harder to achieve, if not outright impossible unless an off-the-shelf component already exists; but the latter suggestion, the one concerning how the Bowden tube is implemented, where it should be a sprung suspension rather than a rigid component, should be easy to implement, just braze/solder a spring to the metal crimp at the end of the tube, then set the other end of the spring in a hole. Implementation of this would require the arm to have a skeleton, as the Bowden tube is not a floating piece located only by a receiving hole, and could fall out in a test setup that does not have the arm's skeleton in place.
Please do a v2 with derailleur cables. They don’t change length. And you can get thicker Bmx brake cable that use this type of outside casing and has nice Teflon inside lining.
What about cutting the Bowden cables from center, then a little bit shorten them, then design a spring mechanism in between to connect 2 bowden cables? This way the spring movement will only occur on the Bowden cables and motor section does not need any moving parts.
The bending/straightening contraction/expansion problem COULD be compensated for in software ... because the controller knows every joints position. Right?
Nokon cable housing may work better, it is aluminium segments with ball joints, the difference in radius between the wire and the housing may be less when bending.
I'm curious if Dyneema shouldn't be used instead, extremely low friction very low elasticity yet stronger and more bendable then cables. You would have to splice it to attach it to anything, but other than that a wonderful material for tension based systems.
So... if the tubes are 180 degrees in front of each other... it would help a lot? Could the friction be due to both tubes pulling in the same direction?
You can account for bend in software so the steppers adjust the end of the cable and zero out bend’s effect. Calibrate the thresholds once the bend is consistent based on the rest of the arm.
It's the same pulley that pulls and pushes the two cables at the same time, so it can't be compensated with software. Unless you have two separated pulleys with one motor each, one pushing and the other pulling at the same time, than you can compensate via software.
You should try to separate motor and gearbox with those bowdens. Try to put gearbox into the wrist, motor on the ground, and transfer rotation from the motor to gearbox with bowden. Btw, KUKA does this, but uses belts to transfer rotation from motor to gearbox.
Couldn't a cable adjuster and a rotation sensor be integrated to negate the extra cable movement? So that when the rotation doesn't match the cable movement the adjuster would adjust the cable.
I love how you start the video straight away. No intro, no speech just pure information. Keep it up.
A RUclipsr I believe named Will Cogley solved the cable lengthening problem by using a servo and having the encoder at the end joint so regardless or stretch the servo would try to maintain position.
I was going to suggest that. Ill check that video out!
Between that, and all the DIY/OS Robotic Arms being made on RUclips right now, i seriously think we need to meet up and share+compare designs and whatnot.
Also Skyentific NEEDS to strap that hand as an end effector one day lol. (Then put it on James Bruton's "open dog" or some bipedal robot (l o n g t e r m goofyness unlike use as an end effector) ?)
that shaft collar idea and using a long bolt. genius. stealing that for my daily work thank you!
zero nonsense straight to the point about beatifully engineered parts and beatiful looking prints. love him
FYI, if you design the holes printed in this orientation with a "tear drop" shape, or just a small triangular wedge cut out of the top, the result on FDM printers will will become much more circular. Can save a lot of time with complex geometry that's difficult to post-process by hand.
Great video!
Great idea! Thank you!
You can also make the very top a bridge, so print the upper half as a half hexagon. Will work just as well, and put more material around the hole
@@TDOBrandano Yeah and some slicers basically do this on their own. That extra bite is great when you're threading screws into plastic parts but still not ideal when clearance is needed for projects like this.
@Lassi Kinnunen 81 I've also found adding a small cutout to the bottom of really small holes can help even more depending layer height. I believe Cura basically adds a bridge automatically to most horizontal holes but it would be nice to adjust how steep of an angle it starts. Changing the orientation is a good thing to note. It's just not always practical and may affect how reliable the overall dimensional accuracy and part strength is.
I've been looking into creating a slicer plugin that automatically applies an optimal tear drop to holes between a set diameter. It may exist already but I haven't seen one. I'd also like to have bridging options for holes that'll be threaded and automatic sacrifical layers for recessed vertical holes .
