Article about the BOSS robot by David Geer

Page 1

SERVO 11.2006

Humble Beginnings;

Hey, Add a Motor to

Me, Will You?

Starting from a “standard, small

frame” shopping cart, the Battery

Operated Smart Servant (BOSS)

creator Greg Garcia — graduate

research assistant in the center for

intelligent machines and robotics, the

University of Florida — modified the

future BOSS robot’s wheels and

chassis by removing the default

wheels using a “4.5 inch angle

grinder with cutoff wheel.”

Then, on a newly fabricated L

bracket, Garcia mounted the robot-to-

be’s motors, having traced out and

drilled bolt holes in a pattern matching

the holes in the motors’ base. He also

drilled a clearance hole for the motor

shaft to traverse. The remaining,

opposite surface of the L bracket was

left untouched for easy welding to

the cart.

BOSS employs Denso 12-volt DC

motors. These geared, right angle

DC motors produce greater torque,

suitable for those ‘high-test’ applica-

tions. “The worm gear [used in these

motors] translates the rotational effort

of the motor about an axis perpendicu-

lar to the motor; this keeps the

motors in profile along the vehicle,”

says Garcia.

These motors can turn at 150 rev-

olutions per minute. Together with the

eight-inch tires that Garcia specified,

this enables BOSS to move fast enough

to keep up with most people.

The motors are powered by a

series of four 12-volt batteries wired in

parallel in order to form one larger 12-

volt battery (to wire them in parallel, all

the negative terminals are connected

each one to the next, and likewise with

the positive terminals).

Keeping the BOSS

Under Control

The motors are controlled by an H-

bridge motor controller — the Tecel

model D200 — which is compatible

with motors requiring up to 60 amps of

current. The controller also uses four

110 amp MOSFETs (metal-oxide semi-

conductor field-effect transistor).

The controller blocks, allows, and

directs current using gates or switches

in order to switch the motors on and

off and to guide the motor’s rotation

directions. You control the H-bridge by

sending pulse width modulated signals

that set the motor duty cycles (speed);

motor direction is set by sending digital

signals that express the selected direc-

tion that the motors — and ultimately

BOSS — should travel.

The primary data in/out for BOSS

is controlled by a MAVRIC IIB microcon-

troller board. The board employs an

Contact the author at geercom@alltel.net

by David Geer

Battery Operated Smart Servants

Know You’re the BOSS!

Robotic Assistance is in the Basket

BOSS at the beginning — just a

simple shopping cart.

This is BOSS’ preliminary structural

design for placement of the battery.

Deciding where to put the

electronics enclosure.

Page 2

GEERHEAD

Atmega128 MCU (microcontroller

unit). The MAVRIC was chosen for its

versatility when applied to multiple

sensors and input and output data. An

analog-to-digital converter converts

analog data from three infrared

distance-measuring sensors to a digital

signal the controller can understand.

Digital I/O also delivers signals

between the sonar sensor and the

warning buzzer. The signals traverse a

coil relay to turn the buzzer on; I/O

also sends data to the LCD. RS232

serial connections are used so that

the microcontroller can talk to the

image-processing computer.

The microcontroller programming

can be coded using a variety of

programming languages. “I used a soft-

ware package called CodeVision, which

consists of a number of C libraries that

contain function calls designed to

interface with the ATmega128. The

programming of the vision algorithm

was also done using the C program-

ming language. It was accomplished

using Microsoft Visual Studio 6 and

Intel’s OpenCV (open source computer

vision) libraries,” says Garcia.

The BOSS Can See You

The BOSS’ vision program pulls

video input from a Logitech USB quick

cam. Garcia used his own image pro-

cessing algorithm runs in order to cre-

ate the proper reactive control action

and send it to the microcontroller. “This

level of processor computation is too

great for the microcontroller to handle,

so it is outsourced to a higher power

computer. This vision functionality is a

very primitive solution to a complicated

problem,” says Garcia.

BOSS uses its vision facility to track

objects based on color recognition. By

adding other functionalities, Garcia was

able to make this simple tracking modal-

ity much more robust. Extraneous

factors like lighting and shadowing

affect the quality of color recognition

that can be performed. To offset this,

Garcia gave BOSS a training function

and built tolerances into his algorithm,

to allow for variances after the training

Drive motors with noise

reduction circuitry.

At this stage, BOSS’ battery is

mounted and secured.

The battery secured.

SERVO 11.2006

Motor controller housing unit.

BOSS with battery kill switch.

BOSS side view, early stages.

During the creation of BOSS, Garcia

learned that noise has a highly detrimen-

tal effect on sensor information. He also

learned how to avoid it:

• Use good shield wiring.

• Run power lines separate from data

lines whenever possible.

• Mitigate noise from DC motors. “There

are many simple ways to filter out this

noise using a combination of capacitors

and resistors. For example, on my drive

motors I used some 10 microfarad

ceramic capacitors to reduce the noise

output. One capacitor was bridged

across the positive and negative voltage

leads, while two more were ran from the

positive lead to the motor case ground

and the negative lead to the same motor

case ground,” says Garcia.

LESSONS LEARNED ABOUT CABLING

Page 3

SERVO 11.2006

phase of color-based object recognition.

