Robot Arm

Sixi Robot developments 2019-07-16

New OSHPark PCBs arrived

With this we can finally finish the hardware preparation for shipping units, after which we can also get a final bill of materials and start to package DIY kits.  There will be two flavors:

  • DIY kits with all the non-printable parts.
  • DIY kits with everything, including the printed parts.

Collision detection

The software only lets each joint of the robot move within a safe range to prevent wire twisting or other potential damage.  That piece of code does not consider the angle of joint A in relation to the value of some other joint B.  So it was still possible to make the arm hit itself.

This new collision detection code prevents each bone of the arm from colliding with other bones of the arm.  At present it is a crude collision detection system, using only the box around each joint.

Also now that these boxes have been calculated, the center of each box is known.  the mass of each bone can be pulled from the Fusion360 model, and then a point-mass physics model can be created to simulate forces like gravity.  This gets us closer to dynamics like push-to-teach and telling the arm “please push on this item with a force of N newtons”


Next we’ll be assembling the new shoulder design with the PCBs that should allow us to unplug the umbilical from the arm and the control box.  I’m thinking Twitch stream?  Maybe just a Youtube Live, given the recent Amazon protests.  Choice of evil companies… hmm…


Delta Robot 3 Assembly Manual

Delta Robot 3 uses three motors turned 60 degrees from each other to achieve XYZ translation motion.  It is most popularly known in the Kossel 3D printer.  You might have seen one of our robots at DEFCON being used to type out pin pad combinations to unlock a cell phone.

While we do not make the Delta Robot 3 at this time, we believe in open source software & hardware.  So here is a copy of our internal Delta Robot 3 assembly manual.

This robot was designed at a time when we did not have reliable 3D printers.  Challenge yourself to design STL models!

Special Thanks

Special thanks to Justin Engler, Jae Hess, Steve, Juho Leinonen, Mark Scheel, and Eddy Hodak for helping improve this Delta Robot 3 documentation.

Tools you will need

A soldering iron.

Crazy glue or wood glue. Crazy glue takes seconds to stick. Wood glue can take up to 24h and may require clamping.

Crazy glue.jpg

A small Phillips screwdriver

Small screwdriver.jpg

a pair of pliers

Needle-nosed pliers.jpg

Solder the wires on the switches

Stewart Platform limit switch soldering.JPG

You can also crimp the wires with blade connectors if you have the tools.  These days we prefer non-contact sensors for more accurate position sensing.

Snap together the ball joints

The grip I’m using is the easiest way I’ve found to snap them together. Pinch with the pliers to get most of the way there and then gently press down with one finger and the plastic should “pop” onto the ball.

DeltaRobot8 55.jpg

Glue the end effector together

These three screws are M3x25.  This part could have been 3D printed in a single step.

DeltaRobot8 end effector parts.jpg
DeltaRobot8 end effector assembly layer 1.jpg
DeltaRobot8 end effector assembly layer 2 with glue.jpg
DeltaRobot8 end effector and arm assembly.jpg
DeltaRobot8 end effector arm parts.jpg
DeltaRobot8 end effector arm assembled.jpg
DeltaRobot8 end effector side view.jpg
DeltaRobot8 end effector arm fully glued.jpg
DeltaRobot8 end effector arm fully assembled.jpg

Glue the biceps together

Please be extra careful to glue these correctly. Exact alignment is very important! The screw is M3x15.  This part is much easier to make with a 3D printer where no glue is required.

DeltaRobot8 arm parts.jpg
DeltaRobot8 arms with glue.jpg
DeltaRobot8 arms being assembled.jpg
DeltaRobot8 arms, end view.jpg
DeltaRobot8 26.jpg
DeltaRobot8 arm set screw and nut.jpg

zip tie & glue the stepper mounts together

Please carefully note the direction of assembly.  It is very easy to assemble the entire robot backwards.  Then the firmware won’t match the physical model and everything has to be redone. 

You will need the motor mount plate

DeltaRobot8 motor mount plate.jpg

and the mounting bracket pieces.

DeltaRobot8 motor mount bracket.jpg
DeltaRobot8 motor mount and bracket 1.jpg
DeltaRobot8 motor mount and bracket 2.jpg
DeltaRobot8 motor mount and bracket 3.jpg

Use some glue to attach the triangle face to the side of the motor mount plate.

install the steppers in the stepper mounts

Use M3x10 screws to attach the motors.

DeltaRobot8 motors in motor mount.jpg
DeltaRobot8 motors in motor mount 3.jpg
DeltaRobot8 motors in motor mount complete.jpg

attach the biceps to the stepper motors

Depending on availability you may have received MOTO-0007 instead of MOTO-0003. Please see the links for details on the differences and what to do about it.

DeltaRobot8 adding arm to motor shaft.jpg

The arm will only fit when the flat side of the shaft is correctly lined up with the screw in the arm.

DeltaRobot8 adding arm to motor shaft 2.jpg

I am using a spacer to make sure that the biceps are attached at the correct distance from the face of the motor. The spacer should not be glued in! It is only temporary.

