THE SENSE OF TOUCH
Like the human hand, robotic grippers often need a sense of touch to determine if and
when they have something in their grasp. Knowing when to close the gripper to take
hold of an object is only part of the story, however. The amount of pressure exerted on the
object is also important. Too little pressure and the object may slip out of grasp; too much
pressure and the object may be damaged.
The human hand has an immense network of complex nerve endings that serve to
sense touch and pressure. Touch sensors in a robot gripper are much cruder with just one
or two points, but for most hobby applications these sensors serve their purpose: to provide
nominal feedback on the presence of an object and the pressure exerted on the
object.
This chapter deals with the fundamental design approaches for several touch sensing
systems for use on robot grippers (or should the robot lack hands, elsewhere on the body of
the robot). Modify these systems as necessary to match the specific gripper design being
used and the control electronics you are using to monitor the sense of touch.
Note that in this chapter the distinction between touch and collision is made. Touch is a
proactive event, where you specifically wish the robot to determine its environment by making
physical contact. Conversely, collision is a reactive event, where (in most cases) you wish
the robot to stop what it’s doing when a collision is detected and back away from the condition.
Chapter 30, “Object Detection,” deals with the (potential for) physical contact resulting
in a collision.
The lowly mechanical switch is the most common, and simple, form of tactile (touch) feedback.
Most any momentary, spring-loaded switch will do; the switch changes state when an
object is touched, completing a circuit (or in some cases, the switch opens, breaking the circuit).
The switch may be directly connected to a motor or discrete circuit, or it may be connected
to a computer or microcontroller, as shown in Fig. 29-1.
You can choose from a wide variety of switch styles when designing contact switches for
tactile feedback. Leaf switches (sometimes referred to as Microswitch switches) come with
levers of different lengths that enhance the sensitivity of the switch. You can also use miniature
contact switches, like those used in keyboards and electronic devices, as touch sensors
on your robot.
In all cases, mount the switch so it makes contact with whatever object you wish to
sense. In the case of a robotic gripper, you can mount the switch in the hand or finger sections.
In the case of feelers for a smaller handless robot, the switch can be mounted fore or aft.
It makes contact with an object as it rolls along the ground. By changing the arrangement
of the switch from vertical (see Fig. 29-2), you can have the feeler determine if it’s reached
the edge of a table or a stair landing.
Figure 29-2 By orienting the switch to the vertical, your robot can detect
changes in topography, such as when it’s about to run off the edge of a table.
One of the most convenient components that you can use for sensing objects around a
robot or in its grippers is the Microswitch (Fig. 29-3). The name comes from one of the first
companies to manufacture switches that can be actuated by external mechanical pressure.
Microswitches come in a variety of different shapes and sizes (many more than an inch long,
which seems contradictory for something with micro in its name).
Figure 29-3 Microswitch with a long actuator arm built into it. This type of Microswitch is ideally suited for use as a Robot’s “whiskers,” sensing when the robot collides with another object.
When you start out, you might want to consider a Microswitch with a long metal actuator
like the one shown in Fig. 29-3 and have a double throw internal switch (three connections—
a common, a normally open, and a normally closed). The actuator can be extended
by gluing on a plastic tube (or straw) or soldering on a metal extension. The double throw
switch will allow you to use the Microswitch as normally open or normally closed or as part
of a debounce circuit that will be discussed later in the chapter. These features will give you
a maximum amount of operational flexibility when adding a Microswitch to your robot.
If you were to read through a catalog of different switches, you would be amazed at the
number of options available in terms of number of positions, number of separate switches
inside the package (called poles), mounting types, and so on. Despite the plethora of different
options, there is one feature that is common to all switches, but is not listed in any
data sheet, and that is switch bounce—the making and breaking of contacts when a switch
changes state. The oscilloscope picture in Fig. 29-4 shows an oscillating signal that takes
place when a pulled up momentary on button is pressed.
Figure 29-4 Oscilloscope picture of the signal produced when a pulled up momentary on
switch is pressed. The spikes in the signal indicate that the switches internal contacts are literally
bouncing off one another until they make solid contact.
