5+ Axis Robot Arm

Building an open source robot arm for makers and small businesses

Similar projects worth following
THIS PROJECT IS DONE. STOP LIKING IT. UGH.Go here instead : September of 2015, a five axis industrial robot arm able to lift a kilo started at $10,000 USD.In October of 2015 Marginally Clever Robots, Ltd introduced the Evil Minion and brought the starting price down to $2000. Since then we've been working to improve our offer with easier manufacture, more reliability, and greater ease of assembly.I see an opportunity for low cost arms to serve small industry, hobbyists, and schools. I want to drive the cost down even more by making an arm that others can tinker with, improve on, and build community around. I'd like to see two arms assemble a third.Tell SpaceX and RocketLabUSA I need their rockets - lets put Marginally Clever Robots to work building a lunar colony.

Like what you see? Support me on Patreon.

View all 14 components

  • Sorry, Mario. Your princess is in another castle.

    Dan Royer04/04/2019 at 17:53 0 comments

    This project is now retired as I have moved (Since November 2018) to a 6 axis model.  Please follow this link to find out more:

  • Updated Render of Sixi 2

    Dan Royer03/09/2019 at 00:24 0 comments

    Thirsty for robots?  Say no more!  This is a render in Fusion360 of our up-to-date model of the Sixi 2 robot arm.  All the Fusion360 files are available through my Patreon so you can get started making it today.

    Also the Robot Overlord app has been updated to do IK and FK calculations on the arm, meaning it will be easy to train it for various applications.  My first goal is to make it pour a beer.  My second is to work on self assembly.  If I can get these arms to assemble their sisters... it's the last time I'll build a robot by hand.

    You can follow my weekly progress reports on my blog or through my youtube channel.

  • Machine Learning for path finding, part 1

    Dan Royer07/20/2018 at 01:51 0 comments

    I'm trying to use Machine Learning for robot path finding.  Read all about it at the link above.

  • Study number 5

    Dan Royer10/26/2017 at 06:59 5 comments

    I was up until 4:30am last night working on the fifth attempt to design the arm.  Previous post on were of (at best) study 4.  I reached a point in the design where I'd found enough mistakes that it was easier to restart and use the existing design files as a reference to do a better job.  I believe that - after three days of work - I've got something that solves all my previous issues and has fewer prints. 

    With lessons from the previous attempts, this time I built things in large passes:
    place all the things that can't be negotiated (bearings, motors, screws in motor faces)

    1. build the bones to hold the parts together as rough blocks
    2. plan where to cut the bones for printing
    3. add the screw holes, screws, and nuts to each block
    4. cut the bones into pieces
    5. calculate bill of materials
    6. order missing mechanical parts
    7. adjust each piece for printer tolerance and print.  <-- I am here
    8. the rest of the Gantt chart

    Speaking of BOM, here are some stats:

    • 218 fasteners
    • 34 mechanical parts off the shelf, including the drivers and the controller
    • 35 printed parts using ~1.8kg of filament at 20% infill will take 65h of continuous printing.

    Cost for raw goods and DIY printed parts is matching my expectations.  The long pole is the 8 larger bearings.  I got them from and they're about $250+shipping.  Ugh!

    So for day to day updates, follow me on Instagram.

    Become one of my Patreon supporters today and you get early access to the Fusion360 files.  Build this yourself!  Contribute ideas!  Automate all the things.

  • shoulder printed

    Dan Royer10/22/2017 at 06:17 0 comments

  • Bicep printed

    Dan Royer10/16/2017 at 21:44 0 comments

    Bicep parts are all printed.  I have to collect a number of missing fasteners and wait for bearings to be delivered.  In the meantime I'm going to assemble the circuitry for the controller and run some tests on a rotation sensor idea that's staying under wraps for now.

  • 6 axis model progress

    Dan Royer10/14/2017 at 22:02 0 comments

    Here are some pics from the 6 axis model I'm currently working on.

    As I print and assemble each part I check it against the model.  More efficient than building an entire model in a vacuum and finding I made mistakes.  Maybe it couldn't be done without an existing, purely-vacuum model as a reference.  Hard to say from the other side!

    The design files for this version are on Patreon.

    You can wait until it's for sale or chip in a little now, get the files early, and support development.  Your call.

  • Hiatus

    Dan Royer03/28/2017 at 17:59 0 comments

    I'm always thinking about building the arm and life laughs at my plans.

    I'm currently building several robots as commissions; running Marginally; and spending a lot of coding time on Robot Overlord. I've got newly designed AS5045 sensors that work well, they now need to be integrated into an updated arm design. I may actually repurpose those sensors in another project first: A new Stewart Platform being built for a local science museum.

