I like what you're doing. 20 years ago anyone, hobbyists and tiny garage-shop companies could create electronics devices just as sophisticated and cool as big corporations. Today, SMT has largely trashed the creativity (and competition) that existed then. Much of this assault on creativity and competition was purposeful. Totally bogus interfaces like USB were developed to replace totally open standards like the standard 8/9/10-bit UART protocols (plus device command setsm, which were usually published by the manufacturer). Today the scam-interfaces like USB require proprietary drivers for just about every freaking device, and definitely every kind of device, and also require a device driver for every OS.
While most modern technologies are "purposeful evil" developed by [consortiums of] large companies, SMT might not be, for it does address the very real "signal density problem". However, without a doubt, the result is similar.
I remember a HoustonInstruments pen plotter I bought about 20 years ago for about $150. It could move anywhere over its 12" x 18" surface and raise and lower the pen. It was not super fast, but it was quite reasonable. I'm guessing it could go fetch components and place them at about 1000 per hour - if it had a vacuum pick-up tip to replace its ink pens. Of course, it lacked a component centering system, which would add expensive and/or slow it down further.
Still, $150 is 30 times less than $4500... and $4500 would indeed be cheap for an automatic pick-and-place machine.
I created dozens of hardware devices 15 to 30 years ago, with multilayer PCBs up to 16" x 24" in size, with up to 400 ICs (plus many more 2-lead components like caps, diodes, resistors, etc). But for the past 15 years I've been developing software. In my absense, SMT has pretty much taken over the market for "cool/modern components".
I just designed a device/product with 2 small PCBs (2.80" square and 5.80" square) with BGAs, QFNs, QFPs, and discretes from 0201 to 0805. I now have 50 + 25 of these two PCBs, plus all the components. So I've been researching what equipment and supplies I need to build these prototypes. Quite a learning experience: stainless-steel stencils (new), solder-paste stencil printers (new), automatic pick-and-place machines (new), reflow ovens (new), hot-air-soldering stations/pencils/guns (new), and so forth.
Without a doubt, precise placement of BGAs, QFNs, CSPs (and any other package with hidden pads) is the most crucial problem to solve to effectively prototype new gizmos. While it *may* be possible to manually assemble PCBs with 0.40mm to 1.00mm ball and pad pitch components, my opinion is, the world *seriously* needs a CHEAP, SLOW, PRECISE automatic pick-and-place machine to unleash the creativity currently bottled up by the difficulties of SMT assembly.
So, I applaud the folks who are trying to make one!
I created several inexpensive products with robotics features in the past, so I understand the difficulties, at least in principle. I can't be sure what tradeoffs you've made, but the only question I have from the little I can infer from the above messages is the choice of stepper motors. I adopted steppers in my earliest robotics devices, but later switched to DC brush motors plus rotary and/or linear optical encoders. Given my experience with both, I later regretted choosing steppers for my early products. Even with microstepping, which I took extreme advantage of, steppers just can't compete overall. And when I learned how cheaply I could make my own optical encoder discs and scales, even "price" favored DC motors plus encoder feedback.
I don't know where in the world you're located, but I'd be happy to discuss your approaches and maybe I'd even offer some opinions and advice-from-experience that you'd find useful. For the next few months I'm in the western USSA, but yahoo-instant-messenger, skype, telephone works too.
Ironically, the device I'm currently making is a 4-camera robotics vision system. Each of the 4 cameras are 3" x 3" x 1" (thick) with 2592x1944 resolution (either bayer-RGBG-color or monochrome). All 4 cameras (each is one of those 2.80" square PCBs I mentioned) are attached by cables to a single controller (the 5.80" square PCB I mentioned), which controls the cameras, captures the images, performs lossless image/data compression, performs some kinds of image processing (on the fly), puts the image-data and some other data (mostly image-processing) into ethernet packets, and spews them across a single standard RJ45 gigabit ethernet connection to a PC.
Like you, I plan to sell direct to keep customer costs to a minimum, and encourage engineers and companies to make my devices part of other products they're making. The hardware interface is totally standard (gigabit ethernet), I will freely provide all command/communications protocols, data/packet formats, lossless compression/decompression source code, and just about everything with the possible exception of [some-of] the FPGA firmware, which nobody needs to adopt the device. I'll probably also supply the FPGA firmware under confidentiality agreement to anyone who can modify it, and has a real need to modify it for their specific purposes.
I'm shooting at a $700 price for a 4-camera system (each camera adds or subtracts about $100 from the price). However, here is one idea that you might want to consider to "protect your butt" AKA "leave your options open". Publish a [more-or-less bogus] "retail price" 30% to 100% above the price you're willing to sell units for. But then offer "introductory discount/pricing" that brings the price down to your original/nominal target selling price. This way, if you later decide you benefit by adding representatives, distributors, retailers and/or outlets for your product... you can end or reduce the "discount" so your middlemen won't complain and say "I can't compete against a manufacturer who sells units at my cost". That's what I'm thinking of doing. Like you, however, that's not my plan... that's my fallback.
I wouldn't worry too much about speed. The primary market for your machine will be one-man and garage-shop businesses who are just starting up, or who want to develop custom (very low volume) devices with BGAs and QFNs. These folks can "go do other things" while your pick-and-place machine operates, or maybe even "go to bed and let the machine run" if you've done a really great job (highly reliable, and/or can keep running and report all [potential] problems when the run is complete).
What you absolutely NEED to provide is placement precision. Without that, too many BGAs and QFNs will end up with shorts, opens and other defects after reflow. This consideration is one more reason to consider DC motors plus optical encoders (linear scales are much better for X,Y[,Z]than rotary for your device, because rotary are inherently "indirect" for X,Y measurement in any pick-and-place configuration). Of course rotary is better for part rotation (if you need it, which you may not given "vision").
Have you published any photos or youtube test videos?
Good luck. Let me know if you'd like to chat privately.
reply »