These oscillations are caused by the contacts within the switch literally bouncing off one
another when the switch is closed. Going by Fig. 29-4, you can see that this bouncing takes place over a few hundred microseconds—much too fast to be detected by the naked eye.
But this bouncing is a problem for computer systems and other electronic devices. Without
being properly debounced, each spike may be recognized as an individual event and instead
of the robot responding as if the switch changed state once, it will respond to many switch
changes, which will potentially make your robot behave erratically.
It is important to note that switch bouncing also takes place when contacts are pulled
apart from each other as well as when they come together. You will have to ensure that
debouncing takes place for both cases to ensure that no unexpected switch transitions affect
the way you expect your robot to operate.
There are some simple hardware approaches to debouncing switch inputs. These methods
eliminate the need for software to be developed to process the incoming signals as will be discussed
in the next section. Unfortunately, these methods increase the robot’s circuitry, so
while there is some savings in software complexity, there will be an increase in the overall
robot circuitry, which can be difficult to add in a small robot or one that has already been built.
The first method to debounce a circuit uses a double throw switch and a couple of CMOS
Schmitt input inverters as shown in Fig. 29-5 (with its parts list in Table 29-1). The two
inverters form a flip-flop that only changes state when the switch definitely changes state.
Figure 29-5 Two Schmitt triggered inverters along with a couple of other components can be
used to make an effective button debounce circuit.
74LS14 |
Hex Schmitt trigger input inverter chip |
10k |
10k resistor |
Switch |
Double throw, single pole switch |
When the common contact is not touching either connector (and is disconnected from Vcc
and ground), the flip-flop circuitry remembers the last common contact position. In the circuit
given in Fig. 29-5, note that the inverters are Schmitt-triggered CMOS gates—the
Schmitt trigger helps ensure that the flip-flop only changes state at a definite voltage level.
The CMOS gates along with the 10k current limiting resistor ensures that there is not a
large current flow or “backdriving” of an inverter’s output.
There are other hardware debounce circuits that you can use in your application as well.
One of the most popular is using a 555 set up as a monostable multivibrator with enough
of a delay to ensure that any bounces are masked by the 555’s output. The circuit shown
in Fig. 29-5 is very efficient, both in terms of operation as well as cost; you will be hard
pressed to find a circuit that is as simple and effective as this one.
Debouncing button inputs in software can be tricky; the need for performing other operations
in an application means that the debounce code must be integrated with the complete
application. This requirement adds a level of complexity to the code that can be difficult to
implement for many new programmers. While debouncing button inputs in software is
more cost efficient than adding hardware, you may find that the amount of work to get
debounce code working in a complete robot application to be much more effort than adding
a couple of CMOS Schmitt trigger inverters and a resistor.
The rule of thumb for debouncing a button input is that the line is considered to be
debounced and the input stable when it has not changed state for at least 20 ms. The code
for debouncing a single button in a stand-alone application is actually quite easy:
This code increments a counter each time in a loop that the button is down. If the button
is not down then the counter is reset. The counter value is continually checked to see if
it has incremented to a value equal to or greater than 20 ms and when it is, the execution
of the program leaves the “while” loop and returns to the subroutine’s caller.
The problem with debugging a button in software comes when there are additional functions
built into the robot. When this happens the time the main robot loop code executes
must be known (and becomes the incrementing value for the button counter) and the main
loop above becomes distributed. If the robot execution loop was assumed to execute in
1 ms, then the basic code could look like:
Note that the button “state” variables are limited to a maximum value of 20. This is done
to ensure that the variables do not overflow and become zero, which indicates the buttons
are not pressed.
While the code in the listing seems complex, you should remember that multiple buttons
could be added to the robot application by simply copying this code for each additional button
(new counter variables will be required as well). You will discover that the execution time
penalty of the button debounce software is negligible, and using this code, you should be
able to add a number of buttons to your robot application relatively easily.