DeltaRobot8 50.jpg

This is a closeup view of all three arms attached.

DeltaRobot8 51.jpg

zip tie the three frame components together

Sometimes humidity might make your unassembled pieces a bit warped. When they are attached together with the zip ties they will straighten out.

DeltaRobot8 frame side.jpg
DeltaRobot8 frame parts zip tie process.jpg
DeltaRobot8 frame parts with loose zip ties.jpg

close the frame around the stepper mount

DeltaRobot8 motor mount into frame 2.jpg
DeltaRobot8 securing motor mount to frame.jpg
DeltaRobot8 securing motor mount to frame 2.jpg

Attach the base plate into the frame

There are many holes in this plate so that you have options when securing your work piece.

DeltaRobot8 base.jpg
DeltaRobot8 frame and base.jpg

Attach the forearms to the end effector

In more recent robots our process is to precisely assemble the plastic rod ends onto the rods, then insert the balls, THEN attach the rods to the rest of the robot.  The results are more consistent and accurate.  What you see here is our process way back when.

DeltaRobot8 56.jpg

These pictures show M3x40 screws, but we have changed the kits to use M3x30 screws.

DeltaRobot8 57.jpg
DeltaRobot8 58.jpg

Attach the forearms to the biceps

The screws are M3x30. The rods are M3x150.

DeltaRobot8 59.jpg

I use the M3x30 screw as leverage to make screwing in the rod a lot easier.

DeltaRobot8 63.jpg

The distance from center hole to center hole should be 165mm.

DeltaRobot8 64.jpg

The two M3x30 screws should be parallel so that the whole assembly makes a parallelogram. When I get close it can take a few tries with half turns to get the measurements just right.

DeltaRobot8 62.jpg
DeltaRobot8 61.jpg
DeltaRobot8 60.jpg

Assemble the pen collet

The collet holds your pen in place in the end effector. If you drive the tool too far down the collet will rise on the screws. It is our hope that this will help prevent damage to your machine. A 15mm screw and a nut are used between the glued layers. The capacitive pen is here for reference, it is not included in your kit.

DeltaRobot3 collet.JPG


Attach the switches on the frame

Make sure the zip tie is tight!

DeltaRobot8 70.jpg

The funny angle is OK, and expected. You should be able to move the arm and touch the switch so that it clicks.  This is the kind of amateur junk that made us move to touchless switches that don’t have “funny angles”

DeltaRobot8 69.jpg

Attach the female plug to the RUMBA

DeltaRobot8 66.jpg
DeltaRobot8 65.jpg

attach the RUMBA controller on the frame

DeltaRobot8 67.jpg

attach the switches to the RUMBA

The switch next to motor 1 should go to X-. The switch for motor 2 should go to X+. The switch for motor 3 should go to Y-.

Just above the pins are three symbols: +, -, and S. The red should go to +, the black to -, and the blue to S.

DeltaRobot3 endstops.jpg

Attach the stepper motors to the RUMBA

Steppers are attached in counter-clockwise order on the frame and wired left-to-right order on the control board.

Depending on availability you may have received MOTO-0007 instead of MOTO-0003. Please see the links for details on the differences and what to do about it.

DeltaRobot8 68.jpg

Teach the Controller

The brain of the robot is the RUMBA circuit board. To teach the circuit board you will need the Arduino software, the RUMBA driver (so windows/arduino can talk to RUMBA), and the firmware sketch that Arduino will load onto the board.

  • Download Arduino software and install it. Arduino software comes with lots of great sample code to help you start learning about programming robots. You can use it to read and edit the firmware for the robot.
  • Windows cannot find the drivers for the RUMBA automatically. Get them here.
  • The firmware is a program that teaches your RUMBA board how to talk to the PC and how to move the motors. Get the latest Delta Robot Firmware here.

Now that you have all the prerequisites, let’s do this!

Start Arduino software.

Choose Tools > Board > Mega 2560


Choose Tools > Serial Port

  • On OSX it should say something like /dev/tty.usbmodem12# You probably want the first one on the list.
  • On Windows it should say something like COM9. You probably want the last one in the list.
   When your board is connected to a Windows computer it will appear in Start > Devices and Printers along with the COM port number.

Choose upload. It should be the second blue button on the left. At the bottom of the Arduino window it will say “compiling” and then “uploading” and finally “uploading done”. if you get timeout errors, do these steps again on a faster computer.

Check the Controller is OK

Open the Arduino Serial Monitor (the magnifying glass in the top right corner). Change the baud rate to 57600 (the drop down list in the bottom right corner).

You should see a message beginning with “Delta Robot v8-“. This means the code has uploaded successfully. It should now be safe to connect 12v power.

Put one hand on the power connector so if movement looks wrong you can pull the plug immediately. This is your Emergency stop. Better yet, have a friend ready.

Type “G28;” This tells the machine to find home by touching each limit switch one at a time.

Type “M114;” This asks the machine for it’s current position. It should start with


The machine believes that the tool in the end effector is 0cm long.