  • Rotation sensor AS5045 upgrades

    Dan Royer12/28/2016 at 17:10 0 comments

    The biggest problems with the last arm design all stemmed from poor sensor construction.

    - the wires were exposed to users where they could catch, fall off, etc.

    - the sensor and magnet have to be held in perfect alignment or the arm loses sensation and goes nuts.

    To fix that I'm working on a single sensor unit that I can test independently and use as needed in my next arm.

  • When will there be more news?

    Dan Royer09/22/2016 at 22:51 0 comments

    I'm deep into preparing for a new product launch, and then finishing the Jigsolve. My winter project is to make a new arm and integrate it with Robot Overlord, which still needs a way to record/play back every robot it supports. I'm also putting together a team to build a new game based on my design doc, which should eat ~6 months and cost ~100k.

    All of which is to say I'm really busy and I need all the help I can get. I'm looking for talent everywhere.

    There is a growing number of open source robot arms, which is super. I only build one because I can't get one at an affordable price. So if you're making an arm, please let me add Robot Overlord support and together I hope we can achieve greater things.

View all 76 project logs

  • 1
    Step 1


View all instructions

Enjoy this project?



129Richard wrote 12/24/2018 at 15:10 point

Hi Simon what is the max payload of the robot?

  Are you sure? yes | no

Simon Merrett wrote 02/03/2018 at 07:19 point

There's a more comprehensive video update on the effectiveness of prtototyped strain wave and hypocycloidal gears at

  Are you sure? yes | no

Jochen Alt wrote 02/03/2018 at 20:42 point

Hi Simon, impressive! I was wondering why you took the timing belt on your last attempt and did not print the teeth into the outer ring. I could imagine that The timing belt is useful for dampening the whole thing, but it might also increase the backlash.

  Are you sure? yes | no

Simon Merrett wrote 02/03/2018 at 22:02 point

I think backlash is definitely a concern but I think the wear will be less if we start with belt against printed cycloidal disc. If we use printed ring pins against printed discs, it will wear out faster.

  Are you sure? yes | no

Jochen Alt wrote 02/03/2018 at 22:19 point

True... Still I'm wondering why most of the hypocyloid drives work with pins instead of two simple gears, the fixed one with internal toothing and a regular one with the excentre in the middle. In total, this requires 4 bearings only and no pins (although I must admit that I did not yet build it yet)

  Are you sure? yes | no

Simon Merrett wrote 02/04/2018 at 08:11 point

Traditional gear teeth profiles aren't suitable for the hypocycloid motion - it's all curves. So making inner and outer gear teeth with those profiles is unlikely to suit traditional gear cutting tooling (I'm totally guessing here). Drilling or otherwise machining holes or slots and slotting a pin in to take care of the radius (at good precision and hardness and relatively low cost) is probably how pins ended up in these gears commercially. Total guesswork though! 

  Are you sure? yes | no

Jochen Alt wrote 02/04/2018 at 12:17 point

I tried to design it in inventor (with involute curves instead of hypocyloid curves). What seems to work in our situation (diameter of 5cm, teeth height of at least 2mm (module 0,8) is a ratio of 120 teeth to 112 teeth which gives you 1:15 gear ratio after all. If I try to increase the gear ratio (my goal would be 1:20, since that is what I required in my bot Walter), either Inventor claims that the involute curves do cut themselves or the teeth need an unprintable small height. If you accept a diameter of 80cm, 1:20 is achievable.

BTW: do you consider to integrate an encoder?

  Are you sure? yes | no

Simon Merrett wrote 02/04/2018 at 23:51 point

50mm is pretty small, both for manufacturing with standard FDM and M3 hardware, but also for the pressure on your teeth and the strength limitations of using plastic gears. 80mm is more like it but I worry that even that will be unable to take e.g. 20 Nm (unless perhaps you use stronger filaments that are harder to print).

I did think about the encoder - see the link I posted above about the "hoverboard" motor and the BLDC I linked to below. Both would need encoders for phase control of the ESC. However, I think you're talking about an output shaft encoder for closed loop control. Thinking about this but I'm not actually making an arm, so it would really depend on the control system that each builder was using. There's definitely potential to build one in to the mechanism, away from the motor perhaps for a magnetic sensor.

