Robotic Linear Rail Build Photos

pl7i92 wrote:

the bars in the First Image are supost to go the other way as they handle then alot more stiffness
good system will it be on Chain ,Timingbelt,RackPinion ???

It is on rack and pinion. In the first post last second last picture you can see the pinion gear meshing with the rack.

RoboDK outputs the code in cartesian space so inverse kinematics are needed on the robot to actually interpret the code and generate the actual joint values and in this you need full inverse kinematics and a way to handle singularities and pose selection etc. Otherwise the output may not match the simulated code since the interpretation of the code and kinematics calculations are done again. All proper robot controllers work this way which is why I think there's no need for it in RoboDK but linuxcnc isn't really a robot controller and can't be directly used as such without caveats. For a commercial project I'd rather look for a proper robot controller that's actually designed for the job.

I couldn't find a way in the software to output the planned trajectories(such as arc moves) as series of single joint space targets instead of cartesian co-ordinates for use with only forward kinematics on linuxcnc, which is what I had hoped, and in essence using RoboDK as my kinematics planner and using linuxcnc as only a controller for syncing joint motion. I could use inverse kinematics for jogging the robot around but not to actually run programs on it. I was hoping to stream the target points directly to linuxcnc out of RoboDK.

I was going to ask them about this when I got around to bying the edu license but I hadn't gotten that far yet. Maybe sometime next year.

Roiki wrote:

RoboDK outputs the code in cartesian space so inverse kinematics are needed on the robot to actually interpret the code and generate the actual joint values and in this you need full inverse kinematics and a way to handle singularities and pose selection etc. Otherwise the output may not match the simulated code since the interpretation of the code and kinematics calculations are done again. All proper robot controllers work this way which is why I think there's no need for it in RoboDK but linuxcnc isn't really a robot controller and can't be directly used as such without caveats. For a commercial project I'd rather look for a proper robot controller that's actually designed for the job.

I have looked at other robot controllers, www.daincube.com/ and www.mintrobot.com/page/mep-pi.php . I have brought the Mep-PI from Mint Robots, but is is not a complete robot controller. Rather it is a hardware platform for building your own. Mint supply the code library's but you need to code your own controller. I brought the Mep-PI because it is open source and I can change things as I need. But I have decided I will focus on LinuxCNC.

I have spoken to Justin the founder of Mint Robots at length about the design of the his controller, he started off using the Daincube controller but decided to build his own because of its proprietary design.

Over the next couple of years I plan to build a LinuxCNC config for industrial robots. That will include dynamic joint compensation, inverse kinematics and singularity detection. I will keep the config open source.

There really is no alternative. Main stream robot manufactures will not sell you a controller cheaply. My last quote from Yaskawa Motoman was $40,000 for a new old stock NX100.

you certainly have your work cut out for you then, I hope you succeed. I have a friend who used to work for ABB on robotic systems and they spend millions to deal with the robot kinematic systems since it's integral in the reliable operation of the robot. It's not easy by any means.

I agree there really is very few alternatives. The controllers that are out there are usually in 5 figure range, the cheapest I've found was a Chinese one on aliexpress around a grand, and that doesn't really seem like a nice idea without access to the software or documentation. Only a few open source ones exist and they don't really fit the bill.

I was hoping I could build a system around linuxcnc/machinekit and RoboDK but that remains to be seen, I'd need to talk to them about how they handle the trajectory generation and if transforming a move to a point cloud is even possible. Or write a post processor that extracts the points from the output code.

I tried looking at the Mep-PI but couldn't find it on sale anywhere, how much was it?

Yeah I know it wont be easy but I am sure after talking to the founder of Mint Robots that it is entirely possible to achieve. Justin is using the Mep-Pi for his own commercial robots he sells.

I paid $360 USD for the Mep-Pi.

I had a conversation earlier in the year with Albert about using LinuxCNC with RoboDK and he assured me what you can program RoboDK in both Cartesian and joint space.

The Mep-PI does seem a little interesting. Did you buy it from them directly? And do you happen to have any documentation on it? their site is pretty bare on that.

Also the issue I had with RobotDK wasn't programming in joint space, it was getting the solved joint space information out of the software which is what I think it can't do. or at least there's no documentation on how to do it.

Do know
www.anninrobotics.com
Annin Robotics - open source 6 axis robots you can build yourself

Mike_Eitel wrote:

Do know
www.anninrobotics.com
Annin Robotics - open source 6 axis robots you can build yourself

I have brought a copy of his small robot arm and I was going to 3d print one when I got a chance but have not gotten around to it yet.

Roiki wrote:

The Mep-PI does seem a little interesting. Did you buy it from them directly? And do you happen to have any documentation on it? their site is pretty bare on that.
.

Yeah I brought it directly from them over email.

Please see attached user manual.

I could send you the software but it is 1gb so I would need to find a way to do that.

Thanks, I'll take a look. If you have google drive or dropbox, you can get a download link from there and just paste it here.