Scikit-decide

# hub.domain.pddl.heuristics

Domain specification

Domain

# atom_to_str

atom_to_str(
  atom,
  task
)

Format a FlatAtomKey as a human-readable string using task metadata.

Args: atom: A PDDL_FlatAtomKey object with predicate_id and args. task: The PDDL Task providing predicate_name() and object_name().

Returns: A string like "on(block1, block2)".

# HMax

h_max delete-relaxation heuristic (admissible).

Takes the MAX over goal atom costs. Lower bound on optimal plan cost. Pre-grounds all reachable actions at construction time via clingo, then computes heuristic values via fast array-based forward chaining.

For deterministic domains (PDDL): computes h_max as defined in "Planning as Heuristic Search" (Bonet & Geffner, Artificial Intelligence, 2001).

For probabilistic domains (PPDDL):

  • discount_factor=1.0 (SSP): applies Proposition 1 from "Extending Classical Planning Heuristics to Probabilistic Planning with Dead-Ends" (Teichteil-Königsbuch, Vidal & Infantes, AAAI 2011) — h⁺_max on the implicit all-outcome determinization.
  • discount_factor<1.0 (DSSP): applies Theorem 2 — discounted formula h^γ_max(s) = c_m(s) · (1 − γ^{h^{1,+}_max(s)}) / (1 − γ).

# Constructor HMax

HMax(
  task,
  discount_factor = 1.0,
  dead_end_cost = 1000000000.0,
  verbose = False
)

Initialize self. See help(type(self)) for accurate signature.

# compute_detailed HMax

compute_detailed(
  self,
  state
)

Return detailed heuristic information for the given state.

Returns a dict with:

  • heuristic_value: the h_max value
  • atom_costs: list of (FlatAtomKey, cost) for reachable atoms
  • goal_atom_costs: list of (FlatAtomKey, cost) for goal atoms

# HAdd

h_add delete-relaxation heuristic (non-admissible, informative).

SUM over goal atom costs. Assumes subgoals are independent. Pre-grounds all reachable actions at construction time via clingo, then computes heuristic values via fast array-based forward chaining.

For deterministic domains (PDDL): computes h_add as defined in "Planning as Heuristic Search" (Bonet & Geffner, Artificial Intelligence, 2001).

For probabilistic domains (PPDDL):

  • discount_factor=1.0 (SSP): applies Proposition 1 from "Extending Classical Planning Heuristics to Probabilistic Planning with Dead-Ends" (Teichteil-Königsbuch, Vidal & Infantes, AAAI 2011) — h⁺_add on the implicit all-outcome determinization.
  • discount_factor<1.0 (DSSP): applies Theorem 2 — discounted formula h^γ_add(s) = c_m(s) · (1 − γ^{h^{1,+}_add(s)}) / (1 − γ).

# Constructor HAdd

HAdd(
  task,
  discount_factor = 1.0,
  dead_end_cost = 1000000000.0,
  verbose = False
)

Initialize self. See help(type(self)) for accurate signature.

# compute_detailed HAdd

compute_detailed(
  self,
  state
)

Return detailed heuristic information for the given state.

Returns a dict with:

  • heuristic_value: the h_add value
  • atom_costs: list of (FlatAtomKey, cost) for reachable atoms
  • goal_atom_costs: list of (FlatAtomKey, cost) for goal atoms

# HFF

h_FF delete-relaxation heuristic.

Builds a relaxed planning graph via h_add forward chaining, then extracts a relaxed plan backwards from the goal as described in:

Hoffmann, J. and Nebel, B. (2001). The FF Planning System:
Fast Plan Generation Through Heuristic Search.
Journal of Artificial Intelligence Research, 14, 253-302.

h_FF = total cost of unique actions in the relaxed plan. Also identifies helpful actions (relaxed-plan actions whose preconditions are all satisfied in the current state).

For probabilistic domains (PPDDL):

  • discount_factor=1.0 (SSP): h⁺_FF on all-outcome determinization.
  • discount_factor<1.0 (DSSP): h^γ_FF via Theorem 2 from "Extending Classical Planning Heuristics to Probabilistic Planning with Dead-Ends" (Teichteil-Königsbuch, Vidal & Infantes, AAAI 2011).

# Constructor HFF

HFF(
  task,
  discount_factor = 1.0,
  dead_end_cost = 1000000000.0,
  verbose = False
)

Initialize self. See help(type(self)) for accurate signature.

# compute_detailed HFF

compute_detailed(
  self,
  state
)

Return detailed heuristic information for the given state.

Returns a dict with:

  • heuristic_value: the h_FF value
  • atom_costs: list of (FlatAtomKey, cost) for reachable atoms
  • goal_atom_costs: list of (FlatAtomKey, cost) for goal atoms
  • relaxed_plan_actions: list of GroundAction in the relaxed plan
  • helpful_actions: list of GroundAction (relaxed-plan actions whose preconditions are all satisfied in the current state)
  • marked_atoms: list of FlatAtomKey extracted during backward pass

# compute_with_helpful HFF

compute_with_helpful(
  self,
  state
)

Return (h_value, [GroundAction]) for the given state.

# helpful_actions HFF

helpful_actions(
  self,
  state
)

Return helpful actions (list of GroundAction) for the given state.