Optical sensors use a narrow beam of light to detect when an object is within the grasping
area of a gripper. Optical sensors provide the most rudimentary form of touch sensitivity
and are often used with other touch sensors, such as mechanical switches.
Building an optical sensor into a gripper is easy. Mount an infrared LED in one finger or
pincher; mount an infrared-sensitive phototransistor in another finger or pincher (see Fig.
29-6). Where you place the LED and transistor along the length of the finger or pincher
determines the grasping area.
Figure 29-6 An infrared LED and phototransistor pair can be
added to the fingers of a gripper to provide go/no-go grasp information.
Mounting the infrared pair on the tips of the fingers or pinchers provides little grasping
area because the robot is told that an object is within range when only a small portion of it
can be grasped. In most gripper designs, two or more LEDs and phototransistors are placed
along the length of the grippers or fingers to provide more positive control. Alternatively,
you may wish to detect when an object is closest to the palm of the gripper. You’d mount
the LED and phototransistor accordingly.
Fig. 29-7 shows the schematic diagram for a single LED-transistor pair. Adjust the value
of R2 to increase or decrease the sensitivity of the phototransistor. You may need to place
an infrared filter over the phototransistor to prevent it from triggering as a result of ambient
light sources (some phototransistors have the filter built into them already). Use an LED-transistor
pair equipped with a lens to provide additional rejection of ambient light and to
increase sensitivity.
Figure 29-7 The basic electronic circuit for an infrared touch
sensor. Note the infrared filter; it helps prevent the phototransistor
from being activated by ambient light.
During normal operation, the transistor is on because it is receiving light from the LED.
When an object breaks the light path, the transistor switches off. A control circuit connected
to the conditioned transistor output detects the change and closes the gripper. In a practical
application, using a computer as a controller, you’d write a short software program to
control the actuation of the gripper.
An optical sensor is a go/no-go device that can detect only the presence of an object, not
the amount of pressure on it. A pressure sensor detects the force exerted by the gripper on
the object. The sensor is connected to a converter circuit, or in some cases a servo circuit,
to control the amount of pressure applied to the object.
Pressure sensors are best used on grippers where you have incremental control over the
position of the fingers or pinchers. A pressure sensor would be of little value when used with
a gripper that’s actuated by a solenoid. The solenoid is either pulled in or it isn’t; there are
no in-between states. Grippers actuated by motors are the best choices when you must regulate
the amount of pressure exerted on the object.
You can make your own pressure sensor (or transducer) out of a piece of discarded conductive
foam—the stuff used to package CMOS ICs. The foam is like a resistor. Attach two
pieces of wire to either end of a 1-in square hunk and you get a resistance reading on your
volt-ohm meter. Press down on the foam, and the resistance lowers.
The foam comes in many thicknesses and densities. Look for semi-stiff foam that regains
its shape quickly after it’s squeezed. Very dense foams are not useful because they don’t
quickly spring back to shape. Save the foam from the various ICs you buy and test other
types until you find the right stuff for you.
You can make a pressure sensor by sandwiching several pieces of material together, as
shown in Fig. 29-8. The conductive foam is placed between two thin sheets of copper or
aluminum foil. A short piece of 30 AWG wire-wrapping wire is lightly soldered onto the foil (when using aluminum foil, the wire is wound around one end). Mylar plastic, like the kind
used to make heavy-duty garbage bags, is glued on the outside of the sensor to provide electrical
insulation. If the sensor is small and the sense of touch does not need to be too great,
you can encase the foam and foil in heat-shrink tubing. There are many sizes and thicknesses
of tubing; experiment with a few types until you find one that meets your requirements.
Figure 29-8 Construction detail for a pressure sensor using conductive foam. The leads are soldered
or attached to foil (copper works best). Choose a foam that has a good "spring" to it.
The resistance of the conductive foam pressure transducers changes abruptly when they
are compressed. The output may not return to its original resistance value (see Fig. 29-9).
So in the control software, you should always reset the transducer just prior to grasping an
object.
