"Best" robotic arm geometry?

[Infraviolet]

I'm thinking of building a 5Dof or 6Dof robotic arm as a project, I've done enough designing of other robotic parts I'm confident I can do it, and I've used cobots (UR5 type) before during some past jobs. The main thing I'm building the arm to do is test some ideas I have about 3d printed parts, I'm not so much in a situation where I'm looking yet to build it with a specific reach and carrying weight, nor yet have I so much decided on exact actual size (but it will be relatively small), so everything I'm asking here is to be proportionally scaled.

What I am trying to work out is what the general "best" kinematic arrangement looks like in terms of joints, Denavit-Hartenburg parameters and relative link lengths. By "best" I mean the one that can most practically reach the greatest variety of positions, the most spatially versatile.

I know that often the really huge industrial robot arms in things like car factories have pretty limited DoF, the design is all about putting heavy motors at the base, and in doing so they often use linkages rather than rotary joints at each joint. And when I used UR5 robots I found that while able to reach complex positions they VERY often went in to "gimbal lock" and other situations where the inverse kinematics couldn't work well, the actual workspaces where the tool end could reach a wide variety of angles was a lot less than the overall workspace that could be simply reached in 3 dimensions.

So I know that the choice of kineamtic layout and link lengths has a really big influence on an arm's versatility.

I can find resources online comparing SCARA/polar/6DoF to each other, but can't find much discussing how different 5Dof and 6Dof layouts compare to one-another.

Can anyone point me to such discussions and articles, so I can get a feeling for whether there is an actual best geometric layout for spatial versatility, or if not whether commercial designs are each optimised in layout for a different parameter than spatial versatility (for exmaple prioritising payload weight instead...).

Thanks

[Infraviolet]

Any thoughts on where I might find discussions of this. The trouble is most searches for kinematics bring up all the equations but not much discussion of the abilities different arrangements and relative link lengths provide, and searches about "best" bring up specific models of arm rated on things like price or programming interface, both utterly irrelavant to the question of the best geometric architecture to go for when designing your own.

Thanks

[Doctorandus_P]

I have not built a robotic arm myself yet, but I've desinged a few in FreeCAD.

The 6DOF (Excluding gripper!) articulated robot arm is the most universal commanly type, but there are some (commercial) robots that add an extra 3rd arm segment. This can be handy for example for installing a dashboard or a seat into a car, where access into the car is restricted because of the surrounding metal of the car frame.

The "palletizing" robots (Based on parallelograms) are simpler and therefore cheaper (quite often just 4DOF) and when you only want to move items in a horizontal plane it may be adequate. The "EEZYbotARM V3" is a nice DIY example of this. This plywood version from fablab ruc: https://fablab.ruc.dk/robot-arm-v-0-1/ can quite easily be made in bigger sizes (But has a very low gear ratio, see below). Another model of this "palletizing" type I quite like is the uStepper. A lot of the construction is made of standard tubing (Aluminium, or carbon if you like) while the rest is made with a 3D printer, for ease of replication. This one also has a quite low gear reduction ratio though.

SCARA has the advantage that most of the motors do not have to carry the weight of the payload. This type is relatively common for big magazines for CNC machines.

Then there is of course also the stewart platform. It seems to fit the build of (airplane) simulators well, but it's relatively stiff and I've also seen some examples of it being used for CNC milling.

Then there is the Delta robot arm, which is a "simplified" version of the stewart platform, with only 3 DOF. These can be very quick and are commercially used for pick and place of lightweight objects (such as picking up cookies and stuffing them in boxes).

And of course the XYZ cartesian version. These can easily made to custom size by bolting linear stages together. It is also much like most of the commercial CNC machines.

There are some (probably) good books about robotics, but I never looked much into them. I suspect they are easily to generic, or too detailed.

Of all the (quite many) DIY robot arms I've seen, they nearly always have a quite low gearbox ratio, combined with stepper motors. I guess that a gear ratio of around 1:30 to 1:40 will be optimal, and this is difficult in a high efficiency single stage gearbox (both cycloidal and strainwave can do it, but those are expensive). Nema23 motors run quite well up to about 600rpm, and it's nice if your motors can reach this when they move at their maximum speed. Note that if you double the gear ratio, the stepper motor has to run twice as fast, and it's output torque wil be significantly lower (maybe even near a factor or two), but this is compensated because the input shaft of the gearbox only needs half the torqe. The main gain is in higher resolution, and in higher payload capability for slow movements. All robot arms based on hobby servo motors are quite garbage. Too much backlash, low torque, and mostly not enough resolution to control a robot arm properly. I would never build such a robot myself. Using a single hobby servo in a gripper may be acceptable though.

