Fast Downward is a classical planning system based on heuristic search, capable of solving general deterministic planning problems encoded in the propositional fragment of PDDL2.2, including advanced features like ADL conditions and effects and derived predicates. Unlike other PDDL planners, it does not use the propositional PDDL representation directly but instead translates the input into an alternative representation called multivalued planning tasks, which makes implicit constraints explicit. This allows Fast Downward to use hierarchical decompositions for computing its causal graph heuristic, which differs from traditional heuristics that ignore negative interactions. The system solves multi-valued planning tasks by extending the causal graph heuristic to tasks involving axioms and conditional effects. It also employs novel search control techniques, such as preferred operators, deferred evaluation of heuristic functions, and multi-heuristic best-first search. Efficient data structures for state expansion and a new non-heuristic search algorithm, focused iterative-broadening search, are also introduced. Fast Downward has proven successful, winning the “classical” track of the 4th International Planning Competition at ICAPS 2004. It performs well on earlier benchmarks and provides insights into the usefulness of new search enhancements. The system is based on three components: translation, knowledge compilation, and search. The translation component converts PDDL2.2 tasks into multi-valued planning tasks. The knowledge compilation component generates data structures for efficient access, including domain transition graphs and causal graphs. The search component implements various search algorithms, including the causal graph heuristic and focused iterative-broadening search. Multi-valued planning tasks (MPTs) are formalized with state variables, initial and goal states, axioms, and operators. The causal graph heuristic uses hierarchical decomposition to compute heuristic estimates, while domain transition graphs encode how state variables can change. The system's approach is rooted in previous work on causal graphs and domain transition graphs, but it is the first to apply hierarchical decomposition with multi-valued state variables in a general planning framework. Fast Downward also applies techniques from earlier work on abstraction hierarchies and causal graphs, adapting them to heuristic search. The system's success lies in its ability to efficiently solve complex planning tasks by leveraging hierarchical structures and multi-valued state variables.Fast Downward is a classical planning system based on heuristic search, capable of solving general deterministic planning problems encoded in the propositional fragment of PDDL2.2, including advanced features like ADL conditions and effects and derived predicates. Unlike other PDDL planners, it does not use the propositional PDDL representation directly but instead translates the input into an alternative representation called multivalued planning tasks, which makes implicit constraints explicit. This allows Fast Downward to use hierarchical decompositions for computing its causal graph heuristic, which differs from traditional heuristics that ignore negative interactions. The system solves multi-valued planning tasks by extending the causal graph heuristic to tasks involving axioms and conditional effects. It also employs novel search control techniques, such as preferred operators, deferred evaluation of heuristic functions, and multi-heuristic best-first search. Efficient data structures for state expansion and a new non-heuristic search algorithm, focused iterative-broadening search, are also introduced. Fast Downward has proven successful, winning the “classical” track of the 4th International Planning Competition at ICAPS 2004. It performs well on earlier benchmarks and provides insights into the usefulness of new search enhancements. The system is based on three components: translation, knowledge compilation, and search. The translation component converts PDDL2.2 tasks into multi-valued planning tasks. The knowledge compilation component generates data structures for efficient access, including domain transition graphs and causal graphs. The search component implements various search algorithms, including the causal graph heuristic and focused iterative-broadening search. Multi-valued planning tasks (MPTs) are formalized with state variables, initial and goal states, axioms, and operators. The causal graph heuristic uses hierarchical decomposition to compute heuristic estimates, while domain transition graphs encode how state variables can change. The system's approach is rooted in previous work on causal graphs and domain transition graphs, but it is the first to apply hierarchical decomposition with multi-valued state variables in a general planning framework. Fast Downward also applies techniques from earlier work on abstraction hierarchies and causal graphs, adapting them to heuristic search. The system's success lies in its ability to efficiently solve complex planning tasks by leveraging hierarchical structures and multi-valued state variables.