Figure 29-9 The response curve for the conductive foam pressure sensor.
Note that the resistance varies depending on whether the foam is being pressed
or released.
For example, the transducer may first register an output of 30KΩ (the exact value
depends on the foam, the dimensions of the piece, and the distance between wire terminals).
The software reads this value and uses it as the set point for a normal (nongrasping)
level to 30K. When an object is grasped, the output drops to 5K. The difference—25K—is
the amount of pressure. Keep in mind that the resistance value is relative, and you must
experiment to find out how much pressure is represented by each 1K of resistance change.
The transducer may not go back to 30K when the object is released. It may spring up to
40K or go only as far as 25K. The software uses this new value as the new set point for the
next occasion when the gripper grasps an object.
Obviously, the home-built pressure sensors described so far leave a lot to be desired in terms
of accuracy. If you need greater accuracy, you should consider commercially available strain
gauges. These work by registering the amount of strain (the same as pressure) exerted on
various points along the surface of the gauge.
Strain gauges are somewhat pricey—about $10 and over in quantities of 5 or 10. The
cost may be offset by the increased accuracy the gauges offer. You want a gauge that’s as
small as possible, and preferably one mounted on a flexible membrane. See Appendix B,
“Sources,” for a list of companies offering such gauges.
The output of both the homemade conductive foam pressure transducers and the strain
gauges is analog—a resistance or voltage. Neither can be directly used by a computer, so
the output of these devices must be converted into digital form first.
Both types of sensors are perfect for use with an analog-to-digital converter. You can use
one ADC0808 chip (under $5) with up to eight sensors. You select which sensor output you
want to convert into digital form. The converted output of the ADC0808 chip is an eight bit
word, which can be fed directly to a microprocessor or computer port. Fig. 29-10a shows the basic wiring diagram for the ADC0808 chip, which can be used with conductive
foam transducer; Fig. 29-10 b shows how to connect a conductive foam transducer to one
of the analog inputs of the ADC0808.
Figure 29-10 a. The basic wiring diagram for converting pressure data into digital data, using
an ADC0808 analog-to-digital converter (ADC) IC. You can connect up to eight pressure sensors to
the one chip. b. How to interface a conductive foam sensor to the ADC0808 chip.
Notice the 10K resistor in Fig. 29-10, placed in series between the pressure sensor and
ground. This converts the output of the sensor from resistance to voltage. You can change
the value of this resistor to alter the sensitivity of the circuit. For more information on
ADCs, see Chapter 14, "Computer Peripherals."
A new form of electricity was discovered just a little more than a century ago when the two
scientists Pierre and Jacques Curie placed a weight on a certain crystal. The strain on the
crystal produced an odd form of electricity—significant amounts of it, in fact. The Curie
brothers coined this new electricity piezoelectricity; piezo is derived from the Greek word
meaning press. Later, the Curies discovered that the piezoelectric crystals used in their
experiments underwent a physical transformation when voltage was applied to them. They
also found that the piezoelectric phenomenon is a two-way street. Press the crystals and out
comes a voltage; apply a voltage to the crystals and they respond by flexing and contracting.
All piezoelectric materials share a common molecular structure, in which all the movable
electric dipoles (positive and negative ions) are oriented in one specific direction. Piezoelectricity
occurs naturally in crystals that are highly symmetrical—quartz, Rochelle salt crystals,
and tourmaline, for example. The alignment of electric dipoles in a crystal structure is similar
to the alignment of magnetic dipoles in a magnetic material.
When an electrical voltage is applied to the piezoelectric material, the physical distances
between the molecular dipoles change. This causes the material to contract in one dimension
(or axis) and expand in the other. Conversely, placing the piezoelectric material under
pressure (in a vise, for example) compresses the dipoles in one more axis. This causes the
material to produce a voltage.
