From Noisebridge
Revision as of 16:33, 11 December 2010 by Jerkey (Talk | contribs)

Jump to: navigation, search

Hereby known as "MC Hawking - The Wheelchair Robot"

Here's the beast:

Videos: rachel driving rachel attacks a wheelchair driving around noisebridge



Improperly used, the robot can destroy itself, its suroundings, and run you over. Please know what you are doing and don't be afraid to ask questions.

  • RTFI! Know how to engage / disengage wheel clutches, and where the power switch is incase it goes crazy.
  • Always warn humans nearby that the robot could lose control and run them over.
  • DO NOT "UPGRADE" UBUNTU OR THE KERNEL ON THE HOST MACHINE! (unless you are going to stick around and make everything work with it...)
  • if someone is sitting in the chair when it is on, warn them that the robot could go crazy and kill them, it has almost happened.
  • always make sure it's plugged into the battery charger, and that the charger is plugged in and powered on. The batteries are throwaways and barely hold a charge as is, so we need to keep it on the charger.
  • We have not implemented even basic safety shutoffs, so you should be watching it all the time. It will very happily drive through its operator and the wall without stopping.
  • If it appears to be getting out of hand, the power switch is just behind the joystick, where you turned it on. This will power down the motors and stop it where it is immediately, with the brakes on. At this point you can disconnect the parallel-port interface and drive it home with the joystick, or disengage the wheel clutches and push it home manually.
  • If it REALLY IS getting out of hand, use furniture to try to stop its movement so you can get at the power switch. DO NOT try to use your body to stop it, it's heavier and stronger than you are and it doesn't feel pain.


M.C. Hawking the Wheelchair Robot, AKA Noise-Bot, uses an electric wheelchair (Action Arrow Storm Series). This is a mid-sized electric wheelchair which can be dis-assembled without tools and transported in a car, which is how it got to noisebridge. It uses two 12-volt lead-acid batteries such as car batteries. The batteries are wired in series for 24 volts, and it is necessary that they are matched so that one doesn't wear-out the other during use and charging.

At this time (nov 2010) the batteries are throwaways which have just enough capacity to play around with the thing for a couple of hours before having to go back on the charger.

The battery charger plugs into the wheelchair on the joystick side, just in front of the wheel. The wheelchair is electronically immobilized by the third terminal of the chargers' plug, which is connected to ground. In the future, the robot will be able to plug into electricity by driving into its parking spot, so that the onboard computer does not have to be shut down or manually plugged into mains AC to avoid draining the batteries. At that point the computer will gain the ability to power-up the wheelchair electronics, so that the robot can be summoned from afar at anytime, without a human present to unplug the charger, turn on the switch, and do the reverse when the remote user is finished.

There is a computer (separate from the wheelchair controller which technically has a computer in it) which is in an extruded aluminum case like a car-audio amplifier. It uses a mini-motherboard and a M3-ATX power supply

The box is currently powered by a custom voltage regulator made by jake. It accepts up to 40 or so volts, but will never deliver more than 23 to the M3-ATX which is hard-limited to 25 volts. If the voltage is at or below 23 volts, its FETs are in full conduction and there is zero dropout. Once on, the M3-ATX will keep the computer running unless voltage goes below 7 volts, at which point the 24-volt battery array of the wheelchair is being seriously damaged.

The voltage regulator's input plugs into the wheelchair's charger port, on the joystick side, just in front of the wheel. The voltage regulator's output cable is soldered directly to the laptop's mainboard.

A circuitboard zip-tied to the arm-rest interfaces the parallel port of the computer to the joystick. This circuit uses transistors to trick the wheelchair controlbox into thinking that the joystick is somewhere other than at-rest in the center. Of the eight bits of the parallel port, two make the chair go forward, two to the right, two to the left, and two backwards. For each direction, there is one bit causing low speed and another causing medium speed. Both bits cause high speed. As of 11/15/2010 a steady-state on the parallel port is not ignored by the circuit, so a computer crash will cause the chair to go out of control. In the future, the circuit will ignore non-changing status bits, but the software has to be changed first.

