Brain Online

PDDL and STRIPS

📝 Zuletzt geändert: (heute) 🆕📊 689 Wörter

PDDL and STRIPS

chatbot
Here is a structured translation of the provided information about STRIPS and PDDL:

STRIPS (Stanford Research Institute Problem Solver) and PDDL (Planning Domain Definition Language) are important concepts in state-based planning. STRIPS is a foundational system upon which PDDL builds.

STRIPS:

  • Foundation for State-Based Planning: STRIPS serves as the basis for modern state-based planners.
  • State Representation: States are represented as sets of ground atoms, meaning the world is described by a collection of properties or facts that hold at a specific time.
  • Operators: STRIPS uses operators defined by three lists:
    • PRE (Preconditions): A list of atoms that must be true in the current state for the operator to be applied.
    • ADD (Positive Effects): A list of atoms added to the state after the operator is applied.
    • DEL (Negative Effects): A list of atoms removed from the state after the operator is applied.
  • Application of Operators: Applying an operator changes the current state by removing elements from the DEL list and adding elements from the ADD list.
  • Closed-World Assumption: STRIPS employs the Closed-World Assumption, meaning that any atoms not explicitly stated as true are assumed to be false.
  • Planning as Search: Planning in STRIPS can be viewed as a classical search problem where the search space is defined by the states and the application of operators.
  • Solving the Frame Problem: STRIPS solves the frame problem implicitly by leaving everything unchanged that is not explicitly modified via the ADD and DEL lists. This means planners do not need to explicitly define what remains unchanged, simplifying the representation.
  • Planning Algorithms: Various algorithms are used for planning in STRIPS, such as forward and backward planning, and heuristics to make the search more efficient. These include the Ignore-DEL and Ignore-Pre heuristics.
  • Example: Blocksworld is a frequently used example to illustrate STRIPS planning. Operators like put(X,Y) (put block X on block Y) have preconditions that must be met (e.g., X and Y are free) and effects that change the state (e.g., X is now on Y).
  • Limitations: STRIPS may not always handle the qualification and ramification problems well. It allows for simple representation, but the representation of preconditions and effects is limited.

PDDL (Planning Domain Definition Language):

  • Extension of STRIPS: PDDL is a language that significantly extends STRIPS, offering additional features.
  • Development by IPC: The development of PDDL has been driven by the International Planning Competition (IPC) since 1998.
  • Numeric Fluents: PDDL supports numeric fluents, allowing numerical values to be considered in the planning model, e.g., the number of items in a PSU Contains(P, I, 100).
  • Derived Predicates: PDDL allows for definitions of derived predicates, based on other predicates, e.g., PSUAt(P, X, Y) :- Carries(R, P), RobotAt(R, X, Y).
  • LISP-like Syntax: PDDL syntax is LISP-like and uses a prefix notation, e.g., (f x y z) instead of f(x,y,z).
  • Declarative Language: PDDL is a declarative language representing knowledge.
  • Domain and Problem Definition: A PDDL representation consists of:
    • A domain definition (types, actions).
    • A problem definition (initial state, goal).
  • Compatibility: PDDL can be used with various search and planning algorithms.
  • Examples: PDDL is used in various scenarios, such as Blocksworld and Magic World.

Comparison:

  • STRIPS is a simpler approach to state-based planning, while PDDL is a more expressive language that builds on STRIPS.
  • STRIPS solves the frame problem with the Closed-World Assumption and ADD/DEL lists, whereas PDDL extends these basic mechanisms.
  • PDDL offers the capability to use numeric fluents and derived predicates, allowing for more complex planning tasks.

In summary, STRIPS forms the foundation for state-based planning, while PDDL is a flexible and extended language for defining planning domains and problems. Both approaches are significant in AI planning, with PDDL being more commonly used in complex scenarios due to its extended features.

See also

Related atomic notes

Course material

Status:
Tags: science methods-of-ai
Superlink: 611 📠Machine Learning
610 🤖Artificial Intelligence, Künstliche Intelligenz

Quellen

Erstellt: 14-02-25 21:27