  Are you sure? yes | no

Jochen Alt wrote 02/05/2018 at 08:24 point

my shoulder arm had a max torque of 30Nm, but I used T2.5 timing belts, meaning in average I had 40 teeth to take the force, equally distributed by the belt. But backlash is poor (arround 3°). I could compensate that with a closed loop via encoder and a fancy non-linear controller, but I never felt comfortable with electronic compensation of a mechanical issue. In my current sketch with 140 / 148 teeth typically 15 teeth would take the force, but very much relying on the print-precision, and not equally distributed. Finally, I think that your idea of a having a timing belt as internal gear would improve force-distribution across involed teeth a lot. I print with high-temperature ABS, never had an issue with broken teeth in T2.5

  Are you sure? yes | no

Simon Merrett wrote 02/05/2018 at 20:29 point

It would be interesting to see whether a change in belt profile from T2.5 to HTD or GT belts would make a difference on its own.

I agree it is far more satisfying to start with if you can explore the solution to mechanical issues with mechanical interventions but at some point the cost/benefit of the precision required for a given performance can make compensating electronics very attractive as a design investment!

It was interesting to hear about your belt gear setup but my point about the strength of printed parts was more to do with how you interface the torque output to the lever and load, not the strength of the teeth. On Walter and other arms, the diameter of the belt gears is quite large to achieve your desired ratio (so force of torque is reduced at increased diameter) but if you look at Dan's prototype hypocycloidal gear ( if any of that was experiencing 20-30 Nm and was 3D printed I would worry about deformation of parts, especially holes where bolts go in. Increase the diameter and the forces naturally reduce, so we just need to be aware of those limitations for now. Perhaps if you hit upon a winning design you might go for resin casting or CNC machining from e.g. acetal/delrin or HDPE sheet.

  Are you sure? yes | no

Jochen Alt wrote 01/31/2018 at 10:07 point

Hi guys, if the 3d plastic is too weak, I can recommend to reenforce it with carbon rods that are glued into designed holes.  I got this trick from the Dexter arm/Haddington. This makes it much stiffer and eases PID tuning, since this dampens the osscillation as well. I had a hard time in tuning the PID to different load-situation: tuning it with a small load is easy, but higher loads need different P&D factors. When gravity force varies as well due to joint angle this gets even worse.

  Are you sure? yes | no

Simon Merrett wrote 02/01/2018 at 23:47 point

@Jochen Alt thank you for the tip - it's great to hear from someone who has been through the whole design - realisation process. I'm still stuck on printing gears but your idea of reinforcement is definitely applicable if we start hitting the mechanical limit of printed gears before the motor stalls. 

  Are you sure? yes | no

Simon Merrett wrote 01/25/2018 at 19:45 point

@Dan Royer have you considered these BLDC motors with planetary gearboxes and encoders?,searchweb201602_4_10065_10344_10068_10130_10324_10342_10547_10325_10343_10340_10548_10341_10192_10190_10084_10617_10616_10083_10618_10615_10307_10301_10303_5920011_10313_10059_10184_10534_100031_10103_441_10624_442_10623_10622_10621_10620_10142,searchweb201603_40,ppcSwitch_5&algo_expid=79bb3727-2d83-441c-a26c-a1f5957b6c93-3&algo_pvid=79bb3727-2d83-441c-a26c-a1f5957b6c93&priceBeautifyAB=5

The driver appears to be included, in that you control through PWM, direction logic line and start/enable logic line. Looking at the sizing, if you can live with lower RPM on your output stage, they would appear to be small/light for a given torque rating. For example, your 20Nm application might select the GSP42M530SH316. At 24V, rated for 18.8Nm at 3.8 RPM, with a max torque of 28Nm (presumably slower RPM/stall torque).

In comparison, I can't tell from your instagram post of the motors when they arrived but it looks like a geared NEMA 23, such as:

would be a torque equivalent(ish). It's 1.7kg vs 0.5kg for the BLDC. I think you're using this one at the shoulder(?) so the weight is less of an issue but that pattern probably scales down to the smaller units. Also don't know what RPM a 47.1 geared down NEMA 23 output shaft could achieve.

I realise that none of this addresses any potential backlash of the planetary gearboxes and you'd perhaps want to combine (ignore) the input encoder built into the BLDC with (in favour of) encoders on the joints for closed loop but they seem roughly the same cost ballpark and through reduced mass may help reduce the moments your lower(?) degree of freedom joints will have to overcome.

  Are you sure? yes | no

Dan Royer wrote 01/25/2018 at 20:32 point

I appreciate that you put in all that research!  At this time motor torque and arm moment are not yet an issue.  The bones of my last design are two weak, the motor keys tear through the plastic.  I have zero experience with tuning PIDs to get straight line motion with DC motors.

  Are you sure? yes | no

Simon Merrett wrote 01/25/2018 at 21:09 point

That's interesting - one advantage of a 3D printed hypocycloidal or strain wave gear is that it allows you to transfer torque without high forces near the axis. I'm sure a keyed shaft adapter wouldn't be too hard/expensive to make. Is that something near the top of your list?