Type “G54 Z-2;” Now the machine thinks that tool tip is 2cm below the end effector. Type “M114;” again and it should say


because the tool tip has moved, even if the end effector didn’t. Now let’s try our first move command.

Type “G0 Z0;” This will lower the end effector until the tool tip touches the bed. Since there is no tool at the moment it should stop before it reaches the bed.

Let’s try different speeds now. Type “G0 X4 F3000;” The F3000 changes the feed rate – the speed – to 3000 microsteps per minute.

Finished Delta Robot 3!

The steppers get very warm after running for a while. We have tested them under load for 48h and they continue to function within spec without unusual behavior. Even so, be smart: do not leave your robot running unattended.

Auto home

The robot does not G28 automatically when it turns on to avoid the chance of accident. If you want to change that you can add


to the end of setup() in the firmware.

200 step per turn motors

in configure.h there is a define for STEPS_PER_TURN. Change it to 200 if you have 200-step-per-turn (1.8 degree) motors.

Final thoughts

These days (being the days during which I write this post) Marginally Clever Robots do everything to avoid the use of glue.  We like robots that can be un-made or repaired.

Did you know Delta Robot 3 is simulated in Robot Overlord?  Connect one to the other and drive both at once!

Robot Arm

Can Machine Learning Improve Robot Kinematics?

I’ve tried several times to hand-code inverse kinematics for robot industrial arms in Robot Overlord.  To make a long story short, there are a lot of complicated edge cases that break my brain.  Many modern methods involve a lot of guess work in the path planning.  I know that a well trained Machine Learning agent could do the job much better, but to date there are none I can download and install in my robot.  So I’m going to try and do it myself.  Join me!

The problem I’m trying to solve with Machine Learning

I have a 3D model of my arm in the Java app called Robot Overlord.  The 3D model is fully posable.  At any given pose I can calculate the angle of every joint and the exact location and orientation of the finger tip.  I have Forward Kinematics (FK) which is a tool to translate joint angles into finger tip.  I have Inverse Kinematics (IK) which is a tool to translate the reverse.

A robot arm is programmed with a series of poses.  Go to pose A, close the gripper, Go to pose B, insert the part, Go to pose C, etc… The robot software has to calculate the movement of the arm between poses and then adjust every motor simultaneously to drive the finger tip along the path between the two poses.  I’ve already solved the firmware part to drive six motors given sets of joint angles.

The problem is that one IK solution there might be many combinations of joint angles – sometimes infinite solutions.  To illustrate this, hold a finger on the table and move your elbow.  Your finger tip didn’t move and you had lots of possible wrist/shoulder changes.  As the arm moves through space it can cross a singularity – one of the spots with infinite solutions – and when it comes out the other side the hand-written solution flips the some or all of the arm 180 degrees around.  A smarter system would have recognized the problem and (for instance) kept the elbow to the side.  I have tried to write better IK code but have not had any success.

My Machine Learning Plan of Attack

My plan is to use a Deep Learning Neural Network.  The DNN is a bit of a black box: on one side there is a layer of Inputs, on the other there is a layer of Outputs, and in between there are one or more hidden layers.  Inputs filter through the layers and come out as Outputs.  The magic is in the filtration process!  The filter can be trained with gradient descent using a cost function – if I can score every input/output combination I can let the DNN play with the virtual arm while the cost function watches and says “good, bad, better, worse” until the two work out all the best possible movements.

My Machine Learning Network design

I believe my inputs should be:

  • Arm starting pose: 6 random angle values.  Because DNN inputs are values in the range 0…1 I’ll say 0 is 0 degrees and 1 is 360 degrees.
  • Arm ending pose: 3 random position values and 3 random angle values.  Position values are scaled over the total movement range of the robot.  So if the robot can move on the X axis from -50 to +50, (x/100+0.5) would give a value 0…1.
  • Interpolation between both poses: 1 decimal number.  0 means at the start pose and 1 means at the end pose.

I want my outputs to be:

  • Arm joint angles: 6 angle values.
  • confidence flag: 1 number.  0 means “I definitely can’t reach that pose” and 1 means “I can reach that pose”.

The cost function should work in two steps:

  1. make sure there is no error in the joint value – that is to say, the finger is actually on the path where it should be, and
  2. seek to reduce joint acceleration.  Adjust the elbow and the wrist ahead of time to avoid the need to suddenly twist.

I’m going to try first with two hidden layers, then experiment from there.  I intuitively guess it will take at least two layers because there are two parts to the cost function.

My Machine Learning code setup

Robot Overlord source code is already written in Java so I’ve added TensorFlow and DL4J.  Currently I’m still walking through the MNIST quickstart tutorials and asking the DL4J chat room for help.  They already solved a few head scratching differences between the DL4J quickstart tutorials and the DL4J up-to-date examples.  You can find my first test in Robot Overlord’s code at /src/main/java/com/marginallyclever/robotOverlord/


I hope that I’ve described my challenge thoroughly.  Please feel free to look at the code and make pull requests, or comment below or in the forums with any tips and advice you might have.  If you’re feeling helpful but not sure how, please share this far and wide so that I can reach the people who have the DNN know-how.

Stay awesome!