RoadmapEUROPA_X

Items to be incorporated in future EUROPA releases

Make PSEngine runtime configuration (what modules to be instanciated, wiring between modules, etc) completely data-driven. This seems to be possible only through the use of dynamic libraries. Find alternatives for static libs.
Streamline and cleanup PSEngine interface. make sure all extension points and event notifications are reachable through it.
allow EUROPA extensions (constraints, propagators, decision points, etc.) to be written in other languages (java/python)
PSEngine
Systematic review
PSVarValue is a very simplified view of a Domain, should be improved
Decide what to do with !DbClient interface, it is redundant
Improve PSList
Resources
Measure performance of constrained-based resources vs token-based ones (migrate UBO benchmarks). Drop token-based approach?
Implement more efficient alternatives to Incremental Flow profile? (see #217). Measure performance of Grounded profile. Make sure default profile is fast.
Implement fast profile for Unaries so that they're competitive with Timelines

Incorporate domain-independent heuristics developed by Sara Bernardini and Dave Smith
Capture required metadata and expose through Schema API
Add new solver that incorporates heuristics
Make sure built-in solver functionality can be reused as a building block for custom solvers
it works in the presence of constraint violations (until it finds the next one, or ideally, until it breaks a violation threshold, that way upper bounds can be used)
choice of next decision and next value can be easily customized from Java or other languages that can use EUROPA through SWIG.
Re-factor Solver, make it easier to extend built-in solver, write solver from scratch or by composition of solvers.
Add explicit support for relaxations?
Add Dave & Sara's work on domain independent heuristics
Add min-perturbation heuristic solver from Dynamic Europa
incorporate and expose classical approaches from CP (variable ordering, back-jumping, etc).
research :
Support for automatic interleaving of multiple solvers
Machine learning applications to improve search (see Mark Roberts paper on benchmark problems should be useful here).
Mixing classical planning and CP,LS,Math programming search approaches

NDDL
provide support for soft constraints (specify penalty values) and violation explanation
provide procedural elements for :
creating problem instances through iteration, branching, etc.
loading data from a database
creating/extending/running solvers
dumping Plan Database/ Schema information
provide API to obtain structural information (predicate relationships, grounded instances, etc) about the model
better AST support
ANML
Move ANML parser to antlr3, reconnect ANML module to build
Merge with latest ANML grammar written by Will Cushing and Dave Smith. Involve Will?
provide full ANML support
provide PDDL support : #106
provide POCL planner for PDDL/ANML

Improve Solver Dialog : #41
!PlanWorks has in effect been deprecated. Come up with a plan to migrate/re-implement any functionality that we maybe missing in PSDesktop tools or Tatiana's new UI tools