Motoman K6SB Retrofit

Have you seen this tutorial:

linuxcnc.org/docs/html/motion/dh-parameters.html

You might only have to replace the lengths with those of your robot.

Huh I actually had not seen that one! Several years ago I tried to figured out the genserkins config for my puma arm before giving up and eventually finding an out of the box puma config that worked; with this page though I actually have a hope of doing it from scratch, or like you said, just replacing the link lengths.

Thanks a ton!

Anybody by chance know the harmonic drive ratios on these arms? Or know which document would contain them? I’m told they are params in the ERC controller, but since I lack said controller that doesn’t help me much

If I had to find the drive ratios, I think that I would use a spirit level on each joint to find the encoder counts for 90 degrees. That's all you need to know, really. And it will be some fairly-round number.

That’s a good idea, thank you.

I guess what set in my mind that DH params would be hard to deduce on my own is that my PUMA arm has mechanical mixing (wrong term?) in the wrist- moving either of the involved servos alone causes both a twist and a bend of varying rates; to move only one joint, both servos need to move synchronously at different rates.

Unrelated- I was planning to use a mesanet real-time ethernet board for this, since I have one on my mill conversion and it’s very good. I see the original poster is using ethercat- is that something I should look into / any advantages?

If I keep the stock AC servos in the arm, AND I can figure out the encoder startup commands with a logic analyzer and a working controller, AND I can hack said commands into linuxcnc, I will need to pwm “dumb” drives (3 half bridges per servo) and do the phasing in software (since the only feedback from the servos is the absolute encoders- no halls or resolvers near I can tell).

If I replace every servo in the thing with a BLDC, I can either do the same thing (half bridges and phase control in HAL/linuxcnc), or use off the shelf drives, feed them halls and +/-10vdc (lowpassed pwm), and just bring the encoders back to linuxcnc.

Any thoughts / hints on the three-halfbridges method, or other ideas?

I presume that the Motoman robot uses Yaskawa motors. If so it might be worth while to try to get some used Yaskawa drives and interface those with one of Dmitry's PCI "Mechatrolink" cards:
yurtaev.com/

That is not a bad idea- I’ll read about his stuff, thanks. I don’t know anything about yaskawa drives but they are def widely available on ebay and I can probably find their docs. I agree it’d be much less work to use them than replumb everything.

If you know the motor types used on the robot then you might want to contact Dmitry through his website. He really knows his stuff.

i'm not that good with yaskawas... if it happens that the original drives are using mechatrolink then i can probably control them. otherwise i don't know how to convince newer yaskawa drives to recognize older/non-standard motors...

I think that the STMBL might work with your motors. But the difficulty is getting a set of STMBL drives, especially with the current chip famine.

phasefreq wrote:

I guess what set in my mind that DH params would be hard to deduce on my own is that my PUMA arm has mechanical mixing (wrong term?) in the wrist- moving either of the involved servos alone causes both a twist and a bend of varying rates; to move only one joint, both servos need to move synchronously at different rates.

I _think_ that this can be solved in HAL, external to the kinematics. I have looked at it in the past and decided that it would work, but didn't have long enough with the robot to get there.