COMP790-072
Robotics: An Introduction

Sensors & Actuators

Introduction to Kinematics
UNC Chapel Hill
M. C. Lin
Sensors

Vision (Review)
– Stereoscopic
– Monoscopic

Sonar (see a later lecture)

Others (bump sensors, LIDAR, etc.)
UNC Chapel Hill
M. C. Lin
Cameras

Charge coupled devices (CCD’s) use arrays of
photosensitive diodes to generate intensity maps
– grey-levels of color devices are available
– a range of image resolutions (pixels per image)
• 800 × 600 pixels is typical
– a range of frame rates (number of images per second)
• 30 Hz (frames per second) is typical

The field of view can be changed
– high-resolution cameras typically view 45 - 60°
– wide-angle (fisheye) lenses may cover 80 - 90°
– curved mirrors increase field further without distortion
UNC Chapel Hill
M. C. Lin
Stereoscopic Vision

Viewing the world with two cameras (eyes)
allows a 3D representation to be formed
– unfortunately the signal is complex and noisy

Each camera receives a slightly different view
– the distance between corresponding points in an
image is known as the stereo disparity
disparity
UNC Chapel Hill
M. C. Lin
Stereo Ranging

The amount of disparity is related to distance
– the difficulty lies in identifying corresponding points

The general principle is
–
–
–
–
left and right images are digitized
raw images are rectified for distortion / misalignment
rectified images are filtered to enhance textures+edges
a stereo matching algorithm is applied
• modern techniques search along horizontal scan lines to find the
best set of matching pixels (e.g. mean-squared-error)
– raw disparity map is filtered to remove noise

This can now be done on modern computers
– e.g. Pentium P-4 @ GHz at interactive frame rates
UNC Chapel Hill
M. C. Lin
Monoscopic Vision

Although stereo vision is popular, it has
problems
– high hardware requirements, camera alignment, etc.
– consequently single camera input may be used also

Monoscopic ranging
– optical flow
• the relative motion between the moving camera and viewed
objects in the environment, seen over a sequence of images
– looming
• as an object gets close, it gets bigger!
• is simple to use this information to calculate distance
– but the object must have been identified and must be totally in view
– depth from focus
• depth-of-field of conventional lens systems can be used
UNC Chapel Hill
M. C. Lin
Object Recognition

Much vision research on object recognition
– so easy for humans, but the problem not yet solved
– humans may use a combination of techniques and
reasoning

Edge detection
– fairly simple filter operations can detect clean edges
• e.g. the discrete Laplace filter
– reliable detection of all edges is much more difficult

Area based techniques
– connected regions of similar color, texture or
brightness probably belong to the same object
UNC Chapel Hill
M. C. Lin
Actuators
Locomotion
 Manipulation

UNC Chapel Hill
M. C. Lin
Actuators
Locomotion
 Manipulation

UNC Chapel Hill
M. C. Lin
Locomotion
Legs
 Wheels
 Other exotic means

UNC Chapel Hill
M. C. Lin
Legs

Two legs seems the most obvious configuration
– but in fact balance is an incredibly difficult problem
• e.g. the Honda Humanoid Project
– need knees, ankles and hips in order to move around
– two legs are inherently unstable: difficult to stand still

Six legs are much easier to balance and move
– stable when not moving
– can work with simple cams and rigid legs
– Brooks et al. (1989) evolved the walking Genghis robot
UNC Chapel Hill
M. C. Lin
Wheels

Any number of wheels is possible
– there are many different configurations that are useful

Two individually driven wheels on either side
– usually with one or more idler wheels for balance
– independently driven wheels allows zero turning radius
• one wheel drives forwards, one wheel drives backwards

Rear wheel drive, with front wheel steering
– the vehicle will have a non-zero turning radius
– for two front wheels, turning geometry is complex
– rear wheels need a differential to prevent slippage

