Largest CNC Router Is Controlled By Hand

Fresh from this year’s SIGGRAPH is a very interesting take on the traditional X Y-table based CNC machine from [Alec], [Ilan] and [Frédo] at MIT. They created a computer-controlled CNC router that is theoretically unlimited in size. Instead of a gantry, this router uses a human to move the tool over the work piece and only makes fine corrections to the tool path with the help of a camera and stepper motor.

The entire device is built around a hand held router, with a base that contains a camera, electronics, stepper motors, and a very nice screen for displaying the current tool path. After a few strips of QR code-inspired tape, the camera looks down at the work piece and calculates the small changes the router has to make in order to make the correct shape. All the user needs to do is guide the router along the outline of the part to be cut with a margin of error of a half inch.

You can read the SIGGRAPH paper here (or get the PDF here and not melt [Alec]’s server), or check out the demo video after the break.

Anyone want to build their own?

[youtube=http://www.youtube.com/watch?v=-UmL7xZZSUk&w=470]

61 thoughts on “Largest CNC Router Is Controlled By Hand

    1. Why? Were you planning on plunge routing steel? With the kind of wood or plastic that router is designed to handle, I doubt you’d be able to maintain that kind of accuracy without, at the very least, keeping the finished product in a temperature/humidity controlled environment.

    2. I can totally imagine walking into Home Depot in 5 years and expecting most every power tool to be “smart” in some way. Check out the PDF where they discuss their results and look at what they were able to accomplish with a vinyl cutter.

      – Robot

    3. Let us know when you actually need a .003″ router.

      And also, Plasma torch anyone? God that guy’s vision of a home depot with smart tools like this is a freaking wonderful dream to live in. I hope it happens. This is a great idea I’d back the kickstarter in a heartbeat.

  1. Whoah… This is cool but I can’t help but think that this completely takes skill out of the equation. Like that Star Trek episode where they find a planet with zero children, one of the natives gives one of the ships children a carving tool where they only need to imagine the shape and the tool does the work for them.

    I like it and I want one, if only to make my projects go a little smoother and more consistent. I just can’t help but feel this is demeaning to the hundreds or thousands of man hours already invested to acquire the skills by those who do it by “hand”.

    1. It’s just a CNC router that you move manually.

      All the usual limitations apply, like the long set-up time and slow speed. If you want to cut maps of the United States out of fiberboard and you only need a couple, this is your tool, but for just about everything else a good set of hand tools and manual skill will get you good results faster.

      1. I have a feeling similar things were said when power tools first showed up on the scene. It definitely opens up things to the average person, but it won’t entirely kill of true artisans.

    1. Super Cool! I wonder if the “QR-Tape” placement important – or just that the camera can see it at all times.

      To Dave who snubs his nose at it’s ability to consistently hold .003″ tolerance:
      first: how do you know it doesn’t? (somewhat retorical)
      second: who cares… just guessing but it looks like it holds less than 0.1″ which is plenty accurate for many projects.

      1. Will, you didn’t read it either. .5″ is the most it is able to compensate for HUMAN positioning error.

        The average error in reproduction(using a pen to draw an image that they scanned for comparison) was .009″ with a max of .023″

  2. Iike the idea, as someone said, smart tools, imagine a jigsaw, router, where you pencil a line like normal, follow like normal, but the tool does fine adjustments to follow the line. I can cut/follow a pretty straight line, but imagine how much better it would be with auto adjustements

  3. This could be one of the most revolutionary things I have ever seen on hackaday. I’m unaware if this is just a modification of things that already have been done, or if this idea is actually new. Immediately I think of all the *onsite work* that could be done in construction for larger pieces that would have required weeks to send offsite and have done somewhere else.

    1. You are correct. This kind of feedback control of meta tasks I have been working on for ages.
      Pipped at the post by MIT students. sigh…

      Still you saw it on hack-a-day first. the future. soon it will ALL be like this….

    2. I have a line following flame cutter from the 1960s. It’s a very old idea. What these guys have done is something completely new and it has a lot of value. As somebody who built his own 4 ft. x 8 ft. CNC machine, I’m very impressed. Get it to track properly on chalk lines instead of that silly tape and I’ll be a lot more impressed. They won’t need my approval though because they’ll be very rich.

  4. I was hoping that this was more of a kind of DIY thing, and that the creator would have a writeup for the build, but the creator is trying to develop this as a product.

