We create our problem domain- Scientific Conference and produce the plan for the domain with Metric-FF and LPG. Both planners are state based with PDDL compatible which could resolve our problem involved with time handling and resource management. After testing with a set of files on our domain, we get conclusion as below.

Firstly, PDDL2.1 level 2 is good at numerics and it is able to handling numeric value. By handling numeric, it is possible to add a currenttime variable that simulate time. With Metric-FF and LPG which support PDDL 2.1 level 2, we can deal with time issue and battery problem. However, the planners do not have a good performance on our domain. Some simple problems with few goals can not be resolved in a short time and take a lot of memory. In some case, they don't succefully make a plan. The main reason is that these planners are not designed to do time scheduling. So, the planner take too much time to order actions. We do not recommend Metric-FF and LPG to handle time scheduling.

Secondly, PDDL 2.1 do not support an action with numeric parameters. Instead, parameters should be in the form as objects in finit set. This is a limitation in the case when time is sencentive.For example, we can not pass a numeric value into Socialize(x)--x is a numeric value. One of alternative ways is we have to add a precondition current-time=firingTime. We also introduce wait actions that have the capability to wait until a fixed time like the beginning of a presentation session. Waiting until a fixed time point seems to be the only way to introduce operators that can consume variable time duration.

Thirdly, Enforced Hill-Climbing algorithm (EHC) used in Metic-FF play an important role in plan research. Generaly, EHC improve performance of the planner. But, in our domain, it is not the case. In our tests, running Metric-FF without EHC (by using -E option) is generaly faster and consume less memory.

Fourthly, LPG has two limitations in our problem domain. The bigger one is that LPG can not make an identical result based on the same input file. The reason may result from the randomly selection for path search. But LPG uses best-first search algorithm which assure a nondeterministic choosing. So we are quite confused with the result. Another limitation is LPG does not support Equal operator. There is a precondition in Make-Presentation and Assist-Presentation: (= (currenttime) (time-pres-begin ?ses)). Because Equaloperator is not fully supported, we believe that LPG assume that this condition is always true. Therefore, the robot always make a presentation without wait even the time is not the current time yet. Also, for bigger problems, LPG fails to resolve what Metric-FF was able to do. So, for this reason, we prefer Metric-FF as LPG.

For our Scientific Conference domain, a Chronical combined with HTN approach seems to more suitable for our scientic conference domain. If we choose Chronical approach, the time problem in our domain will be easier to resolved. The HTN will supply powerful methods to decomposit our tasks. For example, Show-Poster Tasks needs two actions from common sense:  Fix-Poster and Remove-Poster. It is what HTN does if it is applied on our domain. But in Metric-FF, we have to add a predicate poster-fixed to link both actions.