There were teflon lined bowden tubes for racing bikes in the 80s, it was a difference like day and night, compared to the conventional tubes.
material types make differ but kuka and others solved this problem using strong timing belts.
If this worked for robot arms probably all companies would use it.
I like the wrist design. When I saw your design I had the following idea. Replace the cables with belts. A belt runs from a pulley on the wrist to a double pulley on the elbow, then from this pulley to the motor. You can even use the double pulley for gear reduction. Now you can have three motors near the shoulder. Elbow motor pulley -> belt -> elbow rotation. Wrist pitch motor -> belt -> pitch double pulley -> belt -> wrist pitch rotation. Wrist yaw motor -> belt -> yaw double pulley -> wrist yaw rotation. The double pulleys are mounted as idler on the elbow axis.
That's really great design despite the poor performance of the Bowden cables! You can see that great care was taken and a lot of thought was put in. Also the video is well made.
Great content, thanks! You should check rolling diaphragm actuators. This is a hydraulic actuation with barely any backlash and high backdrivability.
use derailleur tube instead of braking tube can solve the bending-shrinking problem, but the tube would be much stiffer
Other option is linear brake cable housing that BMXs use. its just a stronger version of derailleur cables.
I've thought of things like this for years, and often referred to such things as "fly-by-wire prosthetics", using cables to control limb movements while placing the heavy and bulky motors/servos elsewhere. Using Bowden cables as you have is probably the most problematic way to do so. Instead of trying to manage the whole "they get longer when bent" thing, just keep the cables straight at all times. If the cable needs to go around the elbow, use a transfer mechanism to connect two cables, one in the forearm and one in the upper arm, that moves along the same pivot the elbow does. When the elbow bends, use software to ensure the other joints don't move by adjusting accordingly. The actuation of the cables shouldn't be done by wrapping them around a drum as this itself causes the stretching issue. Instead, actuate them with lever arms, or mounting them on the surface of a wheel/disk/circle/sandwiched disks to ensure the cable stays straight. Yes, it means you need a larger disk to actuate the cable a given length as compared to wrapping around a drum, but it means there's bending effects. This should also reduce the amount of wear that occurs on both the cable, and the actuator.
The use of cables to transfer motion does work. Airplanes have used this technology (called Fly-By-Wire, hence the name I use) for almost as long as planes have been a thing. The important bit is that the distance between the actuator and the controlled surface be fixed as much as possible. This means having a structural forearm, elbow, and upper arm (which incidentally is just known as "the arm", but that's confusing...), and likely shoulder to keep everything held in place. It's also important to have the extension and retraction cable anchors to be connected to each other so that extending one retracts the other by the same amount. It's also worth mentioning that this type of device would be best suited to something such as a human hand, as it requires multiple cables per finger to give it the same or similar range of movement that a normal hand has.
If you're willing to revisit this idea, It would be interesting to see how not bending the cables at all works out.
no it is okay to bend them, just have to keep them bent consistently at the same bend and not have them bending during use then unbending again. This way they aren't bending at all any more than a preset bend that is kept consistent. Also, he has more friction since he's using thick metal cable rubbing on a metal interior. He should use teflon tubing with a PE fishing line instead which will cut down friction. He also uses ZERO lubricant on the cable which also greatly would reduce friction.
There is a reason why Bowden cables are used on bikes and not multi-axis robotic mechanisms. "Yes, it seems like Bowden cables sucks" pretty much sums it up @ 15:50. Keep up the good work sir!
For holes that are aligned vertically, I find that a mild tear drop shape helps get them close enough for small holes.
This should work! Thank you for idea!
I use a hexagon for vertical aligned holes. The principle is the same as the tear drop.
Awesome design as always!
Ah That is why I'm subscribed to you, this kind of video!
i'm 50seconds in and that's exciting!
