You are on page 1of 24

Robotics Research

l^hnical Report

pcvi*

"Door Opening" Experiments


Using the Four Finger Manipulator
by

Gerardo Lafferriere

Deepak Mohan
Technical Report No. 280
Robotics Report No. 100
March, 1987

New York University


Institute of

Mathematical Sciences

Computer Science Division


25 Mercer Street New York, NX 00
1

"Door Opening" Experiments


Using the Four Finger Manipulator
by

Gerardo Lafferriere

Deepak Mohan
Technical Report No. 280
Robotics Report No. 100
March, 1987

New York University


Dept. of Computer Science
Courant Institute of Mathematical Sciences
251 Mercer Street
New

York,

New York

10012

Work on this paper has been supported by Office of Naval Research Grant N00014-82-K0381, National Science Foundation CER Grant DCR-8320085, and by grants from the Digital
Equipment Corporation and

the

IBM

Corporation.

"Door Opening" experiments using the


Four Finger Manipulator.

by:

Gerardo Lafferriere

Deepak Mohan

ABSTRACT
Software has been developed that enables the Four Finger Manipulator to "open a
door". The axis of rotation of the door is unknown, and the robot uses force/position
feedback information to perform the task. Two variants of the problem have been
implemented. In the first case the robot "holds a knob and opens a door "(a simple
door with no external forces), and in the second case the robot "pushes open a spring
loaded door".

March

11th 1987

1.

INTRODUCTION
Most robots

in

use today use position control and hence have to be

accurately and in great detail to perform specific tasks.

Such

programmed very

an approach has

many draw-

backs, such as poor adaptability to changes in the environment and great sensitivity to model-

Robots using force feedback control could

ing errors.

However, force feedback

adaptable and safer.

well understood at the present time. (See


associated

with

force

become

control

cooperate to perform a task, or

NYU

in the

in principle

be more versatile, more

control strategies for real time robots arc not

[6] for a

survey of such strategies.) The problems

particularly

complex when several

robots

must

case of dextrous manipulation by several fingers.

Four Finger Manipulator described

in the

next section

is

The

a device built to study control

edgorithms to deal with such situations.


In this report

form.

This task

is

we

may

often have to per-

about an unknown axis of rotation.

Typical examples

focus on one specific type of task that robots

to rotate an object

of such a situation are:


a)

Pulling a door

b)

Turning a crank.

c)

Turning

wrench.

d)

Pushing

door open.

knob

to

In the absence of the

open

a door.

knowledge of the

only on the feedback information that


available (force, vision, etc.).

it

axis of rotation, the robot

must move depending

can gather. This information depends on the sensors

control algorithm must then utilize this information effec-

tively.

We

have developed software that enables our "Four Finger Manipulator"

above task using force/position feedback information.

Two

to

perform the

variants of this problem have

been considered.
1)

Task

1:

"Hold

knob and open

a simple door".

In this case there are

no external

forces acting on the door, and initially an approximate starting direction has to be specified.

2)

Task
door.

2:

"Push open

a spring loaded door".

In this case an external force acts

on

the

-2

2.

THE FOUR FINGER MANIPULATOR


The Four Finger Manipulator (FFM)

fabricated at the

NYU

two dimensional robot designed and

Robotics Laboratory to study dextrous manipulation and problems

associated with real time control of robots.

many

a simple

is

Although the

FFM

is

two dimensional robot,

of the ideas/algorithms developed for this system can be extended to a three dimen-

sional arrangement.

FFM

The

schematic representation of the system (top view)

is

shown

in Fig. 1.

consists of four identical and independently controllable fingers (see Fig. 2)

mounted on the four

sides of a square table.

Each finger can be moved

X-Y

in the

plane

with the aid of two stepper motors connected to the finger through a system of carriages and
In

links.

all

the robot has eight degrees of freedom

and has two pairs of

ant,

strain

The accuracy of

force acting on the finger.


eight stepper motors

attached to

it.

is

gauges that enable

controlled by an

it

(2x4
to

sense the

the strain gauges

INTEL

Each finger

fingers).

is

and

is

compli-

Y component

about 0.01 N.

of the

Each of the

8088 processor and has a position encoder

These processors, along with the other precision components provide a posi-

tion accuracy of about 0.001 cm. Together, the strain gauges and

motor encoders provide

all

the information required for force/position control of the system.

