AI and Robotics

flybittencobwebΤεχνίτη Νοημοσύνη και Ρομποτική

2 Νοε 2013 (πριν από 3 χρόνια και 10 μήνες)

77 εμφανίσεις

2008-05-07
Baran Cürüklü 1
AI and Robotics
• What is a robot?
– Different types of robots.
– Autonomy
– What do you find on an autonomous robot?
• Planning and Robotics
– STRIPS
– Conjunctive goals
– Horisontal architecture ↔Subsumption architecture
• The Future: Social robots.
What does the word``Robot” mean?
• Robot = Forced labor
– Robota = Drudgery (tungt arbete)
– Biologically Engineered
• no mechanical parts!
– Coined by Chech writers Josef and Karel
Č
apek
– http://www.uwec.edu/academic/curric/jerzdg/RUR/
Three laws of robotics by Asimov.
A robot...
• may not injure a human being, or through
inaction, allow a human being to come to harm.
• must obey orders given it by human beings,
expect where that would conflict with the first
law.
• must protect its own existence expect where
that would conflict with the first or second law.
Different types of Robots
• Indoor, outdoor.
• Water, air, ground (wheels, tracks or leggs).
• Degrees of autonomy:
– Shared (industrial robot)
– Semi (human intervenes occasionally)
– Full (can operate without human intervention)
Indoor vs Outdoor:
Water, air, ground
Autonomy depends on the need:
• Shared:Built a car in a factory.
– All the cars must be exactly the same.
• Semi:Sending robot to Mars.
– 30 min delay for the radio signals Earth→Mars.
• Full:Vacuumclean a house.
– Avoid the cat, etc.
2008-05-07
Baran Cürüklü 2
What do you find on an autonomous
robot?
• Sensing (Sense)
• Computing (Plan)
• Motion (Act)
Material covered until know is mainly under Plan
Sensing
• Encoders on the wheel axes (to measure
traveled distance)
• B/W or color camera to ``see” objects and
calculate distances to them.
• Sonars/lasers to measure distance to objects.
• Artificial noses (and other specialized sensors)
– find the land mine or truffle
– in wine and cheese production
Computing using
• Ordinary processors.
• Signal processors.
• Transputers.
…it depens on the task and the budget.
Vision may require a fast processor.
Motion
• Wheels, tracks, vings or leggs for motion
• Arms for manipulating the objects.
Planning and Robotics
• What is planning? A form of search?
• Different techniques/theories.
• Planning and robotics.
What is planning? A form of search?
• Search:Bfs, Dfs are based on state-space.
– A state specifies all the relevant as well as
irrelevant facts.
– Operators(state_in, state_out) apply to a fully
specified state.
• Memory consumtion is huge.
2008-05-07
Baran Cürüklü 3
Planning is more efficient
compared to Search
• Planning is search in the formula space
– State: at(baran,home) ∧ ¬has(baran,bananas) Operators:
buy(bananas) results in has(baran,bananas)
• Planning...
– Does not case about irrelevant facts.
• Like at(baran,home) above
– Interested in the path from the start to goal.
Formal definition of planning
Planning involves choosing a plan by
considering alternative plans and reasoning
about their consequences.
STRIPS -
STanford Research Institute Problem Solver
Example: Block Stacking Problem
• Action: move(X,Y,Z)
– move block X from top of Y to top of Z
• Precondition: on(X,Y) ∧ clearTop(Z) ∧ clearTop(X)
• Effects
Additions
: on(X,Z) ∧ clearTop(Y)
• Effects
Deletions
: ¬on(X,Y) ∧ ¬clearTop(Z)
A
B
C
move(X,Y,Z)
• Action: move(X,Y,Z)
– X = B, Y = A, Z = C
• Precondition: on(B,A) ∧ clearTop(C) ∧ clearTop(B)
• Effects
Additions
: on(B,C) ∧ clearTop(A)
• Effects
Deletions
: ¬on(B,A) ∧ ¬clearTop(C)
A
B
C
C
A
B
STRIPS={objects, initial_facts, operators}
– objects = {A, B, C, table}
– initial_ facts = {on(A,table), free(C), ...}
– goal = {on(B,C), free(A), on(A,B), on(C,table)}
– operators = {oper1(pre,add,del), oper2(pre,add,del), ...}
•``pre” is a list of precondition (facts).
•``add” is a list of facts to be added.
•``del” is a list of facts to be deleted.
C
B
A
The STRIPS Assumption
• World is deterministic
• No exogenous actions
– the world does not change (evolve).
• all changes are described by the operators.
2008-05-07
Baran Cürüklü 4
Conjunctive goals
• Goal = on(B,C) ∧ free(A) ∧ on(A,B)

on(C,table)
– the order of achievements of sub goals does not
matter.
– Not true in many real life problems.
Sussman Anomaly
• Achive first: on(A, B)
– Can we achive on(B,C) ?
• or achive first: on(B, C)
– Can we achive on(A,B) ?
⇒Conclusion
– Order of achievements of sub goals does matter.
C
B
A
A
C
B
B
C
A
C
B
A
Initial state
goal
Partially Ordered Plan
Wake up
Finish
Left sock
Left shoe
Right shoe
Right sock
Planning and Robotics (some
history)
• Planning first! (1969 - 85)
– Focus on generating plans
– STRIPS formalism
– Assume ideal conditions
• But the world is not perfect.
• Action first! (1985 - 99)
– Rodney Brooks at MIT
• http://www.ai.mit.edu/people/brooks/
– Focus on execution of plans in dynamic enviroments.
• Somebody must makes plans.
Brooks’ Outrage (1985)
• The horisontal architecture (see belov) is a
composition of many steps and it takes too
long to generate an action.
taskexecution
modeling
planning
perception
motorcontrol
sensors actuators
Brooks’ alternative to the horisotal
architecture
• The vertical subsumption architecture.
– Higher leves have higher priority.
– Short way from sensors to action.
sensors actuators
reason about behaviors
avoid objects
wander around
built maps of the envoriment
identify objects
2008-05-07
Baran Cürüklü 5
The vertical subsumption architecture
• Everything is behavior
• No need to model the enviroment.
– Remember we are interested in action not in planning.
• Behaviors with higher priorities inhibits whose
with lower priorities.
– But how to decide the priorities between behaviors?
Subsumption example
sensors
behavior 1
behavior 3
behavior 2
actuators
s s
inhibition
behavior 1:wander around
behavior 2:catch the mouse
behavior 3: eat the mouse
The Future: Social robots.
Cog (Brooks)
Coco (Brooks)
Kısmet (Brooks)
Kısmet’s design is…
…more like the cortex of a primate then a simple planning architecture
.