Building an Open Source Self Balancing Scooter

Malcolm Faed malc@ihug.co.nz

Copyright: This document may only be reproduced with the author’s permission.

Introduction

Parts List

• Wheels with motors and reduction gearboxes - Electric Wheelchair - AUD$100
  
• 12mm Plywood to stand on and to hold the electronics - Free
  
• Batteries (3 x 12 volts) - Free with Wheelchair
  
• PWM motor controller – Locked Anti-phase for 2 motors - AUD$280
  
• Sensors – Accelerometer and Gyro - USD$114.95
  
• CPU – Atmel ATMega16 - AUD$43.76
• Firmware from Geoffrey Bennett's Web Site - Free
• Schematic- Interconnect Chart
  
• RC Joystick
• Misc cable and connectors.
  

Chassis

• http://www.metalite.com.au/ (Russell Cragg)
• Old electric wheelchair, cut chassis with grinder cutting wheel to suit
• Has 2x 24v motors already conveniently attached.

Batteries

• Conveniently from old electric wheelchair
• 36 volts total
• 3 x 24ah, 12volt SLA batteries.
• Charged in parallel, run in series.
• 60A circuit breaker doubles as a switch
• Position the batteries in such a way that the scooter is naturally balanced when unpowered.
  

>

Motor Controller

• IBC From RoboWars. http://www.robowars.org/store.html
• Two x 50Amp MOSFET controllers
• Replaced micro with Atmel ATtiny2313 and firmware to give locked anti phase control instead of default RC PWM inout.. (Bascom Firmware Download)
  
• Low dropout switching regulator
• Regenerative braking (inherent from locked anti phase motor control)
• Australian design and open source

Microprocessor

• Atmel ATMega16
• 8 bit
• 16k Flash
• 8MHz Clock
• 512 Bytes EEPROM
• JTAG
• ISP
• 3 Timer – Counters
• 8 channel 10 bit a-d
• Hardware USART
• 32 IO lines
• Many other features
• Development in AVR-GCC
• Prototype board from http://www.futurlec.com.au

Sensors

Software

• Developed with AVR-GCC
• Reads the analogue inputs
• Drive the motors
• Read from RC receiver
• Write and read the serial port for logging and tuning
• Download source from Geoffrey D Bennett’s web site

Balancing

• If the robot is leaning, drive the wheels in the direction of the lean
• f you lean more, go faster
• This is a control systems problem:
• input variable is the platform angle
• output variable is motor speed (PWM). 50% duty cycle is stationary.
• by controlling the motor speed and direction, keep the platform horizontal
• Attempt to keep the platform angle zero
• output = Kp x input + Kd x input’ + Ki x input dt
• Proportional — if you lean more, go faster
• Derivative — if you lean quickly, go faster
• Integral — if you’re still leaning, go faster

Turning

• Joystick for input (also RC)
• Speed up one wheel one wheel
• Slow down the other wheel. 
• by the same amount, to maintain balance!
• Turn faster when stationary. Turn slower when traveling.

• Accelerometer and gyro must be in center between the wheels otherwise it senses turning as tipping.

Measuring the platform angle

• The Gyro input gives the angular rate but it drifts a lot
• The Accelerometer input gives the acceleration due to gravity and due to the scooter accelerating
• When the scooter is not accelerating, the inverse-sine of the accelerometer input gives us angle
• This isn’t very precise
• It is susceptible to vibration
• But it doesn’t drift
• How to combine them?
• The Gyro input averaged over time should be zero
• This lets the software adjust for drift
• We assume gyro starts off stationary
• We assume initial angle is given by the accelerometer
• Track angle changes with the gyro
• If the tracked angle is different to the angle from the accelerometer, track towards it slowly
• Subtract any acceleration force we apply to the motors from the accelerometer reading
• When moving, rely less on the accelerometer reading
• sin(x) ~ (for small x) This omits the need for Trigonometry in the software.

Over Speed

• At some point, the motors can’t go any faster
• Then you fall, because the wheels can’t keep up with you
• Solution: before that happens, “push back” on the rider by making the wheels go even faster

Misc.

• Soft Start
• In the first few seconds of running, slowly ramp up to avoid lurching.
• Voltage Reference - ADXRS401 has a 2.5V precision output
• Used to measure regulator voltage and scale inputs appropriately
• Use a-d to measure battery voltage and scale outputs appropriately to compensate for discharging battery
• Tuning
• Uses Perl-GTK on Linux with graphical UI in Linux from Geoffrey Bennett.
  
• Tipped
• If the platform angle is “too large”, shut down

(Adapted from Geoffrey D Bennett presentation)

Future Improvements

• Better Heat sink for MOSFETs
• Over current monitoring (LED)
• Low battery voltage warning (LED)
• Detect when rider not present / Present and adjust balance parameters accordingly
• If the kill or dead-man switch is tripped, shut down

Links, credits and inspirational people.

This project would not have been possible without the assistance of the following people.

Geoffrey D Bennett – for the source code and tuning software
http://www.netcraft.com.au/geoffrey/meta

Trevor Blackwell
http://www.tlb.org/scooter.html