While natural crystals were the first piezoelectric materials used, synthetic materials have
been developed that greatly demonstrate the piezo effect. A common human-made piezoelectric
material is ferroelectric zirconium titanate ceramic, which is often found in piezo
buzzers used in smoke alarms, wristwatches, and security systems. The zirconium titanate is
evenly deposited on a metal disc. Electrical signals, applied through wires bonded to the surfaces
of the disc and ceramic, cause the piezo material to vibrate at high frequencies (usually
4 kHz and above).
Piezo activity is not confined to brittle ceramics. PVDF, or polyvinylidene fluoride (used
to make high-temperature PVDF plastic water pipes), is a semicrystalline polymer that lends
itself to unusual piezoelectric applications. The plastic is pressed into thin, clear sheets and
is given precise piezo properties during manufacture by—among other things—stretching
the sheets and exposing them to intense electrical fields.
PVDF piezo film is currently used in many commercial products, including noninductive
guitar pickups, microphones, even solid-state fans for computers and other electrical
equipment. One PVDF film you can obtain and experiment with is Kynar, available directly
from the manufacturer (see Measurement Specialists at www.msiusa.com for more information).
Whether you are experimenting with ceramic or flexible PVDF film, it’s important to
understand a few basic concepts about piezoelectric materials:
- Piezoelectric materials are voltage sensitive. The higher the voltage is, the more the piezoelectric material changes. Apply 1 V to a ceramic disc and crystal movement will be slight. Apply 100 V and the movement will be much greater.
- Piezoelectric materials act as capacitors. Piezo materials develop and retain an electrical charge.
- Piezoelectric materials are bipolar. Apply a positive voltage and the material expands in one axis. Apply a negative voltage and the material contracts in that axis.
The ubiquitous ceramic disc is perhaps the easiest form of piezoelectric transducer to experiment
with. A sample disc is shown in Fig. 29-11. The disc is made of nonferrous metal,
and the ceramic-based piezo material is applied to one side. Most discs available for purchase
have two leads already attached. The black lead is the ground of the disc and is
directly attached to the metal itself.
Figure 29-11 Piezo ceramic discs are ideally suited to be contact and pressure sensors for robotics.
You can use a ceramic disc as an audio transducer by applying an audio signal to it. Most
piezo discs will emit sound in the 1K to 10K region, with a resonant frequency of between
3K and 4K. At this resonant frequency, the output of the disc will be at its highest.
When the piezo material of the disc is under pressure—even a slight amount—the disc
outputs a voltage proportional to the amount of pressure. This voltage is short lived: shortly
after the initial change in pressure, the voltage output of the disc will return to 0. A negative
voltage is created when the pressure is released (see the discussion of the bipolar nature
of piezo materials earlier in the chapter).
You can easily interface piezo discs to a computer or microcontroller, either with or
without an analog-to-digital converter. Chapter 30, “Object Detection,” discusses several
interface approaches. See the section “Piezo Disc Touch Bar” in that chapter for more
information.
Samples of Kynar piezoelectric film are available in a variety of shapes and sizes. The
wafers, which are about the same thickness as the paper in this book, have two connection
points, as illustrated in Fig. 29-12. Like ceramic discs, these two connection points are used
to activate the film with an electrical signal or to relay pressure on the film as an electrical
impulse.
You can perform basic experiments with the film using just an oscilloscope (preferred) or
a high-impedance digital voltmeter. Connect the leads of the scope or meter to the tabs on
the end of the film (the connection will be sloppy; there are some suggestions for other ways
to attach leads to Kynar film in the next section). Place the film on a table and tap on it.
You’ll see a fast voltage spike on the scope or an instantaneous rise in voltage on the meter.
If the meter isn’t auto-ranging and you are using the meter at a low setting, chances are that
the voltage spike will exceed the selected range.
Unlike piezoelectric ceramic discs, Kynar film doesn’t usually come with pre-attached
leads (although you can order samples with leads attached, but they are expensive). There
are a variety of ways to attach leads to Kynar film. Obviously, soldering the leads onto the
film contact areas is out of the question. Acceptable methods include applying conductive
ink or paint, self-adhesive copper-foil tape, small metal hardware, and even miniature rivets.