I subbed for the pulli 😂
To avoid the problems you mentioned try to use a hydraulic system - similar to the bicycle hydraulic brake system instead of Bowden ropes
I just started thinking about this yesterday. Thanks for the physical proof. My research on bowden cables came to the same conclusion without making anything.
well you then condemned an approach that if done correctly is going to be a game changer. Your research failed you. He executes poorly in this video.
when i was doing this about 12yrs ago now,i used car window winding mechanisms , the amount of flex in cable was my advantage as it gave very smooth and lifelike movements to the neck and head of a 4 meter long robot butterfly
Wow! This is what I've been waiting for! Classic skyentific video with another cool ingenious design to be tested! Thanx a lot for inspiration to complete my own robotix projectz
Despite the fact that bowden cables sucks, it's very nice to see a proof for that)
I experimented with spectra fishing line running through ptfe tubes, it could only pull of course so they had to be used paired, but it is so slippery, you can run both down the same tube and they slide past each other.
That's awesome, you could have fully rotating effectors on the ends and a single spring for length compensation.
@@freakinccdevilleiv380 I used a spool from my gus Simpson build to actuate the spectra line with a few dozen wraps around it couldn't go forever in one direction but could go several rotations in each direction which was more than enough to wiggle an arm. but no technical reason you wouldn't be able to do a loop for full continuous rotation forever I think. You would need to add a guide to keep the thread from walking off the end of the rollers.... Or actually, if you just loop around a couple times that may not be an issue. Worth experimenting with!
yes this is a much better materials choice. I prefer PE braided fishing line over spectra. It comes in 300lb test down to 10lb test and everything in between. Might even go up to like 400lb test and up a bit more 500lb or 600lb I think is around the stopping point. But yes TPFE teflon tubing. This is what I'm using in my robot.
Great project idea, well done. Probably the best is a hydraulic fluid system or maybe an air vacuum pump like used in the Mercedes central door locking system, very little friction and waterproof. For example 2 syringes connected with a small bore plastic pipe. giving you instant push/pull linear motion.Keep up the good work.
You could fix the extension problem in software. When its attached to an arm its linearly proportional to the bend of the arm, so should be quite easy to calculate.
The friction problem is solved by using pulleys and bearings, instead of tubing.
(there's a lot of friction present in those tubes.)
Right to the point! :D "Hello and welcome to my channel - It is important to keep the weight of the robot arm as low as possible!"
you know i cant help but think that bowden cables and the concept of tensegrity should be able to work really well together to make some pretty strong but flexible robots
I totally agree with you. I thinking in this direction for the long time, but have not find any good solutions yet. But I agree that this should make a great design
That spring is a great idea.. The problem with the flex of the block can probably solved by a dovetail pattern on the base and having the cable retainer slide.. Or have a small linear rail holding them.
some ideas i had that might alleviate some issues
-springs on both ends of the tubes so that all the travel isn't just on one end of the system should help a little
-spring in the middle instead [or in addition to the others], more specifically, at the hypothetical elbow, we basically just wanna give the tube as much room to expand and contract as possible and spread the expansion and contraction out, kind of like a real muscle
-concentrating the springs in places where the cable does a lot of flexing should also keep the tube from needing to compress as much, as the cable has more room to find a better path
Thanks for showing us your experiences.
Should be able to do precise positioning if you put encoders on the wrist joints themselves, so the cable length change can be compensated by moving the steppers. Friction will still be a problem, but at least to start with, simply using larger steppers would solve it. In the long run you may have trouble with cables/tubes wearing out. Bicycles only apply significant tension to the cables after the brakes have engaged, and then there's not much more movement of the cables until you release it. But on a robot you'll be continually sliding the cables in the tubes while under tension.
A couple thoughts here.
1. if you move the motor half way - at the elbow - you still reduce its influence on moment of inertia at the shoulder by a factor of two which is significant.
2. this way transmission from "elbow" to "wrist" does not need to negotiate bends so simple cable pulleys aka capstans or belts are fine, which means is less friction, less flexing and option of having a lightweight reduction gear at the wrist with motor (and first stage reduction) at the elbow, if that feels appropriate.
Bowden cables are good. Actually you can compensate most of bad effects using encoders and software compensation.
You are right. By 'integrating' the cables before the sensor, you include them in the closed loop system and the control system can compensate for the non-idealities. It is of course not as good as if it would not have any non-idealities, but better than using the cables in 'feedforward'.
You will always waste (precious) motor power overcoming resistance, leading to lower efficiency, not being able to reach theoretical torques/speeds and motors getting hot.
