Transcript 030.Deliberative-SPA - Electrical & Computer Engineering

SPA Architectures
(planning,
deliberative)
Science & Reality
–“As far as the laws of
mathematics refer to reality,
they are not certain; as far as
they are certain, they do not
refer to reality” (Einstein)
Remember this dealing with formal
models and deliberative robotics
Artificial Intelligence
One more definition:
• What is Artificial Intelligence?
– “The science of making machines do things that
would require intelligence if done by [people]”
(Minsky, 1968)
Are you sure that robot-frog should have
human-like intelligence to solve her
problems?
“Mind & Body” in AI
• Descartes:
– Mind is distinct from body
• Heidegger:
– We function in the world by simply being a part
of it
• Clarke:
– “mind, body and world act as equal partners”
Classical Artificial Intelligence
• Physical Symbol System Hypothesis
– “Formal symbol manipulation is both a
necessary and sufficient mechanism for general
intelligent behaviour” (Newell & Simon, 1957)
• Computational Representational
Understanding of Mind
– “Thinking can best be understood in terms of
representational structures in the mind and
computational procedures that operate on those
structures” (Thagard, 1996)
Classical Artificial Intelligence
• Shakey [Nilsson, 1969]:
- It failed….
- “the principle drawback of the classical
view is that explicit reasoning about the
effects of low-level actions is too
expensive to generate real-time
behavior” [Russell & Norvig, 1995]
The Old School of AI (and Robotics)
a typical robot
Processor
Sensors
Actuators
Sensors
sense - plan - act (SPA)
• Consists of 3 linear, repeated steps:
– Sense your environment
– Plan what to do next by building a world model through
sensor fusion, and taking all goals into account -- both
short term and long term
– Execute the plan through the actuators
• The predominant robot control mechanism through
1985
Called also sense - think - act (STA) or deliberative or planning
robots have many goals
A goal’s priority naturally will change based on context
I need to inspect
these railroad
spikes
I want
to take a nap
A train is about
to hit me
I just want
to be loved
I am about to
fall over
Tradition approach to slicing the
problem: SPA
• decomposition by function - classical AI
Motor control
Task execution
Planning
Modeling
Perception
Sensors 
 Actuators
All goals are known at each stage, and affect the computation
The Control Cycle: SPA
• A fundamental methodology
• Derived in the early days of robotics from engineering principles
• Sense-plan-act cycle:
– the principle is to continuously attempt to minimise the error between the actual
state and the desired state
• based on control theory
compute
(plan)
sense
think
act
The Control Cycle: SPA
world
world
perception
cognition
Discuss stages
action
modular horizontal SPA architecture
In case of soccer
robot this
architecture looks
as this:
The Control Cycle: SPA
• Agent design can be for instance like this:
– Sequential flow
– Percepts are obtained from sensors in world
(somehow)
– Get a logic-based or formal description of percepts
• E.g., wumpus world percepts
– We apply search operators or logical inference or
planning operators
• General (replaceable) formal goal
– Arrive at some operator or operator sequence
– Apply that operator sequence to world (somehow)
Path Generation
•
•
•
•
k = DOF of robot
C configuration space of robot(set of points)
O configuration space of obstacle
F = C - O free space, the set of
configurations in which the robot can move
safely
Path Generation for mobile and stationary robots
2
c1
c2
2
c1
c2
A workspace with a rotary two- The corresponding
link arm. The goal is to move
configuration space, showing
from configuration c1 to
the free space and a path that
configuration c2
achieves the goal
NAVIGATION AND MOTION PLANNING
• Given analysis of robotics problems as motion in
configuration spaces, we will begin with algorithms that
handle C language directly (no parallel instructions)
• These algorithms usually assume that an exact
description of the space is available,
– so they cannot be used where there is significant sensor error
and motion error
• We can identify five major classes of algorithms, and
arrange them roughly in order of amount of information
required at planning time and execution time
NAVIGATION AND MOTION
PLANNING: Classes of algorithms
• 1. Cell decomposition methods break continuous
space into a finite number of cells, yielding a discrete
search problem
• 2. Skeletonization methods compute a onedimensional “skeleton” of the configuration space,
yielding an equivalent graph search problem
• 3. Bounded-error planning methods assume bounds
on sensors and actuator uncertainty
NAVIGATION AND MOTION PLANNING
1. Cell decomposition method
A vertical strip cell decomposition of the configuration space
for a two-link robot. The obstacles are dark blobs, the cells are
rectangles and the solution is contained within grey rectangles.