And by "last design", are you talking about Study No.5?

I have no idea about PID tuning DC motor drivers either but I reckon it wouldn't be too hard (if you were to use what's already out there, building on the shoulders of giants etc). By "straight line" do you mean of the effector, i.e. the reverse kinematics part? Could it be set by encoder pulse inputs, rather than coordinating motor step output pulses?

  Are you sure? yes | no

Simon Merrett wrote 12/14/2017 at 08:06 point

Hi Dan, when you were coming up with your design, what did you think about strategies to move the motors further "back" to reduce the moments?  Did the motor masses become less relevant once you set yourself a 2kg @ 50cm payload constraint, or could the mass help counterbalance the arm and payload? 

iforce2d on YouTube has been attempting a design that you might find interesting. 

One of the other reasons it might be worth using belts or concentric tube drive shafts to move the motors "back" is that you could remove the motor axle from the pivot axis at key joints and avoid those BoM-busting bearing prices! 

For my part, I'm hoping to soon find time to video the latest (nearly a year old!) timing belt strain wave gear. It's OK but I'm not sure how much it will handle higher moments with rubber belts. I saw the recent AvE video on the hypocycloidal(?) Japanese gear and am inspired. Maybe a cnc plastic version with small deep groove bearings is feasible? 

  Are you sure? yes | no

Dan Royer wrote 12/14/2017 at 17:54 point

What I think is "how?"  Either I use gearing, which doesn't work over long bones; or timing belt, which is hard (for me) to tension and (I'm told) adds a harmonic vibration that reduces overall performance.

What are "concentric tube drive shafts"?  link, please.

No bearing is more than $30.  they're not really that bad.  total cost of motors is greater.

I'm trying to make something that nearly anyone can recreate.  The easier it is to produce, the better.  So I've accepted that 3D printing is OK, but most makers don't have CNCs to produce bearings or metal adapters.  I may have to bend and make one or two metal parts for sale.

  Are you sure? yes | no

Simon Merrett wrote 12/16/2017 at 21:22 point

I admit I haven't tried to tension timing belt and wasn't aware of the harmonic vibration. 

Not got links for the concentric shafts but given the hollow "bones" I thought it would be a good approach for rotational transmission. The idea is to nest one, say wood/carbon/fibreglass dowel or tube inside say a PVC or polycarbonate (slightly shorter) tube, separated by a radial/deep groove bearing. A second pair of larger bearings holds the outer tube in the arm. There's a motor coupler on one end of the inner shaft and a drive gear/pulley on the same end of the outer shaft. 

On the "far" end of the inner shaft you could transfer rotation to the next bone, perhaps with a universal coupler (  and a drive gear/pulley on the outer tube to drive the joint at that location.

I've wondered about two similar shafts in parallel instead. 

I see the bearings aren't individually expensive now. 

Without going too off-topic, I know you are considering encoders in future. What kinds are you thinking would be on the shortlist? 

  Are you sure? yes | no

Simon Merrett wrote 01/30/2018 at 02:56 point

This design has some interesting approaches to keeping the motor mass closer to the previous joint

I think there's even a concentric drive in the forearm - see two steppers at the elbow.

  Are you sure? yes | no

trevorjtclarke wrote 10/26/2017 at 17:16 point

Hey dan! Like the iterations in this! Curious about angular measurement, are you using any encoders? Or using steppers with encoders? And yes, printing iterations for tolerance is definitely annoying. I've just added it as a variable in my fusion360 designs :)

  Are you sure? yes | no

Dan Royer wrote 10/26/2017 at 17:17 point

in my current iterations the robot is totally blind.  once I've proven the basic mechanics I plan to add first some limit switches, then some encoders.  One problem at a time :)

I also use a printerMargin value.  I'm not in love with it.  I'd love a CAM option to add tweaks after design and before printing, so it can be better tailored per-machine.

  Are you sure? yes | no

trevorjtclarke wrote 10/26/2017 at 17:24 point

Right on :) Also, it looks like if you could cnc some of the bigger sections out of hdpe or something for quicker testing?

  Are you sure? yes | no

Florian Festi wrote 09/26/2017 at 13:25 point

Nice Project! But I wonder if you are giving a way much strength and stiffness by using open cross section profiles (H and C) for the arm. In my experience constructions made out of plywood and with finger joints are much stiffer if they are closed all around. Holes in the walls are often not a big problem but missing walls weaken the parts notably - especially wrt (lateral) torsion.
It is a bit difficult to judge just from the pictures but it also seems that the different axis of the arm have different strengths - limiting the capabilities to the weak ones. Especially the NEMA17 motor for the first axis looks much weaker than the linear actuators for axis two and three, given that is also has to deal with the full length of the arm as a leaver.