Play/backlash is something a control system cannot really compensate for very well, and can even lead to control system instability.
@@tHaH4x0r To pick up some mug and to place it in another place - this is more than enough. Really good lubrication and maybe bigger diameter and more flexible cables - can improve efficiency - like in human body : D
@@humanspirit3432 If you can improve the mechanical system of the arm, why not do so? It seems like a waste of effort to design such a beautiful arm and 3D prints, to then not nicely finish it up in the end.
Also, the bigger diameter tubes will only exacerbate the issues with bowden tube getting stiffer when bent, because this is mainly caused by the difference in length between inside and outside of the cable. The resistance is also not caused by friction inside the tubes, but by the cables working against each other. This means that lubrication does not help at all either.
More flexible cables would maybe help, but this would come at the cost of control system bandwidth, as this elasticity is directly in series with the actuator.
@@humanspirit3432 To elaborate on my last comment, what is the point of putting the motors far away from the joint itself? It is to decrease arm inertia. Why do we want to decrease arm inertia? Improved bandwidth and increased safety. If you then have a more elastic link between motor and joint you will decrease the bandwidth, yielding probably worse performance as a result, with higher mechanical complexity (meaning more cost, lower reliability) so it makes a lot of sense to want the mechanical part to be operating without slip, resistance and other non-idealities, as otherwise you are just handicapping the arm and negating any benefits you are trying to get.
@@tHaH4x0r I mean - cables must be flexible for bending, but not flexible for stretching. Steel cables with many wires - are OK. Maybe some synthetic cables will work also good.
This was really good. Thanks for the quality content!
I have played with something of the same kind in a smaller version But instead I used braided fishing line And Servo motor But I have a problem with it moving together when it moved as you showed.
I moved Potentiometer out in the Joint and it worked really well
Nice! I thought about trying this before as well :)
me too, did you consider compound bow risers and artificial muscles? I'm thinking electrolysis could maybe inflate the muscles
@@kingmasterlord shaping the rollers which are guiding the wires in a way to accommodate for the position dependent torque - this would be worth a try 👍🏼
The question is if such a artificial muscle can achieve reasonable dynamics
@@EngineeringSpareTime well I was figuring on how to take this stuff www.fibreglast.com/product/24K_Carbon_Tow_2293/carbon_fiber_all
basket-weaved around strips of alternating rows of these cables and inflatable artificial muscles in the same way muscle tissue and ligaments are.
combine the bow risers and some gearboxes and pulley ratios and the small action of the muscles can have big results.
I've got these 4 ft fiberglass rods that I'm going to make into bones by putting them in a carbon fiber sleeve with some wiring conduits, the reduction cables can go inside of it too..maybe through some 4040 extrusion in the center third
Nice design btw
you could send power down the lines too! This design is really neato
This is beautiful idea!!! Thank you!
The problem that you have is that you transfer the whole 100% force through the cables.
1) make cable to drive a system of blocks of somethign like that and you won't need to tension it too hard.
2) small change in length of bent bowden is divided by the reduction rate of your gearing mechanism.
3) use teflon coated gearshifting ca bles from bikes (those are thinner, flexier and much smoother).
It is suboptimal to pull the entire torque by these cables.
A tip for the holes, you can you use a tear drop shape at the top of your holes to prevent hole collapse. Saw it on maker's muse and it works quite well. I haven't done it yet on that small of a whole, because usually I dont need it for that small of a whole as its using for screws
You could add rotary encoders on the wrist. They have no weight.
I agree. Feedback on the end effector would eliminate the stretching causing inaccuracy problem.
You should try doing something similar to the mechanism of a ‘flexible shaft Dremel’. So instead of the cable moving forward or backwards like the Bowden system, rather use rotation in the cable. Have always wondered if anyone has used flexible shaft mechanisms for robotics.
Those cables are good for the Dremel application (constant twisting/spinning), but for robotics it's not going to be good because they bend/twists too much, so you'll loose precision (backlash). I'm pretty sure I saw a guy that made a thicker version of those cables that reduces the twisting... But can't remember the application... and it's much more rigid.