NAVIGATION AND MOTION
PLANNING ALGORITHMS CONT.
• 4. Landmark-based navigation methods assume
that there are some regions in which the robots
location can be pinpointed using landmarks,
whereas outside those regions it may have only
orientation information
• 5. Online algorithms assume that the environment
is completely unknown initially, although most
assume some form of accurate position sensor
– Instead, one can try to produce a conditional plan or
policy that will make decisions at run time
NAVIGATION AND MOTION PLANNING
A two-dimensional environment, robot and
goal
SUMMARY ON NAVIGATION
• The problem of moving a complex-shaped object( i.e., the robot and
anything it is carrying) through a space with complex-shaped
obstacles is a difficult one.
• The mathematical notation of configuration space provides a
framework for analysis.
• Cell decomposition and skeletonization methods can be used to
navigate through the configuration space.
• Both reduce a high dimensional, continuous space to a discrete graphsearch problem.
• Some aspects of the world, such as the exact location of a bolt in the
robot’s hand, will always be unknown.
• Fine-motion planning deals with this uncertainty by creating a sensorbased plan that will work regardless of exact initial conditions.
SUMMARY ON NAVIGATION
• Uncertainty applies to sensors at the large scale as
well.
• In the landmark model, a robot uses certain wellknown landmarks in the environment to determine
where it is, even in the face of uncertainty.
• If a map of the environment is not available, then the
robot will have to plan its navigation as it goes.
• Online algorithms do this.
• They do not always choose the shortest route, but
we can analyze how far off they will be.
Problems with SPA
(sense-plan-act)
• Its monolithic design makes it slow
– At each step, we have to do:
• sensor fusion,
• world modeling,
• and planning for all goals
• Slow means we almost never can plan at the rate the
environment is changing
• We end up doing “open-loop plan execution”
– inadequate in the fact of uncertainty and unpredictability
ModelBased
Approaches
Model Based Architectures
• A symbolic internal ‘world-model’ is maintained:
– the sub-tasks are decomposed into functional layers
– similar to ‘classical’ artificial intelligence approach
sense
perception
modelling
planning
task execution
Many
levels
assess the
model
motor control
act
Problems with Models
• An adequate, accurate and up-to-date model must be
maintained at all times
– this is very difficult in practice!
– suppose, for example, the sensors detect an object that we
have not got a symbol for (a novel object)
• A model-based system is extremely brittle
– if one of the functional layers fails (e.g. hardware
problems, software bugs), then the whole system fails
• Significant processing power is required
– maintaining the model takes time, so slow responses!?
• Despite much effort, little progress was made!
Problems with traditional
approaches
•
•
•
•
•
•
•
•
Can’t account for large aspects of Intelligence,
Reliant on representation
Rapidly changing boundary conditions
Hard to map sensor values to physical quantities
Not robust
Relatively slow response
Hard to extend
Hard to test
Sources
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
Rodney Brooks
Maja Mataric
Nilsson’s book
Jeremy Elson
Norvig’s book, chapter 2. Good. Stimulus-Response Agents
English PH.D thesis, recent
Jon Garibaldi
Prof. Bruce Donald, Changxun Wu, Dartmouth College
Leo Ilkko
Prof. Manuela Veloso, Dr. Tucker Balch, and Dr. Brett Browning
Carnegie Mellon University
Rabih Neouchi , Donald C. Onyango and Stacy F. President
Axel Roth
Ramon Brena Pinero ITESM
Rhee, Taik-heon, Computer Science Department, KAIST
Brian R. Duffy, Gina Joue
Lucy Moffatt, Univ of Sheffield
Yorick Wilks, Computer Science Department, University of Sheffield