The host computer


running the

UNIX

for the

FFM

is

operating system.

SUN

workstation with a Motorola 68000 processor

This workstation serves as the primary system for

software development, data storage, controller and data analysis.


level controller also with a

real time version of

68000 processor running the

UNIX

NRTX

The

The
etc.)

is

NRTX

is

software incorporates data analysis

etc.

The lower

level

strain gauges.
facilities.

Information (forces, positions,

stored in standard "frames" at each cycle of execution, and these frames can be

archived to the host computer. This archived data can then be interpreted using the "S" ([3]
[4])

analysis package for which several extensions have been written.

data

shown below were obtained

FFM

is

given

in [1].

&

The graphs

using this package.

Details of the system software architecture are given in [2].


the

that provides several features for real time processing such as

communicates with the finger motors and

FFM

also has a lower

operating system.

preemptive scheduling, multitasking, host controller communication


controller

FFM

complete description of

3.

TASK

"HOLD A KNOB AND OPEN A SIMPLE DOOR"

The Problem.

3.1.

The "door"

is

shown

in Fig. 3.

It

has a revolving knob and

placed arbitrarily in the robot workspace.

system (springs, load,

Given

The

etc.).

It is

assumed

center of rotation

is

that

is

hinged to a base and

no external forces

act

unknown.

develop an algorithm that would enable the

this situation, the task is to

on the

FFM

to

"open the door".

Solution to Task

3.2.

1.

The problem requires


a)

Grasp the door knob.

b)

Move

the center of the

To perform

task,

the arc of a circle, maintaining the grip.

each of the three fingers must have a different trajectory of


this

could be very complicated as the experimenter

would

program each finger separately, yet with appropriate high-precision coordination.

to

We
menter

this

knob along

Without force control

motion.

need

the robot to do the following

have developed and implemented an algorithm for the

to consider a

FFM

that enables the experi-

grasped body and the fingers as a single unit. This exploits the fact that

once the gripping points are fixed, both the position (and orientation) of the body and the
external forces (and torques) exerted on

it

can be obtained by a linear transformation of the

positions and forces sensed at the fingers. In this strategy the experimenter chooses a refer-

ence point on the grasped body and issues target motion and force

The algorithm automatically

calculates trajectories for finger

of this algorithm will appear in a separate report.)

commands

for this point.

movements. (Theoretical

details

Thus the independent finger motions

(which must also account for the gripping forces) become invisible to the experimenter.

Our

solution utilizes the above scheme.

the knob.

unknown
to

The problem then


axis of rotation.

be rotated about point

We

take the reference point as the center of

simplifies to that of rotating the center of the

The problem can be viewed

as follows (See fig. 4)

by the robot holding the knob

robot has no information of the radius

knob about an

at point

Line

OT

(reference point).

or the coordinates of point O.

The

However, the robot

can use force/position feedback information, and has an approximate idea of the direction

which

to

move

in

initially.

The algorithm uses an approach

similar to that described in [7].

direction tangent to the circle described by the


tion to a

is

minimum.

We

knob and keep the forces

want

to

in the

move

in the

normal direc-

4-

shows

Fig. 5

the

computed motion TQ and

of a typical experiment.

finger compliances.

is

from the center of


it

which

if

This

is

justified, since

which direction

when

dle the case

to

move

is

the robot

absorbed by the

moves perpendicu-

be moving correctly. After calculating the


fingers to set this force equal to zero.

not have any normal force; thus an approximate direction of

specified.

when humans open

Initially

motion must be

initial

doors, they do ordinarily have an idea

Moreover, the algorithm could be extended

initially.

tar-

to han-

the robot has no idea of the direction of motion, by selecting a direction in

which the force buildup

is

minimum.

Tet Results for Task

3.3.

its

Hence

rotation.

will

get direction of motion, the robot adjusts

as to in

OT

detected by the strain gauges at the fingers and the robot

lar to this force in the desired direction

we do

motion TS after one control cycle

Errors create a normal force along

This force

feels a net force pointing

the actual

1.

This algorithm has been implemented successfully on the Four Finger Manipulator using
existing software and control structure.

