Skip to content

Latest commit

 

History

History
99 lines (61 loc) · 2.77 KB

engines.rst

File metadata and controls

99 lines (61 loc) · 2.77 KB
Raw

Planning engines

LPG

Support Features:

  • classical and numeric state variables.
  • durative actions.
  • Supports the Oneshot Planner in both quality and runtime modes.

Fast-Downward

Support Features:

  • Classical planning support with full information, non-numeric, deterministic instantaneous actions.
  • Supports Oneshot and Anytime Planner engine in both quality and runtime modes.

Advantages:

  • Integration within the unified_planning library facilitated by the AIPlan4EU project, making it accessible for broader use.
  • Recognized as one of the most successful systems for classical planning.

Disadvantages:

  • Specificity to classical planning scenarios may limit its applicability to other types of planning problems.

EnhSP

Support Features:

  • Boolean and numeric state variables, actions, processes, and events (PDDL+ language).
  • Supports Oneshot Planner in both quality and runtime modes.

Advantages:

  • Handles disjunctive preconditions and conditional effects without expensive compilations.

Disadvantages:

  • Optimality only assured for specific PDDL+ fragments (simple numeric planning problems)​

symk

Support Features:

  • State-of-the-art classical optimal and top-k planner based on symbolic search extending Fast Downward. It can find a single optimal plan or a set of k different best plans with the lowest cost for a given planning task.
  • Supported in the Oneshot and Anytime Planners in both quality and runtime modes.

tamer

Tamer is a temporal planner that supports temporal action-based problems.

Support Features:

  • Supported in the Oneshot Planner in the runtime mode.

pyperplan

Support Features:

  • Classical planning based on different search heuristics.
  • Action-based problems with hierarchical typing.
  • Supports the STRIPS PDDL fragment without action costs.
  • Supported in the Oneshot Planner in the runtime mode.

Advantages:

  • Lightweight and written in Python.

Disadvantages:

  • Does not support action costs in the STRIPS PDDL fragment, which could limit its usefulness in certain planning scenarios.
  • The default planning algorithm is a blind breadth-first search, which does not scale well, although other heuristic search algorithms are available.
  • Limited to specific PDDL fragments, which may not cater to more complex or varied planning needs.

fmap

Support Features:

  • Distributed heuristic search.
  • Forward partial-order planning scheme allowing parallel action planning among agents.
  • State-based estimates utilizing frontier state.
  • Supported in the Oneshot Planner in the runtime mode.

Advantages:

  • Enhanced solution plan quality due to parallel action planning.
  • Accurate state-based estimates for improved planning.