 The combination of motor size and gearbox is quite important and has to made early in the design. I've seen too many DIY robots that can barely hold their own weight, or even don't work at all without a significant re-design. You can either do some calculations (which can be inaccurate due to loss of torque for stepper motors at higher rpm, (unknown) inertia and acceleration, and other factors. You can also start with some motor and gearing, and add a broomstick with an (approximated (final arm + payload)) to the end. This can give you an idea if your motor setup is "adequate" for your case quite quickly.

But what type of robot arm you want to design / built is mostly a personal question. For a lot of the DIY projects, the software seems to be a big stumbling issue. Chris Annin does put a lot of effort into software for his (open sourced) AR3. That may be worth checking into. There are quite a lot of youtube video's about building DIY robot arms and the first movements, but more elaborate examples of doing something useful with them, or even fluid motions are quite rare.

For myself I also do not know yet how I'm going to make it. Some of my designs are based on flat wood panels that can be lasercut, other models are designed to be made from steel tubing. Before you decide on a model to build, think of some long term things you could do with it. A lot of the DIY robot arms probably end up in some closet quite soon. A long term use of a robot arm could for example be something simple as a desk (reading) light. Or maybe a (fixed position) microphone or (macro?) photo camera that swings into position as you start some software on your PC.

The "palletizing" / parallelogram based version can be quite nice for this. I'm thinking of making it dual use as a tapping arm.

[Infraviolet]

Thanks for your detailed tips there. I definitely know what you're saying about hobby servos (or even the professional grade ones of similar sizes) being no good for jointed arms. And the "broomstick" trick is something I'm familiar with from other projects I've done with actuators in the past. Working out a use case definitely makes the arm more worthwhile, and gives one a much stronger direction to the design than trying to ponder which arrangement of joints is "best".

[Doctorandus_P]

It's hard to deduce your previous knowledge, and / or knowledge "gaps" from a few short posts, but the things you already knew, may be of use to those other 500+ people who read this thread.

And as I wrote before, I have not built a robot arm yet, but I have seen a lot of them. From youtube, Hackaday, thingiverse, instructables, kickstarters and other sources, and you can learn a lot both from the strong and the weak points of what others have done.

[Hawaka]

I have never built a robotic arm, but I've seen a few. One thing that I noticed in industrial robots is that a specific task is almost always done by the same type of robot. For speed and light payload delta design, for lower speed higher payload the more standard "arm" design, for extreme rigidity look at all the CNC out there. There is a reason that the industry went a direction and not another.

For your project I would recommend to first properly define what you need, what rigidity, precision, speed, payload capability, available space and so on. A 5Dof could be anything between a CNC center and a proper "robot arm" (technically both 5Dof robot).

A well define project will help you all the way during the design phase, otherwise you'll just keep adding or removing features that you don't know if you need and at the end you'll have a lower quality product that you will never re-use because it's either too complicated or too fragile.

[Bicurico]

Here is a suggestion: go to Autodesk, register and download the 30 day trial o PowerMILL (CAM software).

It comes with a module called "PRI - PowerMILL Robot Interface", which has a fairly good database of common Fanuc, Kuka, Yaskawa, etc. robots, including their CAD model (use for the built-in simulator/solver).

You can copy&paste the relevant CAD files to your own CAD software (consider IP and Copyright issues - don't distribute these files) and study why they build antropomorhic robots like that. It is basically the result of optimizing the shape. If a segment is too long, you need stronger motors, which then need more space and power, etc. So all robots represent a compromise between functionality, cost and what is possible to build.

Note that the files are stored DMT format (Delcam Machining Triangles - a different format for the same type of mesh). Just import them as a model into PowerMILL and export as STL.

Hope this helps.

Regards,
Vitor

[libralect]

Skyentific has a lot of info about self build robots, he even has a 3D printed one with some files.

It would help to know the application, because testing 3D prints doesn't need the giant work radius of most robots, so it's perhaps something basic that fits the project.

[liaifat85]

You can check this too: https://www.scribd.com/document/413998965/Types-of-Robotic-Arm-Geometry