New Member - Interested in retrofitting an industrial robot for hobby

[mcardoso]

Hi All,

I tried posting this once, it looks like it didn't go through. Sorry if this is double posted, this is my first post on this forum!

I graduated from college a few years ago having studied Mechanical and Electrical engineering with a focus in controls. I currently work for a large industrial automation company as a project engineer for motion control (servo) systems. I have too many hobbies already (I have a full machine shop in my basement), but I have wanted to purchase and rebuild an industrial 6 axis robot for some time now. My original intention was to look for a unit which came complete with a controller and teach pendant, however I caught on early that finding one of these used was either rare or prohibitively expensive.

Not to be discouraged, I decided that with my background in motion control, PLC programming, and access to industrial servo motors/drives, I would be ok buying just the robot itself and interfacing to the existing motors or replacing them entirely. In doing so I realized that I would be writing the control software and would need to teach myself the kinematics. I purchased “Introduction to Robotics: Mechanics and Controls” by John Craig and began studying it along with any college course material I could find that followed along with the book. My goal was to learn the material well enough that I could derive the forward and inverse kinematic equations and program these into a PLC to control the robot.

Unfortunately after a year or so of working at this, I got busy and lost the kick to keep working through the book. It felt like I was very much on my own and was struggling with a few of the concepts. I passed on a number of opportunities to purchase a robot like I was looking for in the <$1000 range, and kind of put the project away on the shelf.

For whatever reason I got the itch to pick it up again now and see if I can’t actually make this happen. I don’t know why I never got involved with the forums before but I’m hoping someone can answer a few broad questions below to get me pointed in the right direction:

1. Is anyone aware of people trying to do anything along the lines of what I want to do? If so, where can I find information or examples?
2. Which forums would be good to get involved with? I’ll be wanting to discuss the math of robot kinematics, interfacing to existing hardware, and just the project in general. I really get motivated by sharing the projects that I do, so I think getting involved with an online community would be the best way to keep me on track.
3. My understanding of kinematics for these robots is that there is not guaranteed to be a closed-form solution to the forward/inverse kinematics of a serial robot, but in the special case of a robot with a spherical wrist, there are solutions possible. Is this correct or is it a fool’s errand to try to derive these?
4. During all my research, I found very few examples of someone truly working out all the kinematics (the only one I remember was a graduate thesis on the kinematics of a Universal Robots UR5). Does anyone know of an example where someone worked out the kinematics for a 6 axis serial robot with a spherical wrist?
5. Does anyone know a good place to look for buying a robot? There is an industrial surplus company not far from me at all that often has used robots come through. They are often in the $2000-7500 range, but sometimes, smaller units come in without controllers and sit until they sell for a few hundred. I’d prefer to find something in better shape than I see there, but I’m hoping to keep the actuator purchase around $1000.

I feel confident that if I were to derive the forward and inverse kinematic equations AND I were to find a robot in good enough shape to be run, I have the knowledge and resources to interface to/replace the motors and write motion control code in a PLC to actually use it.

Thanks for taking the time to read this. I am very interested in your thoughts!

[jwatte]

This guy is building his own 6DOF arm, mainly using 3D printing: https://www.youtube.com/watch?v=oFCUw1pXlnA
He's using steppers in this series, but he has a newer series where he uses the ODrive series of controller boards and motor/encoder units to get full servo control.
If you have a mill and lathe in the basement, you could do one better :-)

When it comes to the kinematics of rigid articulated assemblies, you may be better off looking at simulation software. There are two levels; one integrated into CAD/CAM packages (which you're probably familiar with) and one integrated into game engines. In these systems, you typically describe the joints/bodies/forces involved, and it solves the constraints based on forces in the simulated world. Many of these engines expose their kinematic software libraries, so you can plug in your joint offsets/orientations, contact manifolds/meshes, and body inertia tensors, and calculate the full output. "working out" a closed form formula for whatever you want, would "just" mean tracing through the code for what math actually happens in each step. It's not actually "hard," as it's just composing homogenous transforms and applying to the various quaternions and tensors in the right order, but it's massively dull work better left to the computer :-)

Deriving the inverse kinematics for a 6DOF robot with a 6DOF constraint always has a closed-form solution, but the solution may have imaginary roots (e g, the position is too far away to be reached, or otherwise outside the range manifold of the assembly.)
The way I've worked out IK in the past is to write the code to solve it, using linear algebra (matrices/quaternions) for each joint. Because each joint only has one degree of freedom, you end up with a 6 row equation system, but those equations are not linear (e g, they contain square roots or arc tangents or whatever.)

[LloydF]

Someone pointed out this Open_manipulator_6dof_application. https://github.com/zang09/open_manip...of_application
I have used this and it works great, I did have the
3-d parts machine out of aluminum, and that was way overkill.

[mcardoso]

Jwatte,

I had initially looked into building my own from scratch, however to have an actuator with a useful payload and decent positioning accuracy looked to necessitate harmonic drive or cycloidal gearboxes. At $200-400/ea minimum on the used market, the cost of the arm in raw materials alone became prohibitive. Simpler designs using belt and pulleys would definitely be doable, but wasn't what I was hoping to find.