    I really wanted to know how much this cost him, and try to build one myself… looks like Im going to have to wait for someone else, or wait until this is actually a product.

      1. I applied to be informed about it on the website, and in the box for it, said that I would probably pay around $150 with the router included, maybe a bit more. I can only hope that it will be inexpensive. Otherwise, I might as well build a CNC machine myself.

      2. At $1k it’d be an amazing deal. For example, the Blacktoe 4’x8′ CNC machine for plywood is $3k and, if you could find a panel big enough, this could cut it up.

        I want this so badly.

  5. I think a lot of people are missing the point here: This doesn’t produce an image within 0.5″ of the drawing, the user merely has to be within 0.5″ of the correct location and the tool will make the exact cut.

  6. I think some people are misinterpreting the half-inch margin of error that was mentioned. That is the margin of error for the human moving the router, not the accuracy of the router.

  7. Hoooooo-ly carp. That is totally awesome and I want to make one.

    I have a Bosch that is almost identical to the router spindle they used. (it doesn’t have that whole spiral thing going on, it adjusts depth straight up and down.)

    I could probably design the physical bits from scratch with just the info in that paper…but I don’t think I’m anywhere NEAR smart enough to be able to develop the software/electronics…

  8. Now that is really, REALLY cool.

    It’s certainly useful exactly as described, as a manually-operated device. But I think this idea goes even further.

    Because the localization scheme and servos can correct positioning errors up to 0.5″, you could *fully* automate this router, and make accurate cuts in the same materials for which this router is intended, even with a terribly inaccurate gantry.

    This means instead of using precision hardened steel rods or beams to guide the axes, you could use 2×4’s. Or maybe cables and pulleys. Scrounged gearmotors instead of servos. Hardware store parts. With precision no longer a requirement, it opens up tons of possibilities for building a large gantry inexpensively.

    1. I was wondering about something like that as well but more from the lines of considering two different tasks:
      – moving the cutting head fast
      – moving the cutting head accurately

      When the table is small then it can be relatively cheap to do both but, on very large areas, unless you splash out lots of money you’re likely to compromise on one of these two. Using this approach I could see that it would be possible to use fast/cheap motors to do the bulk moving and then fine-tune the exact location using a cutter-head correction. In this case they’re using fleshy bodies for large movements.

  9. Makes me wonder if Dewalt, Makita, Milwaukee, Bosch and others are beating a path to the door behind which this better mousetrap has been invented?

    How about similar tech for a handheld power drill or screwdriver? Put the motor and gearbox on a 2 axis swivel behind the chuck so that a pair of actuators attached near the back of the motor can keep the bit perfectly perpendicular, horizontal or at one of a few selectable angles.

    It would make drilling for pocket screws or close to corners much easier.

    A pocket drilling mode could angle the bit down to start the hole easier then adjust it to the angle for finishing the hole.

    Think of it as applying the tech used by surgery robots for drilling femur bones for hip joint replacements. The robots are more accurate than a drill held by a doctor. A handheld drill with an “active angle” chuck could cost less than the robot and be just as accurate.

  10. This reminds me of that Dremel thing that was posted earlier, with the 6 degree sensor thingy. I still think the concept is totally awesome! The one thing I would change would be the range of error. Perhaps make the device turn off outside of the correct range, if it gets too close. (Turning off BEFORE it reaches its physical limit of movement would increase quality, so it slows down as it continues to stay put)

  11. Their solution appears to do the ‘heavy work’ through cameras, image recognition and a load of math to calculate location but could something be more easily achieved through a couple of optical mice?
    They have high-resolution and with similar marking strips can always track their motion. Is it just an issue with keeping the imaging sensor clear or are optical mice just not very accurate? (they are clearly sensitive)

    1. I’ve tried to track table motion with an optical mouse, but the problem is that it’s difficult to measure slow speeds with it.

      It’s a relative position sensor that you poll, and it gives you the number of intervals moved since the last time you polled it. That’s fine for a mouse that simply maps this data directly to the screen and moves the cursor there, but for a router you also need to know how fast you got there and how fast you are going right now so you can stop your motors from overshooting and oscillating around the set point, and generally be able to move in straight lines.

      Put simply, your polling rate will introduce aliasing in the resulting speed measurement you can derive of the data, which means you have to average things out over time, which means your control loop is always living in the past and won’t be able to react very fast.

