Seminaarin aikataulu

1
Seminaarin aikataulu
2
Planning An introduction

• Seppo Törmä
• Helsinki University of Technology

3
Content

• Meaning of planning
• Intentional action, control and planning
• Primary uses of a plan
• Different plan structures
• Some planning techniques
• classical planning
• hierarchical planning
• case-based planning

4
The need for planning

• Examples
• All kinds of projects
• Business strategy
• Behavior of a robots and software agents
• Manufacturing processes
• Factory production
• All intentional (goal-directed) behavior where
  the goal cannot be approached in a monotonic
  manner

5
The meaning of planning

• 1. Creation of the activity network
• 2. Decision making about future action

6
Subtasks

• Project management
• 1. Identify activities --gt set of activities
• 2. Sequence activities --gt precedence
  constraints
• 3. Schedule activities --gt execution times
• 4. Allocate resources --gt resource demands
• In general, not a fixed sequence
• eg, often the resource availability has to be
  determined first to define the correct activities
• different types of decision are interleaved

7
Dynamic environment

• Problem
• every plan makes assumptions about the world
• knowledge is incomplete and world is often
  unpredictable
• assumptions fail and plans become obsolete
• Solutions
• make fewer assumptions
• plan less
• make plans less dependent on volatile conditions
• revise the plan when assumptions fail
• incrementally modify the existing plan
• create a new plan

8
Questions

• What is the purpose of planning?
• What kinds of plans are needed?
• When are plans needed?
• How detailed plans are needed?

9
Intentional action

• Agent that tries to achieve a goal hasgoal
  intentions
• commitment to achieve the goals
• Agent is involved in intentional action
• Intentional action needs control
• take care of the type and direction of effort
• control involves planning, since there is a need
  to make decisions about the future action (even
  if only the immediate activities)

10
Control of intentional action
11
General strategies

• Execute, observe and plan cycle
• flexible few assumptions about the world
• observations can be expensive
• opportunities may be lost
• problems with dead ends
• Plan in advance
• inflexibility more assumptions about the world
• how much in advance are the plans needed?
• how detailed plans are needed?

12
Plans and organizations

• Alternative mechanisms to coordinate intentional
  action
• planning avoid conflicts in advance
• organizations solve recognized conflicts
• Practical combinations
• plans at a high-level
• execution control at a detail level

13
Main uses of a plan

• 1. to provide operational guidance
• agents participating in the execution need to
  know what to do next
• 2. to estimate the performance of the action
• eg, duration, costs, resource needs,
• need to answer the questions
• What alternative course of action to choose?
• What alternative goals to choose?

14
Plan as operational guidance

• Plan as a set of implementation intentions
• committed situation-action -rules
• in situation y I intend to perform action z
• Trigger situations need to be observable
• include the necessary observation activities to
  the plan
• Consistent plan is a set of rules so that
• there are no unobservable trigger-conditions
• when the implementation intentions are carried
  out, goals are achieved

15
Different plan structures

• Time-related plans at 1.12.97 start A
• project management
• Sequential plans first A, then B, then C
• workflows, computer programs
• Queue plans enqueue(A, Q), dequeue(Q)
• different priority dispatch rules
• resource management (eg, operating systems)
• Event-related plans when conflict, call boss
• how to guarantee a desirable global behavior?

16
Plan structures and environment

• Durations uncertain
• sequential plans
• Resource bottleneck
• queue plans
• Multiple resources
• time-related plans
• Unknown dynamics of the environment
• event-related plans

17
AI planning research

• Domain-independent planning techniques
• the essential aspects of the planning problem
• algorithms are computationally complex
• Much effort on generative techniques
• how to build a plan from scratch
• Sequential plans
• Simplifying assumptions
• static environment, no resource contention,

18
Classical planning

• Planning problem
• C, G, A, where
• C current state (a set of conditions)
• G goal state (a set of conditions)
• A a set of activities each represented with
• preconditions
• effects (added and deleted conditions)
• result is a sequence of activities from A

19
Classical planners

• Planning method
• establish goal conditions one by one with
  activities that produce the conditions
• resolve conflicts between activities
• Execution is a sequence of activities
• Assumptions
• static environment
• perfect information about the world

20
Classical planning

• Example traveling from Espoo to NY

A3 drive JFK-NY - pre loc(x)JFK - del
loc(x)JFK - add loc(x)NY
loc(x)JFK
A1 drive Espoo-HEL - pre loc(x)Espoo -
del loc(x)Espoo - add loc(x)HEL
G
A2 fly HEL-JFK - pre loc(x)HEL - del
loc(x)HEL - add loc(x)JFK
C
loc(A)NY
loc(x)Espoo
loc(x)HEL
21
Hierarchical planners

• Planning problem
• T, R where
• T the task to perform
• R a set of task reduction schemes
• eg, build a house
• Reductions can be motivated by the
• structure
• function

22
Hierarchical planners

• Method
• expand an activity into a network of simpler
  activities until all activities are primitive
  (directly executable)
• Better control over the planning process
• in practice (even if not in theory) more
  efficient
• Many practically oriented systems
• extensions such as handling of time and
  resources

23
Case-based planning

• Method
• retrieve a similar past problem
• good match
• good performance
• adapt the plan of the particularities of the
  current problem
• record the performance
• store the plan as a new case

24
Case-based planning

• Advantages
• reuse of the work
• resembles human planners
• learning (both plans and failures)
• Problems
• creating the initial plans
• indexing what makes problems similar
• adaptation methods