Debugging a DIY Segway at the Vancouver Hack Space

Al first contacted me by email.  Apparently someone has been telling him that I’m the go-to guy for anything Arduino.  He is not wrong. Sit right here and I’ll tell you about how we went to work on this machine and how far we got in an evening.

A Segway PT drives by sensing the way you tilt your body.  The brain of the robot can sense which way is up and which way you lean.  It does this with a gyroscope.  This model also has an Dead Man’s Switch for safety: you have to hold the button down to make it go, meaning it can’t make any escape attempts.  In order to test the parts early on there’s also a knob to make the robot turn and a knob to make the robot go forward and back.  In theory you could drive this model without the robot brain, but that’s not as cool.

Al did what anyone would do.  He carefully assembled the entire machine by following the steps in the manual, copied the code into the Arduino line by line, and when it was ready flipped the switch.  One wheel started to turn right away and the other did nothing – and nobody was holding down the Dead Man’s switch!

To attack this problem efficiently we made a plan: if we know the fundamentals work then we can test higher-up stuff.  First we have to check the mechanical and electrical connections match the diagram.  Second, we have to check that the inputs work and the Arduino can “hear” them.   Third, test that the brain can make the motors go the right way.  Fourth, we know we have an input > brain > motors connection.  The last step then is teaching the brain how to balance.

We disconnected the motors and then Al got out a sharpie and some masking tape.  I gently pulled on a wire at the top of the steering controls and when the other end at the Arduino moved, he put a piece of tape on and labelled it.  We did that through all the wires, which took about 20 minutes.  You can see a ‘T’ (for turn) on a piece of tape next to Al’s index finger.

Al, being very clever, had drawn a schematic of how the wiring should be done.  We took out all the wires from the red breadboard shield, effectively disconnecting all the controls.  The breadboard has the gyroscope chip mounted in the center for easy connecting to the Arduino.  The breadboard and the Arduino click together like a LEGO sandwich.  Then we used the traced Dead Man’s wire and reconnected it to the pin in the schematic and went poking in the code.  About 40 minutes later we had the light on the top of the breadboard turning on when the Dead Man’s Switch was pushed, and turning off when it was released.  This was an excellent sign!  We didn’t stop for high fives, there was a long way to go.

The next step was to get the knobs and the button all working at once.  To do this I wanted to display on a screen the position of the knobs and the on/off of the button all at once.  The best way I know to debug an Arduino is the Serial interface.  In the code we have a value called steeringPot and another called gainPot.  I wanted to be sure that when the steering knob is turned that only analogRead(steeringPot); would return a changing number between 0 and 1023.  I also wanted analogRead(gainPot) to do work the same when the forward/back knob was turned.  I told the Arduino,

Serial.print(analogRead(steeringPot));  Serial.print('\t');
Serial.print(analogRead(gainPot));  Serial.print('\t');
Serial.print(deadManSwitchState);  Serial.print('\n');

…This is where we hit the biggest bug of the night.  We had the code reporting

1023   1023   1

over and over, no matter what we did to the knobs.  We retested everything.  The always resourceful Simon joined in with a multimeter and tested that the electrical signal from the knob potentiometers was rising and falling as the knob was turned.  It’s his hand in the left of the photo when we tried removing the breadboard to see if it was somehow the problem.  Everything seemed to be working, and yet the Arduino was deaf.  For nearly and hours various people took a crack at it.  In the end, it was a function I’d never used before called analogReference(INTERNAL); that was the culprit.  Nobody caught that the AREF wire in the schematic was missing; the one in the board was removed because it wasn’t in the schematic; and the Arduino was waiting for voltages it would never get.  We removed the line and suddenly we were getting numbers that made sense.

By then it was 11pm and time to go home.  Al promised to work on connecting the 6DOF sensor over the week.  If he does we’ll take another crack at it Tuesday night and you can read all about it here Wednesday afternoon.

I should probably mention along the way John cleaned a 1000 watt amplifier and started putting together the Solidworks computer; Miriam got her Makerbot Cupcake 3D printer moving; Jack was laser cutting; and there were at least 9 other people with as many projects going on that I didn’t even get to talk to.  Come on down and check it out sometime, it’s really something.