      1. Okay, I think I understand.

        I can certainly see how an interrupt system would be better than a polling system (particularly one hidden behind an OS) but *if* it was driven from a uC at defined time intervals (very small intervals) then surely the same information can be gathered – from any sensor/encoder/etc you’d be gaining distance travelled and time taken to travel it which can be used to calculate instantaneous speed and acceleration from last sample.

        Would you ever be able to poll a mouse sufficiently quickly (lets forget about it going via a PC), i.e. faster than the motor could achieve, to tighten the control-loop?
        (google gave me this page which has a few interesting bits such as sample-rates, accuracy and DPI)

      2. The problem is more profound than that.

        Suppose the sensor resolution is 400 DPI, that is, it gives a “one” every 1/400 inch. Let’s say you’re trying to move it an inch a minute, which gives you one tick of the counter every 150 milliseconds, and you still aren’t sure about the exact speed because you need at least two ticks to confirm.

        So you get maybe three somewhat accurate measurements per second, but the actual position of the table may oscillate in between. For the PID tunings, this means really lazy values because you have to dampen any oscillations above 1.5 Hz.

        In practice, it behaves like you were pulling the table around with an elastic band. To get it to move, you have to pull it tight, and then immediately release as it starts to move, but if you can’t detect when it starts to move and how fast, you’ll just get a wild oscillation. If you do it too carefully, the table won’t move at all.

    2. To put what Dax said even more simply. That would be open loop control whereas this is closed loop control. It might be accurate enough depending on what you are doing, but they probably thought of that first and tried it before using registration marks.

  12. This is one of those flashes of brilliance you rarely see. Eliminate the entire gantry, its genius!

    One of the things I run into constantly as a woodworker is the difficulty of cutting smooth curves for templates. First you draw the curve, which can be quite challenging if it has to be an -accurate- curve. Then you cut the template on the band saw, which of course leaves a rough edge that has to be dressed with a file/sander/spokeshave whatever. Try to get two pieces that match to a 1/32nd of an inch tolerance doing that, its pretty hard.

    This problem of curves is one of the things CNC routers are awesome for. Draw your spline curves exactly the way you want them on Rhino or Autocad, send the file to the machine, it cuts your template in a couple of passes. You’re done. No band saw, no sandpaper, no screw-ups.

    This little machine can do the same thing without taking up 40 square feet of floor space and costing $10k. Absolutely made of awesome.

    I want one RIGHT NOW PLEASE.

    By the way, for the issue of needing to put location tape on the sheet. May I suggest printing something on there? Shouldn’t be too hard to arrange some kind of silk-screen for full sheets. Roofing already comes with a grid printed on it, manufacturers would no doubt jump at the chance to value-add their MDF sheets.

    I had a thought about position sensing as well. Since its a sheet with a known size, if you clipped a cable to the two corners of one end and used a winder to keep them tight, triangulation would locate the cutter fairly closely. Not as elegant as optical, but a more positive sensor much less prone to slippage than a mouse ball. It would also account for changing the angle of the platform.

    Then there’s the old drafting machine concept. Clip a jointed arm to one corner, the hinge angles locate the cutter. Limits the size of the machine, but the arms don’t have to be anything special. Sell the hinge sensors, let the end user make the arms and enter the lengths in a user interface.

    I really want one of these, did I mention?

  13. I want to elaborate on the QR tape placement and localization scheme, since judging by replies, many of you have missed this particularly cool aspect.

    As I understand from the paper, the QR tapes DO NOT have to be placed at exact positions. They can be placed quite haphazardly, as long as the router can initially see QR codes from two different tapes at any position.

    Before cutting, there is a registration step. The router is quickly passed over the entire work area, with the bit unpowered and retracted. While doing this, it builds a map of where each tape and code is located in relation to each other. Doesn’t matter if they’re not parallel, unevenly spaced, etc.

    Thereafter, it can determine its exact position by seeing any one QR code, even if the code is partially obstructed; due to the code’s inherent ECC and redundancy.

    That is incredibly clever.

    Regarding the suggestion that QR strips be preprinted on building materials. It would be convenient, although placing the tape yourself isn’t difficult since exact positioning isn’t required. It would also allow skipping the registration step on dimensionally stable materials, like plastic and sheet metal. However, wood would still require registration for best accuracy; since it expands and contracts in response to humidity.

  14. The video uses a music track from Blue Man Group.

    A use I can think of for this is carving designs in concrete floors to fill with faux terrazo marble chips in urethane or epoxy resin. Screed that into the carved design then grind and polish smooth.

Leave a Reply to John McCormackCancel reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.