The MAV - Mono Axial Vehicle is
a
project aimed at building a 2 wheeled balancing vehicle that can
function as a
human transporter as well as autonomously.
Two wheeled
inverted pendulum vehicles have become practical for amateurs to build
in the
last decade thanks to miniature gyroscopes and accelerometers in
standard IC
packaging. Also the
availability of open
source programming tools on the internet has made it easier to get
started
working with microcontrollers than it has ever been before. Electronics has also
become more open to
amateurs with the valuable resources of electronics newsgroups for
asking
detailed technical or beginner questions, as well as online groups
dedicated to
open source electronics designs, such as motor controllers. The MAV – mono
axial vehicle project has only
been possible to create because of the resources mentioned above.
The MAV is
a two wheel balancing vehicle similar to the Segway™ human
transporter
vehicle. Inertial
sensors are used to
balance the vehicle and keep it upright when the rider drives forwards
and
backwards by leaning. A
two wheel
vehicle can remain horizontal while driving up and down steep hills,
which is a
major advantage over a vehicle that has passive stability with another
axle,
which will tip the vehicle backwards when climbing a steep hill. Another major benefit of
having only two
wheels is the active stability required for balance also allows for
better
dynamic performance as the center of gravity of the vehicle can be
shifted
forwards and backwards allowing for much faster acceleration and
braking times. The
MAV can also be used as a two wheeled
robot, with either R/C control or autonomous operation.
To switch from riding the MAV to operating it
as a robot, simply unbolt the handlebar, turn the chassis 180degrees,
and bolt
on the robots torso. The
accelerometer
detects when the platform is flipped upside down, and the robot will
start
balancing as soon as the key-switch is turned on.
The design is modular so that future robot
torso’s can be developed which may include sensors for
robot-vision,
robotic-limbs for opening doors, and also increased computing power. It is possible that a
robotic torso could be
developed that would be able to provide assistance with chores around
the
house.
Notable Features
The MAV – mono axial vehicle
The MAV is designed using homebrew modular electronic boards which are assembled together into a stack. The assembly consists of the MAV control board, shown as the yellow box in figure1, two MAV motor control boards, shown as the two green boxes in figure1, the MAV inertial sensor board, shown as the red box in figure1, and the MAV power supply board, shown as the light blue box in figure1. There are also two rotary encoder boards mounted on the MAV wheel shafts, shown as the two blue boxes in figure1. There are a total of five Atmel AVR microprocessors in the MAV, shown as MCU1 through MCU5 in figure1. Despite being fairly complex this hardware bundle is very compact and all fits into one corner of the MAV chassis. The hardware bundle can be seen in figure 2.
Figure 1: MAV Overview
Figure 2: The
Complete MAV hardware
bundle shown with its custom case
The MAV
Control Board uses an ATMega32 to gather all of the information from
the
inertial sensors and the wheel encoders and uses this sensor data to
calculate
how much torque is necessary to send to the motors.
The large code space, fast single clock instruction
execution, good peripheral selection and readily available open-source
programming tools made the ATMega32 a good choice, since this is where
the
MAV’s primary control code is located.
The MAV control board also monitors the system’s
supply voltage, and
uses its onboard A/D module to determine the joystick position on the
handlebar. This
board is capable of
logging data for debugging and performance analysis with 2 onboard
16Mbit Atmel
DataFlash chips. This
board also
supplies a filtered voltage supply for the IMU board where several
precision
A/D measurements are taken. Figure 3 shows the Eagle layout of this
board. Figure 4
shows a photograph of the
constructed board, and figure 5 shows the block diagram of the board.
Figure 3 Eagle
Layout of the MAV
control Board
Figure 4
Picture of Constructed MAV
control Board
Figure 5 MAV
control board block
diagram
The MAV
Inertial
Measurement Unit (IMU) Board contains three ADXRS150 gyroscopes and two
dual-axis ADXL202e accelerometers as well as an 8 channel 16bit ADC and
a precision
5Volt reference and supply. The
gyroscope sensors measure the rate of rotation around their vertical
axis at a
maximum rate of 150degrees per second.