At this time Jake is preparing to install sensors on the motor which will allow the computer to track movement and speed. An old ball-mouse has been disassembled and its sensors mounted on wires, and these sensors will be placed at the open end of the drive motors. An optical interruptor will be attached to the end of the motors' driveshafts, so the computer will experience movement on the x and y axes of the mouse when the left and right wheelmotors turn. Since the mouse is/was DB9 serial, it will be possible for the computer to treat it as a serial accessory rather than interpreting it as an actual Human Interface Device to move a mouse cursor. The next step will be a daemon to listen for serial data and to make available speed and movement data to applications such as the telepresence driving program ( to improve control of locomotion over a lagged internet connection.

An LCD monitor is being prepared to be mounted permanently as the seatback.

A robot arm made of mostly aluminum and animated by air-cylinders is fitted to the robot, but few electronic valves and no air compressor or tank have been acquired. Someone will have to invest time and energy to get the arm to do more than make the thing look more menacing. As of 11/15/2010 the arm was removed to make the robot more wieldy/less unwieldy.


The PC is running a stock Ubuntu with PyParallel being used by the controller program to access the parallel port. All the software from the home directory as of 11/15/2010 is here:

The above software is present in the home directory of user tvbrain and mchawking. The difference is that if you log in as tvbrain you are a regular user, but mchawking logs in to an instance of python which is how the wheelchair's motor functions are accessed.


username: mchawking
password: emc
username: tvbrain
password: [redacted]

Network Connection

The embedded PC is using a Generic Realtek USB 802.11b interface, which is sub-excellent, but the WUSB600N by Cisco doesn't work properly under Ubuntu for some strange reason.

In the future, one of those fancy Cisco 1200 wireless routers will be the way onto the net - some of them have four antenna connectors and do "A" which is apparently really good. They have an ethernet port to talk to the computer, and a serial port for configuration, and they run on 48 volts (POE is supported but has opposite polarity of 3com standard). Hopefully they run OK at lower voltages like 16v so they can be powered directly from the wheelchair's batteries. That way the onboard computer can suspend and do wake-on-lan if the wireless device starts sending it packets.

The embedded PC is configured from /etc/networks/interfaces Also, Network-Manager which is the gui-compatible 21st century newschool network configuration system is been pushed to the side, possibly for good. The embedded PC is now configured to find the wireless network called noisebridge and take the internal IP below.

Noisebridge internal IP address

The robot has claimed an IP on the internal network of

Static world IP address

For those not inside the noisebridge internal network, a static IP from has been assigned. The static IP is


Access from outside 2169 Mission

Configuration of the firewall is done (thanks Jof), so that connections to the machine can be initiated from the outside world. This is important if people are expected to log in and drive it around from Germany or Singapore.

DNS entry

In the future someone will assign that IP address a DNS entry like

Local Operation

The electric wheelchairs' controller brain will not allow movement unless the joystick is at rest for a second when power is switched on. If the computer has not initialized the parallel port to the "stop" bitpattern, and the switch is flipped on, the green LED will flash endlessly because it thinks you are leaning on the joystick and don't intend to do so.

The drive-wheels have clutches which can be disengaged from their motors (and brakes) by pressing a button and rotating a collar until it can be pulled outwards. When the clutches are disengaged, the robot can be pushed around like a shopping cart, and the motors can be activated without actually causing movement. Sometimes the robot must be pushed forward or backward in order to engage or disengage the wheel clutches, because they are like gears and only fit together in some positions.

Disengaging the clutches before testing movement software is highly recommended, as you will be able to see the motors trying to run you over without the robot being able to actually do it. This is like taking the laser off of Johnny Five before sending him to the lightning-prone recharging booth.


If you are badass, you can compile the code in simulation mode following the instructions here:

The resulting binary should run and give you a UI on your system. There's actually nothing machine-specific that you need at this point: your UI will look pretty much exactly what we have now. Select the "axis.ini" config file.