When the vision program begins, a

prompt asks the user to select the color

she/he is wearing from a list of avail-

able colors. The user must then obscure

the camera’s view with what they are

wearing. “This allows the computer to

know its color under the given lighting

conditions,” says Garcia. Tolerances are

built into the Red-Blue-Green color

values to mitigate minor lighting

changes. If BOSS can’t gather enough

color information based on the image

for the algorithm to make a decision,

the robot will stop and sound its buzzer

to let the user know there is a problem.

Makes Sense to

Follow Along

More than a shopping cart with a

brain and sensors, BOSS homes in on

its master’s image to follow and deliver

its cargo.

BOSS uses both IR and sonar rang-

ing sensors for positioning. In this way,

BOSS can know its relative distance

from objects in it environment and

know its own positioning. The sonar

itself is specifically for ranging the

distance between BOSS and its master.

“This sensor is mounted inline with

the camera so it measures the distance

to whatever the camera is pointing at.

This sensor controls the cart’s following

behavior. If the cart is too far away from

the target, it will speed up. Conversely,

if it is too close it will slow down or

stop. But, it will stop close enough to

you to not hit you and yet allow you to

put things in the cart,” says Garcia.

The sonar is a Devantech SRF 05

ranger. It works by transmitting ultrason-

ic pulses. It counts the time between

sending the pulse and hearing its echo to

determine distance, as sonars generally

do. The sonar’s pulse can’t be detected

by the human ear. This particular sonar

has a range of up to four meters.

The IR sensors are Sharp GPD 12s.

These are mounted close to the

ground. The mounting allows BOSS to

see in front of itself and to each side.

These sensors locate minor obstacles in

BOSS’ path. They are limited in range,

which is sufficient for their task. These

sensors can sense up to 30 centimeters.

“When an obstacle is detected, the

cart will stop and sound an audible

tone alerting the user that the path is

blocked. These sensors are currently

used to prevent the cart from having a

collision. Based on empirical tests, it

has been shown that if the user follows

a clear path, that the cart too will

follow said path,” says Garcia.

Why, BOSS?

Garcia is most frequently asked

why he built BOSS and how he got

the idea. It was mostly to satisfy

a class requirement (EEL 5666c

Intelligent

Machines

Design

Laboratory). The freedom to design

and create what he wanted — with-

in the class requirements — was a

factor in his enthusiasm for the job.

GEERHEAD

Here the warning light has

been mounted.

BOSS computer brains, open and

ready for inspection.

BOSS front view; IR sensors

have been mounted.

A rear view of BOSS’

electronics enclosure.

This is BOSS’ motor actuation box.

“Nothing is as it seems,” says

Garcia, to borrow an old adage. Greg

learned that actions and motions like

pushing a shopping cart or following

someone become very difficult when

you try to reproduce them in machines

alone. It requires “incredible” computer

power and sensing. These behaviors

are easy for us because we recognize

not only color but also shape and other

characteristics. “We don’t have to pull

out a tape measure and check each

doorway, we glance at it and know

whether we can pass or not. We can

also step over small boxes which a

wheeled robot cannot do,” says Garcia.

BROADER LESSONS LEARNED

Page 4

“I believe that robots should be

used primarily to save human lives

where at all possible. But, if that is not

an option, they should at least be able

to enhance our quality of living,” says

Garcia. It was the latter motivation that

lead Garcia to come up with plans for

BOSS. People can easily get caught up

in the drudgery of every day chores and

tasks, for example, when Garcia had to

go along to the grocery store as a kid.

“You see, I have an older sister. We

always wanted something to do in the

store while we were with our mother

so we would ask to push the cart. I

guess she really didn’t like doing it

because she would always agree.

Anyway, my sister would somehow

always find a way to “accidentally”

bump into me. So I started thinking of

how to avoid this,” explains Garcia.

The more Garcia pondered the

problem and those memories, the

more he remembered the nuisances

of using a shopping cart. Sometimes

people forget where they’ve left their

carts. Then, they have to go back and

get them. That’s how Garcia arrived at

the solution he calls BOSS.

Beyond this, Garcia sees the possibil-

ity that a smart servant could help the

disabled. “The same day the AP story

broke about our robots, I got an email

from a woman who is disabled. She has

limited use of one arm and suffers

intense pain when having to single-hand-

edly push a shopping cart. She requested

that I keep her up to date on the devel-

opment of this technology,” adds Garcia.

Conclusion

Garcia simply added eyes, brains,

and mobility to a shopping cart. And,

yet, it’s clearly one more step toward

the close proximity we will all share

one day with the many servant and

co-worker robots to come! SV

SERVO 11.2006

GEERHEAD

At this stage of the build, the

warning light has been repositioned.

NOVA 7896 SBC with CIMAR PS,

MAVRIC IIB, and LCD all mounted

and powered in box.

No, the BOSS wasn’t invited to join

the Beach Boys, but it does “get

around” in a really cool way. The BOSS

uses skid steering to move and groove.

This means that two front caster

wheels and two rear multidirectional

orientation wheels are employed.

By telling the rear wheels to spin at

varying speeds or opposite directions,

BOSS can make zero turn radius turns,

according to Garcia. BOSS doesn’t do

reverse yet, as there are no sensors in

the rear to guide it.

BOSS GETS AROUND

The BOSS at home on its

Web-based crib

http://grove.ufl.edu/garcia/

BOSS in action. Catch the video here:

http://grove.ufl.edu/garcia/BOSS_

internet.wmv

Tecel motor driver board

www.tecel.com/d200

MAVRIC controller board

www.bdmicro.com/mavric-iib

Logitech

www.logitech.com

Devantech

RESOURCES