The accelerometer sensors measure the force of
acceleration at a maximum
of 2 G’s (1G being the force of gravity).
There are two small daughter-boards each containing one
gyroscope and
one dual axis accelerometer soldered to the IMU board perpendicular so
that
three axis's of rotation can be measured by the sensors. Figure 6 shows
the
block diagram of the IMU. Each
gyroscope
has an internal temperature sensor which is sampled by the MAV control
board. We have
calibrated the gyroscopes
over a wide temperature range to correct their rotation rate outputs. Figure 7 shows the IMU
stacked under the MAV
control board.
Figure 7 6DOF
IMU board (below MAV
control board)
There are two identical modular motor controller boards in the MAV, one for each wheel. These boards were designed with flexibility and robustness in mind and they are able to operate with both brushed DC motors, surplus brushed wheelchair motors in the case of the MAV, and also with brushless DC motors. There is an AVR ATMega168 on each of the motor controller boards and it handles command packets sent to it over the SPI bus or the UART and controls the motor by applying the requested duty cycle to the motor. The PWM (pulse width modulation) frequency is programmable in a range from 300Hz to 30 kHz, and there is also programmable heat sink temperature limits and shunt current limits. The bridge drivers used (IR21844’s) are high voltage devices so the motor controller can operate at a voltage range from 10Volts up to 600Volts. The mosfets and filter capacitors must be rated to the voltage used, which sets the maximum operating voltage of the motor controllers to 63Volts which is the voltage rating of the filter capacitors used. Although the surplus wheelchair motors used in the MAV are only rated to 24Volts. For operation with brushed DC motors, two of the three mosfet half-bridges are used to power the motor, hooked up in an H-bridge or full-bridge configuration. The mosfets used are IRFP2907 75Volt devices with a maximum on-state resistance of 4.5mOhm’s and a continuous current rating of 209Amps, although to maintain this large current level requires more heat sinking than currently used. The board also supports brushless DC (BLDC) electric motors in hall sensor, encoder and also sensorless 6-state phase commutation modes. The BLDC feature was added in anticipation of the next version of the MAV which will have homebrew BLDC motors integrated into the wheels.
Figure 8: MAV
Motor Control Board
block diagram
Figure 9: Eagle
Layout of the motor
control board
The MAV
motor controller board has 3 half bridges made using high power mosfets. For the MAV we use only 2
of the half bridges
since we are controlling 1 DC brushed motor.
This board was designed with several possible applications
in mind:
Figure 10
shows the MAV motor control board during assembly.
Figure 10 MAV
Motor Control Board
The Rotary
Encoder board uses an ATMega168 to decode the Quadrature signal from
the
optical rotary encoder. The signal from the rotary encoder is decoded
into a wheel
position, and the wheel position is sent via the TWI Bus to the MAV
control
board. Figure 11 shows the block diagram of the Rotary Encoder board.
Figure 12
shows the 4 possible Quadrature or gray code states of the optical
encoder. The 2
channels of the
quadrature encoder trigger external interrupts on the ATMega168. The current state of the
encoder is compared
with the last state to determine the rotation direction. The number of interrupts
determines the total
amount of rotation of the wheels.
This
information is stored in a signed long int to give a large rotation
range
forwards or backwards. This
information
is transmitted via the two wire interface to the MAV control board. The MAV control board
requests the current
state of the long int rotation
variable
every 10ms for each of the wheel encoders.
Figure 11:
Rotary Encoder Board
block diagram
Figure 12:
Quadrature States
The
MAV’s
power supply is an efficient step down buck switching circuit with an
input
range from 12 to 80 Volts and a 10V output.
It powers everything in the MAV except for the motors. Efficiency is a
consideration since it powers
5 MCU circuits off of 24Volt battery power.
Figure 14:
Power Supply Board block
diagram
The
MAV’s
software is organized in a simple task loop, with each task running
sequentially. The
frequency of each
tasks execution is set by programmable task timer variables so that a
task can
be set to run at the desired frequency.
