6.1.1. Experimental results
All the tests were performed using a dual core machine with a 2.50 GHz Intel i5 processor, 6 GB of RAM, Ubuntu 12.10
operating system with 64 bit support. We used a time cap of 130 seconds for all the runs. To solve the encoding of the
problems in PDDL we used two planners: Metric-FF [37,38] and Madagascar-p [39].11 The first solver is based on Graph
Plan [40], a standard planning algorithm to prune the search space. The second solver instead belongs to the Satplan [41]
family and encodes the planning problem into a SAT formula and then uses state-of-the-art SAT solvers to find a solution.
Since for decidability purposes the use of PDDL requires a finite use of objects, for reducing the search space of the solvers
we set in the PDDL encoding the number of components that could be used concurrently to the minimum possible value.
The performance of the two planners is summarized in Table 2 where error indicates that the solver exited with an error
state without computing the plan, timeout means that the solver took more than 130 s and was interrupted, a dash means
that the test was not conducted because the previous execution already timed out or ended in error.12
The performances of the general planning solvers are quite limited: they are able to compute plans for just a small
number of components. These poor performances are due to the fact that the size of the encoding of the planning problem
increases exponentially w.r.t. the number of components that need to be deployed concurrently. In particular, Metric-FF
6.1.1. Experimental resultsAll the tests were performed using a dual core machine with a 2.50 GHz Intel i5 processor, 6 GB of RAM, Ubuntu 12.10operating system with 64 bit support. We used a time cap of 130 seconds for all the runs. To solve the encoding of theproblems in PDDL we used two planners: Metric-FF [37,38] and Madagascar-p [39].11 The first solver is based on GraphPlan [40], a standard planning algorithm to prune the search space. The second solver instead belongs to the Satplan [41]family and encodes the planning problem into a SAT formula and then uses state-of-the-art SAT solvers to find a solution.Since for decidability purposes the use of PDDL requires a finite use of objects, for reducing the search space of the solverswe set in the PDDL encoding the number of components that could be used concurrently to the minimum possible value.The performance of the two planners is summarized in Table 2 where error indicates that the solver exited with an errorstate without computing the plan, timeout means that the solver took more than 130 s and was interrupted, a dash meansthat the test was not conducted because the previous execution already timed out or ended in error.12The performances of the general planning solvers are quite limited: they are able to compute plans for just a smallnumber of components. These poor performances are due to the fact that the size of the encoding of the planning problemincreases exponentially w.r.t. the number of components that need to be deployed concurrently. In particular, Metric-FF
การแปล กรุณารอสักครู่..
