time Last we saw:
DC motors
inefficiencies, operating voltage and current, stall voltage and current and torque current and work of a motor
Gearing gear ratios
gearing up and down combining gears
Pulse width modulation Servo motors
Lecture Outline
What are sensors? Types of sensors (many examples) Sensor complexity Signals -> symbols Levels of processing Poor and good design of perception Biological perception and lessons Sensor fusion
Why is Robotics hard?
Sensors are limited and crude Effectors are limited and crude State (internal and external, but mostly external) is partiallyobservable Environment is dynamic (changing over time) Environment is full of potentiallyuseful information
What are sensors?
Sensors constitute the perceptual system of a robot Sensors do not provide state Sensors are physical devices that measure physical quantities Examples:
Physical property -> sensor: contact -> switch distance -> ultrasound, radar, infra red
Examples of sensors
More examples:
Physical property -> sensor: light level -> photo cells, cameras sound level -> microphones strain -> strain gauges rotation -> encoders magnetism -> compasses smell -> chemical temperature -> thermal, infra red
More examples of sensors
Even more examples:
Physical property -> sensor: inclination -> inclinometers rate of change of inclination -> gyroscopes pressure -> pressure gauges altitude -> altimeters and many more...
Note: the same property can be measured with different sensors
Types of Sensors
Sensors range from simple to complex in the amount of information they provide
simple: an on/off switch (1 bit of input) complex: the human retina (> 100 million
photosensitive elements!)
A sensor provides "raw" information,
which usually needs to be processed
Sensor Complexity
The output of a simple sensor can be used directly, without processing
(e.g., if switch closed, stop, else go)
The output of a complex sensor
must be processed We can ask: "Given the sensory reading I am getting, what was the world like to make the sensor give me this reading?" => reconstruction
Signals -> Symbols (State)
Sensors do not provide state/symbols, just signals A great deal of computation may be required to convert the signal from a sensor into useful state for the robot This process bridges the areas of electronics, signal processing, and computation
Levels of Processing
to find out if a switch is open or closed,
we need to measure voltage going through the circuit => electronics
using a microphone to separate voice from noise and recognize => signal processing using a surveillance camera, find people
in the image and recognize criminals, perhaps by comparing them to a large database => computation
Requirements
The more processing that needs to be done, the more computation is required Thus perception requires:
sensors (power and electronics) computation (more power and electronics) connectors (to connect it all)
Poor Designs of Perception
It is not a good idea to separate
what the robot senses how it senses it how it processes it and how it uses it
If these are separated, the resulting robot design is typically large, bulky, and ineffective.
History of Poor Designs
Historically, perception has been treated poorly:
perception in isolation perception "king" as perception as reconstruction
Traditionally these approaches came from computer vision, which provides the most complex data
Good Designs of Perception
Instead, it is best to think about these as a single complete design:
the task the robot has to perform the best sensors for that task the best mechanical design that will allow the robot to get the necessary sensory information to perform that task
(e.g., the body shape of the robot, the placement of the sensors, etc.)
A New & Better Way
Perception in the context of action and the task Action-oriented perception Expectation-based perception use
knowledge about the world as constraints on sensor interpretation
Focus-of-attention methods provide
constraints on where to look
Perceptual classes partition the world
into useful categories
Biological Perception
Nature solves this problem cleverly: it
evolves special sensors with special geometric and mechanical properties. Consider facetted eyes of flies, polarized light sensors on birds, horizon/line sensors on bugs, the shape of the human ear, etc. Biological sensors use clever mechanical designs that maximize the sensor's properties, i.e., its range and correctness.
Proprioception
Origin of received sensory information divides perception into
Proprioception: sensing internal state (e.g., muscle tension, limb position) Exteroception: sensing external state (e.g., vision, audition, smell, etc.) Examples of proprioception
path integration (dead-reckoning) balancing all movement...
Affordances
Affordances are "potentialities for action inherent in an object or scene" (Gibson 1979, psychology) The focus is the interaction between the robot and its environment Perception is biased by what needs to be done (the task) E.g.: a chair can be something to sit in, avoid, throw, etc.
Lessons from Biology
As a robot designer, you may not get the chance to make up new sensors But you will always have the chance (and the need) to design interesting ways of using the available sensors Utilize the interaction with the world and always keep in mind the task Food for thought: how would you detect people in an environment?
Example: detecting people
temperature: pyro-electric sensors detect special temperature ranges movement: if everything else is static or slower/faster color: if people wear uniquely colored clothing in your environment shape: now you need to do complex vision processing
Example: measuring distance
ultrasound sensors (sonar) give you
distance directly (time of flight) infra red provides return signal intensity two cameras (i.e., stereo) can give you distance/depth use perspective projection with 1 camera use a laser and a camera, triangulate use structured light; overlying grid patterns on the world ...
Sensor Fusion
A powerful strategy is to combine different sensors => Sensor Fusion Sensor fusion is complex because sensors have:
different characteristics different accuracy different complexity
Computation is necessary to combine them effectively (in real-time)
Biological Sensor Fusion
The brain processes information from many sensors (vision, touch, smell, hearing, sound) The processing areas are distinct in the brain (and for vision, they are further subdivided into the "what" and "where" pathways) Much complex processing is involved in combining the information