Effect

The add list specifies the set of logical propositions that become true in the new state after the action's execution. It defines the positive changes the action introduces.

• Example: For an action PickUp(BlockA), the add list might include Holding(BlockA) and ¬HandEmpty().
• In the STRIPS formalism, the add list is a core component for modeling state transitions.

The delete list (or del list) specifies the set of logical propositions that become false in the new state after the action's execution. It defines what the action removes or negates.

• Example: For PickUp(BlockA), the delete list would include HandEmpty() and OnTable(BlockA) if it was previously on the table.
• Managing the delete list is central to solving the frame problem, as it explicitly declares what changes, implicitly stating everything else remains unchanged.

A conditional effect is an effect that only occurs if a specific condition holds at the time of the action's execution. It allows for more expressive action models where outcomes depend on context.

• Structure: When(Condition, Effect).
• Example: An action TurnOn(Generator) might have the conditional effect: When(FuelLevel > 0, Running(Generator)). If the fuel level is zero, the Running proposition is not added.
• Conditional effects are supported in extensions of STRIPS, such as PDDL.

In stochastic planning models like MDPs and POMDPs, actions have probabilistic effects. Instead of a single deterministic outcome, the action leads to one of several possible next states, each with an associated probability.

• Example: An action Navigate(Dock) might have a 90% probability of reaching the dock (effect: At(Dock)) and a 10% probability of getting lost (effect: At(UnknownLocation)).
• These probabilities are defined in the transition function T(s, a, s').

Numeric effects modify continuous, real-valued fluents (numeric state variables) rather than Boolean propositions. They are essential for modeling resources, time, and other metric quantities.

• Operators: Increase, decrease, assign, scale.
• Example: An action Drive(Truck, CityA, CityB) might have the numeric effect decrease(Fuel, 50) and increase(TotalCost, 200).
• Numeric effects are a key feature of PDDL 2.1 and are critical for temporal planning with durations.

In temporal planning, effects can be durative, meaning they occur at specific time points during a lengthy action's execution, not just at its end. This allows for modeling processes where state changes at the start, during, or at the conclusion of an action.

• Key Points: at start, at end, over all.
• Example: A durative action Charge(Battery) might have an at start effect of Charging(Battery) = true and an at end effect of BatteryLevel = 100.
• This requires sophisticated planners that reason about time, such as those supporting PDDL 2.2.

A precondition is a logical condition that must be true in the current state for an action to be legally applicable. It defines the permission to act. While an effect describes the change an action makes, the precondition describes the state the world must be in for the action to start.

• Example: For the action Stack(A, B), a precondition might be Clear(B) ∧ Holding(A).
• In formalisms like STRIPS, preconditions and effects together fully specify an action's semantics.

The state space is the set of all possible configurations or situations the world can be in, defined by the truth values of all relevant propositions. An effect operates within this space, causing a transition from one state to a successor state.

• Each state is a snapshot (e.g., {On(A, Table), Clear(B)}).
• The transition function, defined by actions and their effects, maps a state and an action to a new state.
• Planning algorithms search this space for a path (a plan) from the initial state to a goal state.

STRIPS (Stanford Research Institute Problem Solver) is the foundational formalism that introduced the standard representation of effects. An action in STRIPS is defined by:

• Precondition List: Propositions that must be true.
• Add List: Propositions the action makes true (positive effects).
• Delete List: Propositions the action makes false (negative effects).

This explicit representation directly tackles the frame problem by specifying only what changes, implicitly asserting everything else remains unchanged.

The frame problem is the challenge of efficiently representing and reasoning about which aspects of the world remain unchanged when an action is performed. The STRIPS representation of effects provides a classic solution:

• By explicitly listing add and delete effects, it implies a frame axiom: any proposition not mentioned is assumed to persist.
• This avoids the need for a massive set of axioms stating what does not change for every action.
• Modern planning languages like PDDL inherit this solution.

The Planning Domain Definition Language (PDDL) uses STRIPS-like effects within a more expressive, first-order logic framework. Effects are defined in action schemas that use variables.

• Example Schema: Action(Move(?x, ?from, ?to))
  • Precondition: At(?x, ?from) ∧ Clear(?to)
  • Effect: At(?x, ?to) ∧ not(At(?x, ?from))
• During instantiation, variables (?x, ?from, ?to) are bound to objects, creating ground actions with specific add/delete lists.
• PDDL also supports conditional effects and numeric effects.

Plan validation is the process of verifying that a proposed sequence of actions, when executed from the initial state, will achieve the goal. This is done by simulating the effects of each action in order.

• The validator starts with the initial state.
• For each action, it checks preconditions are met.
• It then applies the action's effects (adding and deleting propositions) to compute the successor state.
• After the final action, it checks if the goal state conditions are satisfied.
• Tools like VAL are dedicated plan validators for PDDL.