In all instances, use small-gauge wire—22 AWG or smaller. You can expect the best
results using 28 AWG and 30 AWG solid wire-wrapping wire. The following are the best
methods:
- Conductive ink or paint. Conductive ink, such as GC Electronics’ Nickel-Print paint, bonds thin wire leads directly to the contact points on Kynar film. Apply a small globule of paint to the contact point, and then slide the end of the wire in place. Wait several minutes for the paint to set before handling. Apply a strip of electrical tape to provide physical strength.
- Self-adhesive copper-foil tape. You can use copper-foil tape designed for repairing printed circuit boards to attach wires to Kynar film. The tape uses a conductive adhesive and can be applied quickly and simply. As with conductive inks and paints, apply a strip of electrical tape to the joint to give it physical strength.
- Metal hardware. Use small 2/56 or 4/40 nuts, washers, and bolts (available at hobby stores) to mechanically attach leads to the Kynar. Poke a small hole in the film, slip the bolt through, add the washer, and wrap the end of a wire around the bolt. Tighten with the nut.
- Miniature rivets. Homemade jewelry often uses miniature brass or stainless steel rivets. You can obtain the rivets and the proper riveting tool from many hobby and jewelry-making stores. To use them, pierce the film to make a small hole, wrap the end of the wire around the rivet post, and squeeze the riveting tool (you may need to use metal washers to keep the wire in place).
Fig. 29-13 shows a simple demonstrator circuit you can build that indicates each time a
piece of Kynar film is struck. Tapping the film produces a voltage output, which is visually indicated when the LED flashes. The 4066 IC is an analog switch. When a voltage is
applied to pin 3, the connection between pins 1 and 2 is completed and that finishes the
electrical circuit to light the LED. For a robotic application, you can connect the output to
a computer or microcontroller.
You can easily create a workable touch sensor by attaching one or two small Kynar transducers
to a thick piece of plastic. The finished prototype sensor is depicted in Fig. 29-14.
The plastic membrane could be mounted on the front of a robot, to detect touch contact,
or even in the palm of the robot’s hand. Any flexing of the membrane causes a voltage
change at the output of one or both Kynar film pieces.
The human body has many kinds of touch receptors embedded within the skin. Some
receptors are sensitive to physical pressure, while others are sensitive to heat. You may wish
to endow your robot with some additional touch sensors like:
- Heat sensors can detect changes in the heat of objects within grasp. Heat sensors are available in many forms, including thermisters (resistors that change their value depending on temperature) and solid-state diodes that are specifically made to be ultrasensitive to changes in temperature. Chapter 34, “Fire Detection Systems,” discusses using solid-state temperature sensors.
- Air pressure sensors can be used to detect physical contact. The sensor is connected to a flexible tube or bladder (like a balloon); pressure on the tube or bladder causes air to push into or out of the sensor, thereby triggering it. To be useful, the sensor should be sensitive to increases in air pressure to about 1 lb/in2, or less.
- Resistive bend sensors, originally designed for use with virtual reality gloves, vary their resistance depending on the degree of bending. Mount the sensor in a loop, and you can detect the change in resistance as the loop is deformed by the pressure of contact.
- Microphones and other sound transducers make effective touch sensors. You can use microphones, either standard or ultrasonic, to detect sounds that occur when objects touch. Mount the microphone element on the palm of the gripper or directly on one of the fingers or pinchers. Place a small piece of felt directly under the element, and cement it in place using a glue that sets hard. Run the leads of the microphone to the sound trigger circuit, which should be placed as close to the element as possible.
To learn more about . . . |
Read |
|
Designing and building robot hands |
Chapter 28, “Experimenting with Gripper
Designs” |
|
Connecting sensors to computers and microcontrollers |
Chapter 14, “Computer Peripherals” |
|
Collision detection systems |
Chapter 30, “Object Detection” |
|
Building light sensors |
Chapter 32, “Robot Vision” |
|
Fire, heat, and smoke detection for robotics |
Chapter 34, “Fire Detection Systems” |