You should now just be able to:

  • open one of the sample gcode files
  • power the machine 'on' (unset emergency stop [red X], and set power [orange square] buttons)
  • 'home' the various axes (must home all axes, use radio buttons to select)
  • 'run' the gcode. (blue "play" triangle)

For most of the configurations there is a smaller window w/ an inverted cone representing the spindle (this is the default view). This will show the progress of your milling run. You can change the view and pan/zoom/tilt with the mouse.

For a sanity check, try simulating a run of the NB logo gcode from Identity

Then try gcode from the toolchain of your choice.


Many tools exist for generating/converting to gcode. Here's a good list:

In particular, here's how you generate gcode from

Bitmaps, Images

The EMC2/Axis application will import most image formats and create a raster scan tool path with milling depth proportional to image intensity. Haven't tried this but it looks very cool!

Text /TrueType

text / truetype :


DXF is a standard file format used by many CAD programs (as well as exportable from inkscape and Adobe Illustrator). It's a good choice, especially as gcode export from Inkscape seems to be buggy.

QCAD is an open-source 2d CAD program. Linux users can apt-get qcad, there is open-source executable for Windows as well (but it's hard to find as Ribbonsoft took it closed-source (wtf?) and most searches end up at their page.) QCAD is very solid and a great place to start with CAD, especially if you can't afford/don't need Autodesk/Solidworks.

We use a DXF-GCode conversion called, unsurprisingly, dxf2gcode [URL is not a typo]. In the /home/nb/dxf2gcode directory, run the command.

Open your dxf file from File->Read DXF. It seems to barf on DXF exported from Inkscape, but is fine with QCAD, so a workaround is to load the Inkscape dxf in QCAD and save it from QCAD (it's also easy to scale and rotate if necessary).

I am still working through the various dxf2gcode options but setting "infeed depth" to the same value as "mill depth" gives you a one-pass toolpath (otherwise it seems to do it in two passes.)

Coordinate units are kind of a crapshoot and don't make much sense right now.

Note that it may be doing a lot of math without benefit of the numpy library so it can be SLOW. It has no progress indication, and doesn't redraw the screen, so it may look like it crashed. Sometimes it actually does. But be patient!


Cadsoft EagleCAD is a schematic editor and PCB layout tool. It's not open source, but there's a free version that is very decent and limited only to the size of board you can lay out.

To generate GCode from your Eagle layout, get the scripts from These do "Isolation routing," that is, they will generate GCode to mill away copper outside the traces you laid out on a copper-covered PCB (as well as drill it).

I've added MAXCNC mill-specific commands to the pcbgcode config stuff, including setting the spindle speed to maximum and turning on the motor. (Config file is /home/nb/eagle-5.6.0/ulp/gcode-defaults.h on the mill PC).

First experiments milling PCBs = success! Jtfoote 23:57, 17 October 2009 (PDT)

I used .01 carbide end mills from here:

On 1-sided copper-plated (1-oz) circuit boards from here:

To start, create a layout in Eaglecad. To work best, use one layer (which can make routing a pain), and use a minimum trace size of 0.012 inches. Make sure you do a DRC check with 10 mil isolation spacing (thouugh I have found it misses trace-pad distances).

From the Eagle command line,

run pcb-gcode-setup

This creates several gcode .ngc file in the same directory as your eagle .brd file. They have suffixes like "top" "bot" and "d" for top copper, bottom copper, and drill. Run AXIS2 and load the appropriate file. Look CAREFULLY at the tool path. Does it really isolate everything you want it to? If not, you may have to increase spacing and re-run. Check especially places of small pitch and tight clearances, especially between traces and pads.

Milling the copper is a tradeoff between quality and speed. At fast speeds, the copper tends to get rough at the milled edges. Slow speed rates help a lot but can lead to impractically long milling times.

2.5 ips gives beautiful results, but a good compromise is something like 5 - 8 ips (set in the pcb-gecode setup). If you look at File->properties in the EMC2-Axis tool it will give you a rough idea of the milling time.