From your statement about using simulation software, it seems like you might be suggesting generating a lookup table of values from computer output rather than calculating the values on the fly. This seems reasonable, however it would leave you to interpolate between values not in the table. I would have no idea where to start doing this?

Your second statement sounds more along the lines of what I am hoping to do. I understand that there are always binary solution roots to the equations which lead to something along the lines of 8-16 (I forget at the moment) joint solutions for a given desired position and orientation of the end effector. I was always thinking that I would solve all of these solutions, discard those which exceed the mechanical limits of the robot or are otherwise invalid, then select the one with the shortest distance to the robot's current position.

I never made it to deriving a full set of kinematics for a 6 Axis articulated robot, but I did complete the kinematics for a much simpler 3 axis articulated robot (think of a KUKA palletizer) and implemented them in my PLC. Fortunately the PLC I have to use is one of the fastest on the market and it can handle the level of computation required to do this in real time. I don't think you could have any hope of doing this with an Automation Direct PLC or even the previous generations of high end PLCs. I extrapolated the performance requirements of these limited equations to the full set of kinematics for the 6 axis robot and figured I could reasonably run the calculations and motion control on a 5ms update rate.

It seems as though you mentioned that you have in fact done the solutions to these equations in the past. Do you have any examples that I could look at? I don't want to jump in and spend a bunch of money on a robot if I can't get the final equations figured out.

LloydF,

I'm looking at this software and having a hard time understanding exactly what it is doing. It definitely seems to be doing the kinematics. Is it supposed to run on a micro controller to interface with hobby servos?

[LloydF]

HUM. what do you know about:
1). ROS ( Robot Operating System ) ?
You can run the simulator but for me $250 or so a servo is much better than a grand ea.
2). Are you trying to build a industrial strength arm?
3.) Just curious are you a Software or Hardware Engineer?
I have found most if not all the software for Industrial strength Robotic Arms is proprietary and the Manufactures do not share.
If you are not all of the above and then some, you may find it very difficult to do what you are thinking.
Oh wait you are most of them there things, still not a easy task. I will follow your progress.
Good Bless and Good Luck.

[mcardoso]

LloydF

1) I've heard about it, that's about it. If you care to share what it is then I'm interested!

2) Preferably not build (although I certainly could if I changed my mind). I would like to take an existing robot and either interface to or replace the servos and do my own controls. I'm kind of in the boat that if I can't go full out and get an industrial robot, then I will probably just simulate on the computer. I am mostly into this for the educational content, but if I do get it working, I have plenty of tasks in my home machine shop that I could set this up to do (even if it were just because I could).

3) I studied Mechanical and Electrical engineering in college. I currently work as a controls engineering doing a lot of electrical panel design and complex PLC programming (including motion control). I have a background in C++/LUA/VBA but find this the weakest of my skill-sets. My access to controls equipment (lots of servo stuff) and my familiarity with PLC programming (and its built in support for motion control) is the reason I have planned to use a PLC for control of the robot.

4) I am equally frustrated at the lack of open information in the industry, but I guess it makes sense from their perspective.

Thanks for the information. My stumbling block right now is the math and kinematic derivation. Maybe I'm deluded but I feel like once I have that nailed down, getting the code written is just time expenditure and waiting for a decent used robot to show up for purchase. Also from the educational standpoint, the math is the most interesting part!

[LloydF]

Take a look at this book and see what is a good start. The download link is at the bottom of the page.
https://community.robotsource.org/t/...ok-for-free/51

[jwatte]

From your statement about using simulation software, it seems like you might be suggesting generating a lookup table of values from computer output rather than calculating the values on the fly.

No, I mean run the calculations on the fly! You can use the simulation software to come up with a solution, and then back-calculate what the actual angles of actuation are by reading out the state of the simulation.

Also: Most people on this forum do not use PLCs, but instead use a variety of microcontrollers. Favorites include Arbotix, Teensy, Raspberry Pi, or NVIDIA Jetson, depending on needs. Choose one with computing power you need, and add the necessary bus interfaces/transcievers. (TTL UART, RS-422/485, CAN, Ethernet, ...) The Arbotix is an 8-bit 16 MHz device (Arduino compatible) and thus not great for floating point, the others on the list all have floating point processing units and generally high clock speeds (Teensy 4 is 600 Mhz, Raspberry Pi is 1.2 GHz quad-core, and Jetson AGX Xavier is 2.2 GHz octo-core with a neural network/vision acceleration GPU.) Obviously pricing varies, too -- the Teensy 4 is $20, the Raspberry Pi is $35, and the AGX Xavier is $700. Plus whatever transciever and sensor hardware you need.

[mcardoso]

Huh, I have to think about what you're describing. Sounds like using simulation software running on a computer and outputting the data in real-time to a microcontroller for interfacing to servos. Interesting idea. What simulation software would you use? Maybe this is something I could do in conjunction with what I've been thinking of.

I totally get that. If I want to use any of the servo equipment I have, I am stuck using the PLC since the interface between the two is effectively proprietary.

I found a Fanuc 200iC plus controller going on sale. I'm not willing to put a ton of money towards it but maybe just maybe I'll end up getting it.

[jwatte]

What simulation software would you use?

The ROS platform uses the ODE library I think (Open Dynamics Engine) which is open source.

the interface between the two is effectively proprietary

The bastards!