v0.1.26

What's Changed

• Fixed constraint checking in Forward and BreadthFirstSearch planners by @aemartinez in #22
• Added LMCut heuristic by @danielz9999 in #21
• Memory reductions for relaxed planning graph heuristics (HMax, HAdd, FFHeuristic, LMCut) through buffer reuse
• Used faster priority queue implementation in relaxed planning graph search
• Added AlternatingRealTimeHeuristicSearch planner
• Added MultiSolution for combined planning solutions
• Dropped support for Julia 1.3

New Contributors

• @aemartinez made their first contribution in #22
• @danielz9999 made their first contribution in #21

Full Changelog: v0.1.25...v0.1.26

v0.1.25

• Fixed bug in RTHS Djikstra update.

Full Changelog: v0.1.24...v0.1.25

v0.1.24

Planner Improvements and Changes

• Support for Fringe-Retrieving A* (FRA*) search as a solution refinement method for ForwardPlanner.
• Support for either RTAA*-style cost differencing or Dijkstra backpropagation when updating policy values in RTHS.
• RTHS can make use of FRA* search tree refinement by setting reuse_search=true.
• RTHS can be configured to run search from either all neighbors, none of them, or all unexpanded neighbors.
• The default behavior of RTHS has changed (Dijkstra-style updating instead of RTAA* updating, searching from only unexpanded neighbors instead of all neighboring states).
• RTHS now always returns a ReusableTreePolicy, which stores both a forward search tree, and a goal tree if reuse_paths is true.
• RTDP now ensures Q-value consistency during the post-rollout update.

Policy Improvements and Changes

• Add EpsilonMixturePolicy as a mixture policy over epsilon greedy policies.
• Define get_mixture_weights to return prior and posterior mixture weights for mixture policies.
• The methods has_values, get_cached_value and get_cached_action_values are now part of the PolicySolution interface.
• Add HeuristicVPolicy as a less convoluted way to use heuristics as (default) value functions.
• Implement tie-breaking among best actions in epsilon-greedy and Boltzmann policies.

Miscellaneous

• Type stability and runtime improvements when computing relaxed planning graph heuristics (HAdd, FFHeuristic, etc.)
• Pretty-printing for all specifications, solutions, planners and heuristics.
• Handle missing actions / dead-ends when simulating policies with a Simulator.

Full Changelog: v0.1.23...v0.1.24

v0.1.23

• Add interface for action filtering/pruning using heuristics via filter_available and filter_relevant
• Add PruningHeuristic to combine heuristics with another pruning method
• Improve type stability in ForwardPlanner and BackwardPlanner

Full Changelog: v0.1.22...v0.1.23

v0.1.22

• Support for path reuse in RealTimeHeuristicSearch by storing a reusable tree (as in Tree Adaptive A*)
• Documentation improvements

Full Changelog: v0.1.21...v0.1.22

v0.1.21

• Added BoltzmannMixturePolicy
• Added MixturePolicy
• Ensure that unavailable actions have zero probability under a policy
• Fix compatibility with latest PDDL.jl v0.2.17

Full Changelog: v0.1.20...v0.1.21

v0.1.20

• Fixed bug with planning graph construction for actions with no preconditions.
• Support reachability checking / fast failure in ForwardPlanner and BackwardPlanner if a heuristic returns an infinite cost.

Full Changelog: v0.1.19...v0.1.20

v0.1.19

• Fixed compatibility issue with latest versions of AutoHashEquals.jl

Full Changelog: v0.1.18...v0.1.19

v0.1.18

• Fixed bug in RTDP callback.
• Updated docs, GitHub workflows.

Full Changelog: v0.1.17...v0.1.18

v0.1.17

• Added test cases for Specifications, fixed associated bugs.
• Added MinPerAgentActionCosts and ExtraPerAgentActionCosts specifications.
• Support planning to take an action with the ActionGoal specification.

Full Changelog: v0.1.16...v0.1.17