4WD is possible, but it is even more complex
UNC Chapel Hill
M. C. Lin
Exotic Wheels & Tracks

Tracks can be used in the same way as two
wheels
– good for rough terrain (as compared to wheels)
– tracks must slip to enable turns (skid steering)

In synchro drive, 3+ wheels are coupled
– drive in same direction at same rate
– pivot in unison about their respective steering axes
– allows body of robot to remain in the same orientation

Tri-star wheels are composed of 3 sub-wheels
– entire wheel assembly rolls over a large obstacle

Many other exotic wheel configurations
– Multiple-degrees-of-freedom (MDOF):
– going side way, tight turns, etc.
UNC Chapel Hill
M. C. Lin
Mobility Considerations
A number of issues impact selection of drive
 Maneuverability - ability to alter direction/speed
 Controllability - practical and not too complex
traction sufficient to minimize slippage
 climbing ability - traversal of minor
discontinuities, slope rate, surface type, terrain
 stability - must not fall over!
 efficiency - power consumption reasonable
 maintenance - easy to maintain, reliable
 environmental impact - does not do damage
 navigation - accuracy of dead-reckoning
UNC Chapel Hill
M. C. Lin
Actuators
Locomotion
 Manipulation

UNC Chapel Hill
M. C. Lin
Manipulations

Degrees of freedom
– independently controllable components of motion

Arms
– convenient method to allow full movement in 3D
– more often used in fixed robots due to power & weight
– even more difficult to control!
• due to extra degrees of freedom

Grippers
– may be very simple (two rigid arms) to pick up objects
– may be complex device with fingers on end of an arm
– probably need feedback to control grip force
UNC Chapel Hill
M. C. Lin
Manipulation Actuator Types

Electric
– DC motor is the most common type used in mobile robots
– stepper motors turn a certain amount / applied voltage

Pneumatic
– operate by pumping compressed air through chambers

Hydraulic
– pump pressurized oil: usually too heavy, dirty and expensive to
be used on mobile robots

Shape memory alloys (SMA’s)
– metallic alloys that deform under heat and then return to their
previous shape: used for artificial muscles
• see http://www.sma-inc.com/SMAPaper.html
UNC Chapel Hill
M. C. Lin
Measuring Motion: Odometers

If wheels are being used, then distance
traveled can be calculated by measuring
number of turns
– dead-reckoning or odometry is the name given to
the direct measure of distance (for navigation)

Motor speed and timing are very inaccurate
– measuring the number of wheel rotations is better
– shaft encoders, or rotation sensors, measure this
– Different types & technologies of shaft encoder
UNC Chapel Hill
M. C. Lin
Motion Types

holonomic: the controllable degrees of
freedom is equal to the total degrees of
freedom, e.g. manipulator arm

non-holonomic: the controllable degrees
of freedom is less than the total degrees
of freedom, e.g. car (although it can
move laterally, but no mechanism to
control lateral movement)
UNC Chapel Hill
M. C. Lin
Introduction to Kinematics

Kinematics: study of motion independent of
underlying forces

Degrees of freedom (DoF): the number of
independent position variables needed to
specify motions

State Vector: vector space of all possible
configurations of an articulated figure. In
general, the dimensions of state vector is
equal to the DoF of the articulated figure.
UNC Chapel Hill
M. C. Lin
Manipulator Joint Types
1 DOF Joint types
 Revolute

Prismatic
UNC Chapel Hill
M. C. Lin
More Joint Types

Many higher order joint types can be
represented by combining 1-DOF
joints by making axes intersect
UNC Chapel Hill
M. C. Lin
Forward vs. Inverse Kinematics

Forward kinematics: motion of all joints is
explicitly specified

Inverse kinematics: given the position of
the end effector, find the position and
orientation of all joints in a hierarchy of
linkages; also called “goal-directed motion”.
See handouts for a simple 2D example.
UNC Chapel Hill
M. C. Lin