It's a bit of a pity that you do not have a video with the arm under load or fast movement, so the strength and stiffness is more visible.

  Are you sure? yes | no

DontStalkME wrote 04/05/2017 at 08:57 point

If you want it to be cheap use 3d printed pneumatics. Or just home built pneumatics from pvc pipe etc.

Counter-weights will get you more strength / $

  Are you sure? yes | no

George I Fomitchev wrote 08/15/2016 at 06:37 point

maybe smaller hand to mount to a DIY SelfieBot

  Are you sure? yes | no

Peter wrote 09/25/2015 at 19:12 point

Do you have any STL or STP files for this Project? 

  Are you sure? yes | no

Dan Royer wrote 09/26/2015 at 02:27 point

they're in a zip in the Evil Overlord project so the arm can the rendered in 3D

  Are you sure? yes | no

chetan wrote 09/02/2015 at 12:11 point

Nice project dan. I am regular visitor to your marginally clever website to learn CNC basics and when i was looking for DIY robotic arm then i found you again here.. :) 

  Are you sure? yes | no

Dan Royer wrote 09/02/2015 at 15:31 point

Do you think I should post this stuff to the MC blog?

  Are you sure? yes | no

chetan wrote 09/03/2015 at 11:41 point

Yes, It would be great if you post it to MC blog and link it to MC tutorials page. 

  Are you sure? yes | no

Andrew Becker wrote 06/17/2014 at 10:50 point
Hi Dan, sorry for the long delay, I took my family on holiday and decided to leave the hobbies behind. I have already had a good look at your project and it looks great, I would be more than happy to share my work with you, I'm also very curious as to what control system you're using.

  Are you sure? yes | no

Dan Royer wrote 06/17/2014 at 15:57 point
The software is home-grown. The electronics are a RUMBA controller - think an Arduino MEGA and a super sized RAMPS board put together.

  Are you sure? yes | no

Andrew Becker wrote 06/17/2014 at 21:02 point
So you are doing all of the forward and inverse kinematics from first principles? I've managed to logic out the geometric relationships for my arm because the axis pairs intersect which simplifies the maths. I've then moved all of that into excel to visualise it and plan on using that to implement it in Linux CNC. It works out fairly simple because the alpha, beta and gamma angles fix the position of the 3rd link in space, thereafter it's a matter of working back through each joint position. Because of the geometry I chose there is only one solution to the set of equations and it's not necessary to use matrices to solve. How are you handling it?

I have a Reprap Ormerod and initially I had some trouble with the software. It turned out that the PSU supplied was not supplying a stable voltage to the controller and when the heated bed switched on it would freeze. I went for an industrial PSU with a much higher rated amperage and the problem disappeared. It did however put me off using an arduino based controller considering how easily my first Linux CNC set-up went.

  Are you sure? yes | no

Kevin Harrington wrote 06/05/2014 at 04:51 point
I had a random thought:
If you take that whole arm section and put it in a 2 axis gambol, centered around the elbow joint (with the elbow being the outside joint), you could get a true R-R-R spherical wrist, without twisting the belts. The entire section you have now would rotate around the center axis. With a circular track around the edge the track race could then form the elbow joint on the out side. You could actuate it with a third counterbalance motor and a gear interface to the outer track.

  Are you sure? yes | no

Dan Royer wrote 06/05/2014 at 05:30 point
That's the first thought I had. I've skipped it for now to ... Work my way up? Also as the gimbal turns the wrist motors could collide with the shoulder assembly. I'm looking for a method closer to what existing arms do where the ulna rotation is between the elbow and wrist. I have some drawings and I'm saving them for later.

  Are you sure? yes | no

Dan Royer wrote 06/03/2014 at 14:35 point
One of the things that drives my business most is customer feedback. You asked for it and I can do it, so I'll start posting here more. Given how busy I am it will probably be a copy/paste from my website, which may hurt my SEO.

  Are you sure? yes | no

zakqwy wrote 06/03/2014 at 18:15 point
Thanks Dan. Great work so far, it's been fun to watch your progress as you develop the 4th and 5th axes.

  Are you sure? yes | no

zakqwy wrote 06/03/2014 at 12:11 point
What are your thoughts on putting more info on I'm less likely to click through for project logs or other updates, it's easier to read everything in one place (and comment on it here, too).

  Are you sure? yes | no

Similar Projects

Does this project spark your interest?

Become a member to follow this project and never miss any updates