Setting the Z home at precisely zero (at the board surface) is very important, because we are working at depths of hundredths of an inch. After several attempts I don't think you can do this well with just the manual jog, and risk breaking the tool. Here's what I did that worked:

  1. Raise the head so you can put the tool in the collet. Put it in as far as you can.
  2. Tighten gently, only enough to keep from falling out.
  3. Drop the head to about 1/4 an inch above the workpiece. Don't touch!
  4. Loosen the mill tool in the collet
  5. GENTLY let the mill tool drop to touch the workpiece surface
  6. finger-tighten the collet -- tightly!
  7. Home the Z axis VERY IMPORTANT DON'T FORGET!!
  8. Raise the head and wrench-tighten the collet

I think the default Z Down dimension of -0.01 may be too much given that 1-oz copper is supposed to be only 1.4 mils thick (0.0014 inches). I've had good results with -0.008; .0.005 led to problems because I think the PCB I used was not perfectly flat at at some places this depth did not hit "bottom." With further experiments this may be because the clamps are compressing the PCB: at least it seemed to work fine more than 1/2 inch away from them.

Drilling vias

For through-hole components, you'll need to drill vias. Doing this is a multi-step, involved process. Because etching is sensitive to the trueness/truing of your board's surface, you only want to etch the board when it's mated directly to a metal block, but you need a plastic sacrificial stop for drilling. So, etch as normal, then add two registration marks to the top of the board. Unmount it, add the plastic backing, change tools, remount, and rezero, respecting the registration. From there, you can run the drill program after suitable modification (you need to manually comment out all M06/tool change commands).

Generating the drill file:

Settings specific to drilling:

  • Machine:
    • Tool Change: you can't actually change tools at the moment, so this is just to expedite your cutting process, and stay within the limits of the mill
      • X: 0.0
      • Y: 0.0
      • Z: 0.2
    • Drill depth: Measure your board with calipers, then add 10mil for good measure (I used -0.100" for the single-sided board I did yesterday)
    • GCode Options:
      • Ensure "Do tool change with zero step" is UNCHECKED. If this is checked, EMC hangs. No, really, you don't want to deal with this problem. It's awful close to "EMC hangs, you have to hard kill it, and then rezero the whole setup."

Once you've got the GCode files for etching and drilling generated, you need to tweak the drill file. As it's currently configured, our milling rig doesn't handle tool changes well (the CNC doesn't support automatic tool change, but the EMC software thinks it does, so the whole setup gets Very Unhappy when it runs into these commands). To remove these, open your drill ngc/tap file in a Text Editor Of Your Choice, and comment out all lines containing the tool change command M06. In GCode, comments are parenthesized lines, and they cannot be nested.

Cheat sheet:

sed: 's/^\(M06.*\)$/( \1 )/;'
vim: :g/^M06/s/^\(.*\)$/( \1 )/
emacs: M-% ^\(M06.*\)$ <enter> ( \1 )

Open your drill file in EMC and make sure it looks sane. There will be a bunch of tool change steps, where it runs itself back to (0,0,0.2), then goes back to work. This is the tool change position specified above; you may want adjust the point to be closer to the centroid of your drill pads, to shave off a minute or two.

After your etch file looks good, and your drill files look reasonably good, go ahead and set up for etching as above. When etching is completed, don't take the board out! We need to add registration marks for the remounting. Manually walk the head down to (0,0,0.5). Once you're there, turn on the spindle, and gently lower the head til it just bites the copper. Personally, I went to (0,0,-0.008), which is the depth used for etching. Leave the spindle on, and raise the head back to (0,0,0.5). You can turn the spindle off if you'd like, then slew over to (1,1,0.5). Once there, turn the spindle back on, and go down to (1,1,-0.008) to make the second registration mark. Raise the head out of the board, then turn off the spindle. I'd also recommend slewing back to (0,0) at this point, to make re-registration easier.

Next, load up the drill file in EMC. You should still have the outline of the paths the head just etched on the display, with the drill file overlaid. This is a good chance to sanity check the line-up of things, etc. Once that looks right, move to adding a plastic backstop and re-rigging the board.

