Please keep replies on-list. For pick-and-place, there are a few open-source projects working on solving the problem, such as OpenPNP:
http://openpnp.org/ Often, the "back-end" performing the motion does use gcode. On 2/26/2019 2:01 PM, Chris Albertson wrote: > I just read the TI's paper on this. They describe the workflow for using > the machine learning and vision subsystem. And the workflow is not > super-horrible as I feared. It fact it seems straight forward if you are > already familiar with machine learning and vision. For others, the > 15-second summary is this: > > You develop your machine learning or vision system on high-end PC hardware >> (at least a nVidida GTX10XX GPU) under Linux. You use familiar tools like >> TensorFlow and openCV. Then there is TI software that takes what you >> have running on the PC and translates it to run on the much smaller device >> on the Beagle Board. The beagle is about 100x slower but speed is really >> only needed for training a network, not needed to run it. > > > For those not in the field. "Machine Learning" is almost a misnomer. The > training and learning happens on the big PC then we "freeze" a snapshot and > move it the tiny chip where it never leans or changes behavior . The > learning happens only in the lab. > > > Nowhere is my question: The most simple use case of this that relates to > Machine Kit is a pick and place machine. This is a 2-axis machine that > picks up a tiny part from a surface and drops it some other place on the > surface. It does not even have a true z-axis. But the catch is finding > the part and finding the *exact* place to drop the part. For that we need > a camera. No-one uses g-code for these machines because we can't know in > advance how the machine is to move. So what we do is tell the machine > where the parts are in general and where relative to the final assembly the > part should go. > > It seems to me this machine would replace the g-code interpreter with > different logic but otherwise could work exactly the same. Or perhaps the > g-code is not read from file but there is a process that generates it in > real time based on the camera. > > What I'd like is to use my Harbor Freight Mili mill as a picking place > machine. This would be REALLY popular., simply place a little vacuum > picker on the spindle and for very little cost you have a slow slow pick > and place machine suitable for hobby level PCB assembly. The HF mill > should be accurate enough. > > In any case, how were you planning to this vision and machine learning > hardware to MK? > > > On Tue, Feb 26, 2019 at 8:15 AM Charles Steinkuehler < > char...@steinkuehler.net> wrote: > >> FYI: >> The BeagleBone-AI may be a good fit for your project: >> >> >> https://www.facebook.com/photo.php?fbid=10218976824519992&set=a.2907631578284&type=3&theater >> >> It should do machine vision _much_ better than the BBB. It's >> basically the SoC from the X15 in a BeagleBone form factor. I'm >> working on getting Machinekit working on this board and verifying >> capes work as expected. >> -- Charles Steinkuehler char...@steinkuehler.net -- website: http://www.machinekit.io blog: http://blog.machinekit.io github: https://github.com/machinekit --- You received this message because you are subscribed to the Google Groups "Machinekit" group. To unsubscribe from this group and stop receiving emails from it, send an email to machinekit+unsubscr...@googlegroups.com. Visit this group at https://groups.google.com/group/machinekit. For more options, visit https://groups.google.com/d/optout.
