PDDLica.jl

PDDLica.jl parses the component-oriented PDDLica language into the versioned PDDLica-JSON surface IR, elaborates component assemblies, and simulates supplied plans with discrete and durative actions, timed initial literals, urgent events, continuous processes, PDDL3 validation constraints, and linear acausal connector equations.

The simulator validates supplied plans independently of optimization. The executable profile includes typed finite quantification, conditional and quantified effects, derived predicates, concurrent happenings, durative conditions/effects, trajectory constraints, preferences, and metric reporting. Preferences and metrics are evaluated by the simulator and may be used by an optimizer when its transcription supports them. Acausal interfaces remain restricted to affine systems with a unique active solution.

The default optimizer unrolls a bounded horizon into a time-indexed hybrid MILP using JuMP and solves it with HiGHS. Binary variables select grounded actions and component methods; state variables and constraints encode discrete frame conditions, affine process dynamics, component requirements, and linear acausal connector equations. The recovered plan is always replayed through the independent simulator:

result = optimize(checked.model;
    max_steps=10,
    max_makespan=10.0,
    max_simultaneous_actions=2)

result.status       # :FEASIBLE, :NOT_FOUND, :TIME_LIMIT, ...
result.plan         # ordinary PlanDocument when feasible
result.validation   # mandatory simulator replay
result.objective    # incumbent, bound, and MILP termination information

An OPTIMAL solver termination means optimal for the configured finite grid, bounds, and supported transcription—not for every horizon or the unrestricted PDDLica problem. NOT_FOUND likewise means the configured unrolling is infeasible, not that the unrestricted problem is.

The MILP profile includes component lifecycle, presence-dependent connection regimes, timed initial literals, bounded urgent-event closure, durative start/end and continuous effects, conditional/quantified effects, derived predicates, optional singleton potential memory, and PDDL3 trajectory constraints, preferences, and metrics. Autonomous guards and process dynamics are sampled on the configured fixed grid. Affine expressions, finite grounding, explicit numeric bounds, and a bounded event-closure depth remain required. Every recovered plan is replayed with the event-locating simulator, so a grid witness that does not induce the same continuous execution is returned as INVALID_WITNESS.

The earlier simulator-driven enumerator remains available as a semantic fallback when fixed-grid MILP transcription is unsuitable:

result = optimize(checked.model;
    backend=NativeSearchBackend(),
    options=OptimizationOptions(max_macrosteps=4, max_makespan=10.0))

using PDDLica

parsed = parse_model("domain.plca", "problem.plca")
checked = elaborate(parsed.document)
plan = read_plan_json("plan.json")
result = simulate(checked.model, plan)

write_execution_json("execution.json", result)
restored = read_execution_json("execution.json")

Inspect and plot recorded time series directly in the Julia REPL:

available_variables(result)
series = trajectory(result, "destination.volume")
series = trajectory(result, "destination", "volume")

plot_trajectory(result, "destination.volume")
plot_trajectory(result, "destination", "volume")
plot_trajectories(result, ["source.volume", "destination.volume"])

Plots use UnicodePlots and therefore render in the terminal. Continuous values are sampled every 0.1 model-time units by default, while action and event boundaries retain duplicate pre/post timestamps. Configure the cadence with SimulationOptions(trajectory_interval=0.01), use nothing to retain only semantic boundaries, or set save_everystep=true to retain solver steps.

Run the reference example:

julia --project=. bin/pddlica simulate \
  examples/pumped_fluid/domain.plca \
  examples/pumped_fluid/problem.plca \
  examples/pumped_fluid/plan.json

Optimize a plan from source with the default JuMP/HiGHS backend:

julia --project=. bin/pddlica optimize \
  examples/pumped_fluid/domain.plca \
  examples/pumped_fluid/problem.plca \
  --backend milp --max-steps 5 --makespan 5 --output optimization.json

Inspect a parsed model or report backend restrictions without starting a solve:

julia --project=. bin/pddlica check model.pddlica.json
julia --project=. bin/pddlica inspect model.pddlica.json
julia --project=. bin/pddlica capabilities domain.plca problem.plca \
  --backend milp --max-steps 5 --makespan 5