Bicycle shifter cables are built differently than brake cables for exactly the reason you show. Brake cables use a tightly coiled spring and contact between the turns to support the reaction loads of the drive cable. Shift cables uses a parallel wires with a twist every 1-2cm stabilized by the plastic jacket. This isn't as strong, but it's stiffer and doesn't change length when flexed.
I'd have a try at using bicycle gear cables instead of break cables, the housings are constructed in a different way to minimise the issue of the length increase as it bends and the cables are thinner and more flexible.
Yes, this is great idea. But there are two issues: gear cable is more rigid and it can transfer less torque.
@@Skyentific the rigidity is a good point, but I don't think the torque would be an issue, they can take 500N+ without breaking
How about analyzing "rotational bowden wires" ? Removing the length from the equation and only work with rotation of the wire? Guess you need higher speed on the wire and gear reduction at the hand to get some torque.
Yes, this is great idea. But with this you cannot move the gearbox. The gearbox would have to stay close to joint. (Because, such cables have some “backlash”)
@@Skyentific have seen some 3d printer s working with this kind of rotation wire. Moving the high weight from the important point. Not sure on how they solve the backlash. Worm gear connected in two stages? But yes, gearbox is needed in the wrong end of the cable
Not a fan of flex-shafts either. They tend to have very low stiffness and significant friction. Much better to snake a timing belt in with a bunch of idlers or use a shaft with a clever gear layout.
Small hydraulic cables with a cylinder attached to a wire in the end and you can draw the lines how ever you want ( like hydraulic steering in a boat )
You could try the bowden cables from a flexible shaft dremel-tool...
using encoders on output for closed loop can help solve problem with lenght change,and friction can be useful for holding position.
Since the cable will not be free-floating like this in the final product, and you can calculate the cable path from other motor positions, probably you can compensate for the cable-lengthening in software.
I had an idea for using this idea to drive an extruder motor 2 days ago. Coincidence is funky.
As a former bicycle mechanic, I know that the cables stretch and housing needs to be "seated" a bit when they are brand new. After some initial stretching the cables and housing will no longer have show the gaps that you are experiencing.
I'm wondering if you use a smaller circumference at the motor, the effective gear reduction would help overcome the friction. Very nice design, too bad it doesn't work better. Thank you for trying it, a design I've been thinking of myself. That saves me some time!
I was thinking if he put a gear reduction between the cable driven shaft to the arm past the previous degree of freedom, this could help with the change of length of cable when boden tube bends.
Have you thought about using bigger bowden cables, like Teleflex cables you use on small outboard boats with steering wheels, they are made for a lot more power. Maybe there is a small version that is not too big(or expensive)
You may replace the Bowden tubes with big PTFE tubes and steel wire or dynema. They can carry less force overall but don't have the problem of the metal spring, which makes the Bowden tubes longer when bent. Or incorporate a spring system which makes both ends slightly flexible. Then you also have some kind of shock protection for the gearbox. And if you measure the movement of the springs, you can also do force control :)
Oh, I hadn't watched till the end
Great video again, thank you... you have gone to sooo much work to show that what you initially thought was correct.
That is a work of art! Great video, very, very informative - thanks so much for providing this!
Very detailed work. Love it. However, it sounds to me like solving this stuff in software is the long term optimal solution for a lot of the issues introduced by these cables.
Very nice wrist build. Can you not ensure the cable only extends along the forearm? The elbow can have a cable-to-cable connector.
Thanks for the informative video. It was a nice demonstration. I would like to request the CAD files of the bowden driver since I would like to try doing this experiment.
Cool experiment! Have you considered something like beaded chain cord that’s used in drapery? If they can hold up to the high tension it might work better.
install some encoders on the end effector axis and spin the motors in such manner in order to compensate for the movement of the cable inside the bowden tubes. thanks for the video!
Bowden cables are generally intended to be used as push / pull cables. IOW a single cable. You are using them in a pull / pull arrangement. With the pull / pull setup do you really need one long continuous outer tube? It seems like bearing surfaces for the cables -- think pulleys -- at various points along the arm would be fine.