The
task timer variable value sets the task period directly in
milliseconds, so a
task timer variable of 50 will cause that task to run once every 50
milliseconds. A
task can also be disabled by setting its
timer variable to zero. For
the MAV
control board, there are several tasks that are running sequentially;
these are
shown below in flowchart 1. It
is
straight forward to add more functionality by adding new tasks, this
has many
advantages to a single thread of code for this reason as well as the
ability to
turn on/off each task as well as change the frequency of each task
independently.
The main
tasks that the MAV runs on the main control board during its balancing
operation are:
Figure 15
shows the software flowchart including a breakdown of the above 4 tasks.
In addition
to this, the other 4 boards that have Atmel microcontrollers on them
also run
software. Each has
an ATMega168 MCU:
A PID feedback
loop is a commonly used method used to control unstable systems. It consists of 3
mathematical terms that are
added as feedback to the output of the system.
The proportional value is the current error of the system,
in this case
the amount of deviation of the platform angle from level. The integral term is the
sampled error
integrated over time. The
derivative
term looks at rate of changes in the system to project the future
correction
factor needed to help stabilize the system.
A PD loop using only the P and D terms was found to be
sufficient to
balance the MAV well enough for both robotic use as well as human
riding modes.
torque = (platform_pitch_degrees *
K1) + (platform_pitch_degrees_per_second *
The MAV
control loop has three parts:
In the
human riding mode the proportional term is the angle of the platform
relative
to the direction of gravity. This
value
is computed by integrating the gyroscope readings and using periodic
corrections from the accelerometers to prevent the integrated gyroscope
angle
from drifting over time. The
derivative
term is the platform rate of tilt which is the calibrated and filtered
output
of the gyroscope.
In the
autonomous mode the calculations are executed in the same way, except
that the
proportional term has a position correction factor on each execution. This term is the distance
away from a user or
AI controlled point. If
the MAV’s
current position is different than this point, then it moves towards it. The P, I and D terms for
position holding in
autonomous mode are shown here:
torque = torque + (platform_position_meters
* K3);
//position proportional
torque = torque +
(platform_meters_per_second * K4);
//position
derivative
torque = torque +
(platform_position_meterSeconds * K5); //position
integral
The C code
for sending a dutycycle update from the MAV Control board to the motor
controller 1 over the SPI Bus is shown here:
MOTOR1_CS_active;
//enable
the motor1 chip select line
printf("$i%05u&i%05u\r",
motor1_dutycycle_int,
motor1_dutycycle_int);
MOTOR1_CS_inactive;
//disable the motor1 chip select line
First the motor1 chip select line is enabled, then the updated dutycycle is sent in packet form using the printf function, which looks like this: “$iVALUE&iVALUE\r”. This sends the same 16bit dutycycle twice as VALUE with ‘$’, ‘I’, and ‘&’ being packet protocol identifiers. The dutycycle value is sent twice each update to prevent dutycycle value errors that we thought may result from motor noise on the SPI bus. Luckily the SPI bus seems to be very robust and motor noise has not been an issue.
Figure 15
– System Overview (MCU1
ATMega32) Flowchart
The MAV
project successfully reached its goals of making a working mono-axial 2
wheel balancing
vehicle that can function robotically and is also human useable. Electronically it is a
compact design that
uses 5 AVR Atmel microprocessors to process analog sensor data and
perform
rapid feedback control calculations.
The
MAV is also a robust design that performs many self checks to ensure
the
various components are working as intended.
The MAV electronic and software design can also be used as
a starting
point for enthusiasts to enter the hugely popular realm of robotics.
ATMega32
“8-bit AVR Microcontroller with
32KBytes In-System Programmable Flash”
ATMega168
“8-bit AVR Microcontroller with
16KBytes In-System Programmable Flash”
WinAVR - open source
development tools for
Atmel AVR microprocessors for Microsoft
Windows
platforms, including GNU GCC compiler and UNIX-like utilities
http://sourceforge.net/projects/winavr/
Eagle Cad - EAGLE Layout
Editor http://www.cadsoft.de/
Olimex - Fast PCB
Prototypes http://www.olimex.com
Digi-Key Corporation,
on-line electronics
part distributor http://www.digikey.com/