The modular nature

of the system

incorporate additional top-level planning routines to the system.

A new

makes

it

easy to

such "planner"

was

written for this algorithm.

Several experiments were run to study the dependence of the algorithm on various

parameters.

step size greater than 0.03

cm sometimes

causes excessive force buildup on

the fingers and sometimes causes the safety limits for motion to be exceeded.

0.02

cm was found

to

be appropriate.

Sensitivity to specified initial direction

smaller value slows

of motion

was found

to

down

step size of

motion needlessly.

the

be low, so even a crude

esti-

mate works.
Figures

6-8 give experimental data.

along the trajectory.

As soon

threshold, the robot

moves

It

shows the forces on the reference point

as the force reaches the threshold value of 1

a correction in the estimate of the tiingent.

jectory at various points.

Fig. 6

Fig. 7

the robot

makes

shows the estimated tmgent along the

tra-

can be seen from Fig. 6 that once the forces build up past the

in

such a way as to

make

corrections in the estimated tangent as

well as to reduce the total forces on the reference point.

Figures 8 a,b,c show the forces on each finger.


after each cycle to maintain the grip

The system of

three fingers adjusts

and sustain the normal forces exerted by the door. The

force sensed by each finger differs due to their different locations with respect to the door.

An

inner control loop adjusts the fingers to give the desired end result.

finger remain within safe limits of about 3 N.

N, which

is

In fact values mostly stay

the expected variation due to the force threshold value used.

The

forces

on each

between 0.5 and 1.5

TASK

4.

4.1.

"PUSH OPEN A SPRING LOADED DOOR"

The Problem.
In

some

cases a robot

example of such a

may need

situation

is

to

acts.

An

before, the center of rotation

may

push an object on which an external force

a spring loaded door.

As

be unknown.

This task can be handled by using just one finger, but here

more complex

situation in

the case

when

the door

is

which the robot needs

beyond the reach of

does not permit any of the fingers to touch


surface

is

dangerous

it

all

if

the orientation of the door

or

in the right direction,

to the robot (hot, reactive, etc.).

Use of

consider the

push the door. This can be

to use a tool to

the fingers, or

we

a tool

case

in

when

the door

can also improve the

force sustaining capability of the robot, since the total force could be distributed on

than one finger. For simplicity

4.2.

here that the

tool

a disk and the door

is

shows the schematic arrangement of the door and the

straight line segment. Fig. 9

assume also

we assume

that the friction at the contact point

more

between the tool and the door

is

is

We

tool.

negligible.

Solution to Task 2.

The algorithm
the fact that there

is

continuous tangential force on the tool. As was said before,

can be determined using the forces sensed by the fingers. To open the door
tion the target

motion of the reference point should be

However

being sensed.

we

for this task utilizes the system described above, but this time

the external forces

in a direction

exploit

this force

such a situa-

in

opposite to the force

must now be overcome by the robot.

impor-

It is

tant in this task to obtain accurate estimates of the tangent to the trajectory to avoid excessive

slippage of the tool along the door. Slip can also be reduced by using a small step size.

The

physical setup used

is

very similar to that of Task

The center of

1.

rotation

is

at

O.

At

T on

a position

TP which

is

the circumference the tangential force

opposite to

TF

is

is in

is

7"

senses.

The magnitude of

The

FFM

moves

the force

in a direction
is

then

moves

discrete

The

direction

The robot senses

the

is

the

opposite to the direction of the force that

not important as long as the fingers can sustain

door exerts a force on the tool which


its

h T P

and the target motion vector

grasps the tool using three fingers and brings

senses this force using

forces at each point and then

TF

At another point T

the desired direction of motion.

tangential force vector

a direction

it

in contact

it.

with the door.

tangential to the trajectory of the door.

it

The

The robot

force sensors, calculates the direction opposite to this force, and

amount

changes and the robot repeats

in

this

this step,

direction.

Then

the direction of the sensed force

and thus keeps moving

in the

open the door, meanwhile overcoming the tangential force exerted on

appropriate direction to
it

by the door.

Test Results for Task 2.

4.3.

This algorithm has also been implemented successfully on the

software and control structure.

A new

FFM

using the existing

top level "planner" was also written for this algo-

rithm.