On bicycles, there's a tensioning, screwable mount that goes over the bowden tube on the outside. This solves your problem of the cable lengthening/shortening. "Stelbout remkabel"
If you use 2 steppers per Bowden Pair ... I think on an arm/joint/bend, the movement could be calculated close enough that you wouldn't need the springs and could make it more rigid. This would also allow you to put springs on the effector/joint side on the end of each cable push/pull, and you could adjust compliance/rigidity in realtime by adjusting/matching the the cable lengths independently.
How about adding some position encoders in the wrist to provide feedback and compensate for the changes in length as the arm bends?
I like how bicycles overcome this by using a long-enough bowden for each bend that changing the bend doesn't change the bowden length much - and then for any straight long sections it relies on the compression strength of the arm to run the cable in a straight line.
Brake handle > top of bike frame, one sweeping curve that doesn't change length much when you steer
Top of frame> fixed point to bottom of frame with no tube, just open cable
Bottom of frame > brake actuator is straight line with no changing bends.
I think a lot of the problems can be overcome by relying on this kind of geometry.
Awesome design, we like them.
perhaps you can make some 'guide-grooves' in the blue printed part connected to the motors to prevent the motion of the tension blocks
This system simply needs a closed loop control. No need to try to address the length variation on the mechanical design side. Slap encoders on the joints. It will also tackle the friction variation by just applying more torque.
awesome! Cool to see these explorations.
I thought you made that work quite well!
Assuming you can solve the bowden cable "stretch" problem, I think this could work although as you have pointed out, it has drawbacks. Perhaps you need independent blocks for each tube rather than a single block for both? Once that problem is solved, tracking the position of the end effector is all you need to prevent this problem from causing unwanted motion on the outputs. Perhaps hydraulics are the way to do then...
A possible extension could be to use a bowden cable loop (for example 5 turns) with a reducer at the wrist joint. This could greatly increase the torque and accuracy despite the flaws of the bowden tube. Another type of cable way (pulleys) would solve the tensionning problem. A stiff pretensionned polymer cable / rope could also work better in a loop system.
you could try to use a differential mechanism in the wrist. Whatever error is created, will probably be created equally in both cables!
Good implementation for a non critical movements like an actuator.
Awesome design as always! Thank you 👍😊
IME brake cables will also work harden, fray and eventually break.
Great video! I think you are there! Maybe I'm being too optimistic, but I think that the movement of the cables when they're bent could be modelled. Then, with the rotation of the joints between the steppers and the articulation, you always would know the change of length of the cable. So each time an intermediate joint moves, you should compensate the change in length by rotating that amount in the stepper.
correct, that's one of many things this video fails to acknowledge.
Can that be a feature rather than a problem. Wanted or not, it's feedback. You could perhaps determine position and force based on that cable "backlash"?
Also, can single-sided help? Boden cables aren't very good transmitting a "push", but they _do_ work a bit. Might be worth an experiment: half friction at a cost. Also removes the non-parallel motion of the spring block. Think of an alligator's mouth: really weak to open, but damn good at closing.
It's just a matter of finding the application that isn't concerned for the limitations. An end-effector gripper, do you care you have no power opening the grip?
I had same idea, thanks to keep me from frustrating do it by my self. I am also not convinced.
But then we use it for the gripper? This sound more usable.
Amazing work I learn so much from your videos
Perhaps a solution to the bowden backlash so to speak (where bending changes length), is to measure at the end and closed loop correct for it.
The motors could still be at he base and you'd only need encoders at the end still cutting down on end effector weight.
Agree, even with the friction and the "backlash", it still carries the force pretty well, so measuring the position at the moving end rather than the motor end makes sense. I had the same issue with my canards, but fixed it with potentiometers, and generalised the principle to all my ailerons/elevons, even using chain belts and tensioners from my servos.
you can use bike brake hose with teflon tube inside , used to reduce friction i think it's better
There are round saw blades for cutting tiles and ceramics, they should work better if you need to form holes like that in the future
You saved me a lot of time, well there's goes that idea.
The friction issue could potentially be fixed by some kind of lubrication system. It's not an ideal solution, but it would work in industrial applications where industrial machinery is already lubricated on a scheduled basis. Alternatively, different materials could be used, such as a jacket with a higher material lubricity, or a cable with a higher material lubricity.