With those two registration marks in place, you can re-rig the board with a sacrificial plastic backstop underneath of it. There's a few of these in the document trays next to the mill. Change out to your PCB drill-end, re-zero as for etching, and ensure you have good registration (make sure the drill-end is sitting on the 0,0 mark when you insert it, raise to (0,0,0.1), and slew to (1,1,0.01) to eyeball the match up. If you want to be totally certain, you can turn the spindle on and just bite the board, but, in doing so, you're likely to obliterate your registration mark, so I'd advise against that.

Make sure you're happy with everything, and, assuming you are, hit Run. Keep an eye on things and be ready to hit the kill switch (if babysitting feels tedious, just imagine yourself doing all these holes on the drill press, and then try not to hug your newfound robot slave/friend). -- User:Jbm 2010-05-07 13:20; a big thank-you to User:jtfoote for all his help in familiarizing me with the setup.


Latest: the gcode export from Inkscape is buggy. It's python so could possibly fix, but it looks kind of involved. Workaround: File->Save As-> .dxf, then use the QCAD->DXF toolchain.

For Inkscape 0.46, use this: (Do "get source", put all files in /usr/share/inkscape/extensions/)

IMPORTANT: Read the "Help" tab when you Effect->Export->Export as Gcode This tells you how to set the origin, and how to name your layers for different tools.

In the exported GCode, you need to add/change the following lines:

Set the cut height to -0.01 (for example, YMMV)

#2=-0.0100 (Height to cut at)

Add commands to turn on the motor after the first line:

G00 Z#1 
S 1000 (set motor speed to 1000 RPM)
M3    (turn spindle motor on)

And you are good to go. ---

This is cool but only works for Inkscape 0.43:


Cutting fiberglass (PCB) will eventually destroy any kind of steel cutting tool. The glass in the resin matrix is much harder than steel. You need carbide tooling. Note that carbide is too brittle for most metals and will just snap.


*  carries Niagara carbide end mills in fine pitches suitable for circuit boards. Available via Amazon as well. 
* also sells carbide tooling locally, which you can pick up at their store on 760 Market street #905
* High Speed Stainless (HSS) tooling available here:
* Spring loaded tool holders account for warped PCBs and mills:  2linc <-- expensive.  Cheap ones exist.
* Really small end mills bitsbits

Search for "small parts inc" on They have reasonable prices and reasonable shipping for endmills. (You can also search for endmills) I believe we have a 1/4" and 1/8" collet for the mill, so you're looking for tools with a 1/4" or 1/8" shank. I mostly bought HSS for cutting plastic, but I have some harder ones with exotic coatings for aluminum and brass. --lamont

Parallel port DB-25 pinout

OK, got a definitive pinout, seem to conform to those above. Jtfoote 23:11, 28 September 2009 (PDT)

Pin  1:  Tool speed PWM, active low
Pin  2,3: A quad input
Pin  4,5: Y quad input
Pin  6,7: X quad input
Pin  8,9: Z quad input
Pin 10: 7420 quad nand output (motion complete ack?)
Pin 11: NC
Pin 12: X,Y,Z limit (or'ed active high when at limit)
Pin 13: spare digital (unused header labeled "probe")
Pin 14: Enable, active high
Pin 15: NC
Pin 16: input to 4x PICs: enable, active low (reset?)
Pin 17: Enable, active high
Pin 18: GND
Pin 19-25 - NC

Misc other info


This guy rebuilt his, has derogatory things to say:

This other guy has some interesting upgrades. Especially notable is the new motor mount, and appropriate dxf

User:seph was hacking on one. Some notes from him are at There's a pinout, and an emc2 config. He also recommends thinking about eventually upgrading a lot of the components. Seph's work is what I (mikew) based the current functional HAL and INI files on.