Tests were again conducted to study dependence of this algorithm on internal

Again a step

ters.

size of 0.02

cm was found

to

be appropriate.

parame-

Figures 11 -13 give experi-

mental data. Fig. 11 shows the trajectory of motion of the reference point. Fig. 12 shows the
estimated tangent at various points.
tangents

lie

Due

to the small step size of 0.02

nearly along the arc of a circle.

Fig. 13

cm, the estimated

shows the forces on each

finger.

The

magnitudes are different on each finger due to varied distribution of the door spring force on
each finger.
1

5.

The

position of finger 3

was such

that

maximum

force.

Fingers

CONCLUSIONS
FFM

is

successfully able to open a "simple door" as well as a "spring loaded door"

using only force/position feedback information.


fingers

was kept

cm and

to

in

a rotation angle of

The speed of

NYU

minimum, and

The time taken

obtained.

In both the cases the force build

accurate estimates of the tangent

both the cases was about 40 seconds to

move

up on the

directions

were

a distance of about

40 degrees.

the system can readily be increased using a multiprocessor, such as the

Ultracomputer

tion of this

6.

sustained the

and 2 maintain the tool grip and sustain forces depending on their position.

The

it

[5], to distribute the

improvement

soon

is

various computational tasks involved. Implementa-

to begin.

ACKNOWLEDGEMENTS
The

FFM

project

depended

project of the

NYU

Mr. Fred Hansen

criticeilly

on extensive support provided by our colleagues

Robotics Laboratory. In particular


for providing excellent

we would

at the

like to thank:

hardware support, fixtures and mechanical

drawings.

Maw Kae Hor for providing useful suggestions and


thesis for the FFM description section of this report.

Mr.
his

Prof.

James Demmel

for providing useful suggestions and ideas through various stages

of this project.

Mr. James Fehlinger

permitting us to use parts of

for software support.

Mr Dayton

Clark

for software support.

Mr. Shidan Tavana

2.

for

REFERENCES

Maw

NYU

1)

Hor,

2)

Fehlinger, James

Kae,

dissertation (in preparation).

Experimenter's Guide to the Four Finger Manipulator, Robotics

Report #102, Courant


3)

hardware support.

Institute,

NYU,

Becker, R.A. and Chambers J.M.

S:

1986.

An

Interactive

Environment

Data Analysis

for

and Graphics. Wadsworth Advanced Books and Software, Monterey, CA, 1984.
4)

Becker, R.A. and Chambers J.M.

Extending the "S" System. Wadsworth Advanced

Books and Software, Monterey, CA, 1985.


5)

Gottlieb, Allan:

#100, Courant
6)

An Overview

Institute,

NYU

of the

NYU

Ultracomputer project, Ultracomputer Note

July 1986.

Whitney, Daniel E.: Historical Perspective and State of the Art


Proceedings of the

IEEE Conference on

in

Robot Force Control,

Robotics and Automation

St.

Louis, pp. 262-

268, 1985.
7)

Raibert,

M.H. and

tions of the

ASME,

Craig, J.J.: Hybrid Position/Force Control of Manipulators, Transac-

Vol. 102, pp. 126-133, June 1981.

CHi
o

ZL
JSl

Finger
(see fig. 2)

Linic

w~^

LHJ

Encoder

Figure

Motor

1.

Four Finger Manipulator

overviev

Connector

71

o
o
Sirajn Gage Mounting Surface

Force Sensing Finger

LJ

Figure

2.

Four Finger Manipulator

Force Sensing Finger

-5500-3.500-

.500

500
.320

.125 Dowel Pin Hole


(Typical of 6 Places)

Figure 3

Door Knob Fixture

o
R

Figure 4

0^
R

Figure 5

c
o

c
a
o
E

o
CL

O
U
c
c

u
c
o
o
o
o
t.

U3

a
u
c

Id

t/1

fe

Vi

Spring

-o-

Axis of rotation

Figure 9

A P

Figure 10

c
o
c
(0

T3

tu

D>

C
o
a.

o
o
c
c
i.
)

>^

ID

D.

D>

NYU COMPSCI TR-280

iJoor

opening" exoprin,^
using the four
ffnge^
^^"^^r
^manipulator

..

'

You might also like