Problem with MaximizeExpressionOnFinalState in HierarchicalProblem with the Aries backend

[nicola-gigante]

Following the answer by @arbimo on #614 I've tried to use the MaximizeExpressionOnFinalState metric and I'm having troubles making it work.

This is a minimal example:

from unified_planning.shortcuts import *
import unified_planning.model.htn as htn
import unified_planning.engines as engines
import unified_planning.model.metrics as metrics

problem = htn.HierarchicalProblem()
problem.epsilon = 1

f = Fluent('f', IntType())
problem.add_fluent(f, default_initial_value = 0)

a = InstantaneousAction('a')
a.add_increase_effect(f, 1)

nop = InstantaneousAction('nop')

t = htn.Task('t')
m = htn.Method('m')
m.set_task(t)
m.add_subtask(a)

m_nop = htn.Method('m-nop')
m_nop.set_task(t)

problem.add_action(a)
problem.add_action(nop)
problem.add_task(t)
problem.add_method(m)
problem.add_method(m_nop)

problem.add_quality_metric(metrics.MaximizeExpressionOnFinalState(f))

problem.task_network.add_subtask(t)

with AnytimePlanner(
    name='aries', anytime_guarantee="INCREASING_QUALITY"
) as planner:
    for solution in planner.get_solutions(problem):
        print(solution)

When I run it I get this output:

status: INTERMEDIATE
engine: aries
plan: Hierarchical TimeTriggeredPlan:
status: SOLVED_OPTIMALLY
engine: aries
plan: Hierarchical TimeTriggeredPlan:

That is, the engine says the optimal plan is empty. This means the m_nop method is used to perform the task t. But the m method would have executed action a which has an increasing effect on f, which is the objective to maximize.

So I don't understand what's happening here. What am I doing wrong?

Thanks,
Nicola

[arbimo]

Thanks for the report, this is definitely a bug on Aries side, fixed here : plaans/aries#139
In a nutshell, maximization was still interpreted as minimization.

The fix should be available in the development version in a few minutes.

[nicola-gigante]

Thanks!