Here's the link to how to drive the mill with a pc joystick:

Workshop notes, 1.15.2010

Machine parts and geometry

  • The bed is the blue anodized aluminum extrusion to which the workpiece is affixed.
  • The collet is mounted at the end of the motor. It holds the milling bit. Note: do not lose the collet, as it is basically impossible to obtain a replacement.
  • The milling bit, which superficially resembles a drill bit, is used to machine the workpiece. To setup the machine, bits are inserted into the collet and then tightened with two wrenches.
  • The workpiece is whatever you're milling - copper-clad board for PCBs, aluminum, acrylic, etc.
  • Workpiece holders are small threaded rods with metal "clips" that are designed to hold the workpiece firmly to the bed. They should not be over-tightened, as it can cause workpiece distortion.
  • Backing material is a piece of scrap material used between the bed and workpiece. It's important to use backing material to protect the bed.
  • Limit switches are located at the end of each axis. These act as safeties in case the milling machine reaches the end of its travel. Note: if you're milling and manage to trip a limit switch, the mill shuts off, in which case you'll need to redo your work. Be diligent.
  • The X axis runs side to side; e.g. the "width" of the bed.
  • The Y axis runs forward and back; e.g. the "depth" of the bed.
  • The Z axis runs up and down; e.g. the space between the bed and collet.

Affixing the workpiece to the mill

  • Use a piece of aluminum stock between the PCB and the bed.
  • Use adjustable wrenches to cinch down the workpiece holders to the PCB. You only want the bolts to be finger-tight; the fiberglass substrate can actually be compressed if too much pressure is used, causing distortions in the milled part. Do not use pliers.
  • The collet faces upwards (like the tip of a rocket) into the spindle.
  • Keep the plastic guard attached to the milling bit. Place the bit into the collet, and bolt the collet into the spindle.

Zeroing the machine

  • Note: In the MaxNC control program, it's important to zero the mill - to set it to 0,0,0 - before routing a board. Occasionally, the milling machine will not home to 0,0,0 in the MaxNC control program. To fix this, go to Machine > Zero Coordinate System > P1 G54.
  • In AXIS' Manual Control tab, adjust the X, Y, and Z axis feed controls until the tool is at the upper right corner of the circuit board.
  • Chuck the bit, finger-tight.
  • It is critical to zero the Z axis correctly. The idea is to drop the bit down onto the surface of the copper, with zero force, and then zero the axis. Use the manual controls to drop the bit, step by step, until it is about 1/4" above the surface of the PCB.
  • Loosen the collet so that the bit drops onto the board. Tighten the collet.
  • Immediately zero the Z axis.


  • Controls for starting / stepping / stopping milling operations are located in AXIS' top toolbar.
  • At this point, all of the axes should be zeroed.
  • Click the Go button in the toolbar to start milling.


I'd like to upgrade the maxnc. Right now its a bit sloppy. I want to (1) fix it up, (2) make a locating system for two-sided PCBs, (3) make a sweet document with tons of pictures so its easy. -moo

Hey moo, who are you? The mill seems plenty accurate for our purposes, are you finding it isn't? (though a pcb jig would be sweet). I think someone bent the X spindle through misuse, if we fix anything let's fix that... Jtfoote 05:37, 13 May 2010 (UTC)

Ideas for maxnc accuracy improvement

In order of utility and least cost

  • tear-down + lube, wax x-y interface
  • inspect replace x-axis gib
  • replace set-screws, with hex cap with nut to lock down
  • x-axis motor mount (stiffer)
  • do swivel ball bearing set screws. (make sense with a gib? ) (anyways grease + locktite)
  • add bearings at ends of lead-screws to stabilie (original MaxNC has these... )
  • z-motor mount (plans at
  • spindle replacement

dual layer PCB fixture & locating system

Part 1 work plate for prepared pcbs (i.e. has holes for locating)

  • make pin indexed work plate that attaches with t-nuts (no need to drill holes in the maxnc)
  • add large dowel pins for indexing PCBs with pre drilled holes.
  • release pockets to scoop under pcb

part 2 prep work-plate

  • something sacrificial with indexing corner, to drill holes in PCBs needed to locate in dowel pins above
  • Face mill both to get flat surface.

This could make it really easy to get good dual sided pcbs. Process would be:

  1. load prep pate, clamp down pcb in corner
  2. home
  3. run pre-written g-code
  4. swap plates
  5. home.
  6. Place pcb in dowel pins, glue, mill side 1
  7. Flip pcb (release, reglue)
  8. Mill side 2
  9. Remove, release CA glue with acetone
Personal tools