The actuation issue that occurs when the tube and cable are coupled should be a non-issue, as the tubes should only be there as a guide, not coupled to the cable. If you're rigidly mounting the tubes to a frame, you should add a spring, or a bellows if you need the cable to still be enclosed, this would allow for stretch and compression in the tube, which would allow the cable to not actuate when the tube bends, due to decoupling of the tube and cable.
Your spring solution was close to working, but wasn't implemented properly, you want springs on the tubes themselves, so that the tubes essentially 'float' between the two ends of the robot's frame. What needs to be tensioned is the cable itself, not the tube. Think about water running through a pipe, the pipe is rigid while the water is not, the water is free-flowing within the pipe; now inverse this logic, where the interior material is the rigid component, the conduit around the rigid material needs to be flexible in comparison. This is exactly the situation here, the Bowden tube needs to be flexible in comparison to the rigid Bowden cable, and to achieve that you need the Bowden tube itself to be held in a sprung suspension when it's coupled to the frame of the robot; as it currently stands in the design, even with the spring tensioner, you are making both the tube and cable rigid components, when the tensioned cable should be the only rigid component.
The Bowden actuator will work, it just needs design refinements and proper implementation of the components. I hope you read this, are able to understand this (I'm not the best at explaining things), and implement the latter suggestion in a version 2 of this robot arm. I know the former suggestion, the one concerning lubricity, is much harder to achieve, if not outright impossible unless an off-the-shelf component already exists; but the latter suggestion, the one concerning how the Bowden tube is implemented, where it should be a sprung suspension rather than a rigid component, should be easy to implement, just braze/solder a spring to the metal crimp at the end of the tube, then set the other end of the spring in a hole. Implementation of this would require the arm to have a skeleton, as the Bowden tube is not a floating piece located only by a receiving hole, and could fall out in a test setup that does not have the arm's skeleton in place.
Please do a v2 with derailleur cables. They don’t change length. And you can get thicker Bmx brake cable that use this type of outside casing and has nice Teflon inside lining.
What about cutting the Bowden cables from center, then a little bit shorten them, then design a spring mechanism in between to connect 2 bowden cables? This way the spring movement will only occur on the Bowden cables and motor section does not need any moving parts.
You can put the black tube at the ends only. Like some bicycles
The bending/straightening contraction/expansion problem COULD be compensated for in software ... because the controller knows every joints position.
Right?
Nokon cable housing may work better, it is aluminium segments with ball joints, the difference in radius between the wire and the housing may be less when bending.
I'm curious if Dyneema shouldn't be used instead, extremely low friction very low elasticity yet stronger and more bendable then cables. You would have to splice it to attach it to anything, but other than that a wonderful material for tension based systems.
I get the feeling, with some redesign and compensating software, (a bit like backlash compensation) these cables could still be useful.
So... if the tubes are 180 degrees in front of each other... it would help a lot? Could the friction be due to both tubes pulling in the same direction?
You can account for bend in software so the steppers adjust the end of the cable and zero out bend’s effect. Calibrate the thresholds once the bend is consistent based on the rest of the arm.
It's the same pulley that pulls and pushes the two cables at the same time, so it can't be compensated with software. Unless you have two separated pulleys with one motor each, one pushing and the other pulling at the same time, than you can compensate via software.
gives lifelike movement but will fray and break after sustained use usually at the ends where i soldered on brass ends
Will the springs in the block affect the springiness of the bending moment arm when loads are applied?
I think: yes.
Great jobs and helpful vedio! And I wonder which CAD softwares you used. It looks like simple and functional.
Yes, it is! This is Fusion 360.
You should try to separate motor and gearbox with those bowdens. Try to put gearbox into the wrist, motor on the ground, and transfer rotation from the motor to gearbox with bowden.
Btw, KUKA does this, but uses belts to transfer rotation from motor to gearbox.
Excellent !!
Really interesting video
Couldn't a cable adjuster and a rotation sensor be integrated to negate the extra cable movement? So that when the rotation doesn't match the cable movement the adjuster would adjust the cable.