Competition Domains for the temporal track

Also problems for Transport, Woodworking and Modeltrain domains from IPC 2008 were created, but they were finally discarded as they contain numeric preconditions which few participating planners were able to handle.

CrewPlanning

There is no description of this domain at the IPC 2008, but here you are a couple of papers describing it. Although SGPlan 6 solved all the problems at IPC 2008, the remaining planners only solved half of them, so we have reused all the problems. The selected problems are:

Elevators

For a domain description click here.

At the last IPC, there were two versions of this domain. Only 2 planners participated in the numeric version, while in the STRIPS there were 6 planners (including baseline) solving at least one problem. This year we have used only the STRIPS one. The quality of the found plans was quite poor and worse than those of the baseline.

We will select 10 old problems and create 10 new ones. In the 2008 version, problem one starts with 9 floors, 4 passengers, 2 slow elevators and 2 fast ones. Until problem 10, passengers are increased by one (13 passengers in problem 10), while the other parameters remain constant. Problem 11 has 17 floors and 8 passengers and passengers are increased by 2 (26 passengers in problem 20). Problem 21 has 25 floors, 12 passengers and 3 slow elevators, which are increased by 3 (39 passengers in last problem).

At IPC 2008 problems were exactly the same as in sequential satisficing track, so do we have created new problems with the same parameters than in that track: problem 11 starts from characteristics of old problem 30 (current problem 10), adding 2 fast elevators, 1 slow and 1 passenger. The next 4 increase the number of passengers by 3. From problem 16 floors are 40, 4 fast lifts with capacity for 6 passengers, 4 slow ones for 4 passengers, and 40 passengers, increasing the number of passengers in 5 per problem.

Floortile

Author: Tomás de la Rosa.

Temporal version of the Floortile domain. For the temporal track, 5 problems for each configuration have been created:

Matchcellar

• Author: Bharat Ranjan Kavuluri

This is a STRIPS version of the domain proposed by Bharat Ranjan Kavuluri. Domain is inspired in this paper. The main feature of this domain is that a lighted match is concurrently required to fix a fuse.

Openstacks

For a domain description click here.

There were 4 domain versions at IPC 2008: numeric, adl, adl-numeric and STRIPS. Most of the planners were only able to handle the STRIPS version. Most of the temporal planners participating in IPC 2011 only support STRIPS, only one supports ADL, so in this IPC we have created only STRIPS problems. Note that, at least in STRIPS, this domain has a different domain file for each problem.

At IPC 2008, three planners (including baseline) were able to solve all the problems. For this year competition we have reused the 10 most difficult problems and have generated 10 more difficult ones. In the IPC 2008, first problem has 5 objects and objects are increased by 1 from problem to problem (problem 30 has 34 objects). The density parameter has been lost.

We have generated problems with 80% density. From old problem 30 (current p10) we increase objects by 1, till 44 objects in problem 20. Given that, the selected problems are:

Parcprinter

For a domain description click here.

There is no generator for this domain, so new problems have been by-hand generated using IPC 2008 ones. New problems add some sheets to old problems or make two-sided some of them. Like in sequential tracks, there are 3 different printers. It seems that printer 3 is the easiest and number 2 the toughest, so 4-3-3 new problems have been generated. It also seems adding an extra image is more difficult than an extra sheet.

We have selected the following problems:

Parking

This domain is a temporal version of the domain created for the learning part of IPC2008. This domain involves parking cars on a street with N curb locations, and where cars can be double-parked but not triple-parked. The goal is to find a plan to move from one configuration of parked cars to another configuration, by driving cars from one curb location to another.

For the temporal track the following problems have been created:

Pegsol

For a domain description click here.

At IPC 2008 problems in temporal satisficing, sequential satisficing and sequential optimal were the same. They were taken from a pool of 105 problems. In this year's competition we have reused 20 problems.

Sokoban

For a domain description click here.

Performance of planners in this domain was quite poor at last IPC(12 problems unsolved), so instead of generating new problems we have reused the most difficult problems of last IPC. The problems are:

Storage

Corresponds to the "time version" from the IPC-2006 domain where actions have duration and the plan quality is total-time (plan makespan). This domain deals with moving a certain number of crates from some containers to some depots by hoists. Inside a depot, each hoist can move according to a specified spatial map connecting different areas of the depot. The test problems for this domain involve different numbers of depots, hoists, crates, containers, and depot areas. The domain has five different actions: an action for lifting a crate by a hoist, an action for dropping a crate by a hoist, an action for moving a hoist into a depot, an action for moving a hoist from one area of a depot to another one, and finally an action for moving a hoist outside a depot.

Five problems for each configuration have been created:

Temporal Machine Shop

Author: Frédéric Maris

The "tms-k-t-p" domain (temporal machine shop, first proposed in [2]) is inspired by a real-world application. It concerns the use of k kilns, each with different baking times, to bake p ceramic pieces (bake-ceramic) of t different types. Each of these types requires a different baking time. These ceramics can then be assembled to produce different structures (make-structure). The resulting structures can then be baked again to obtain a bigger structure (bake-structure). We have defined too a "light" version of these domain for temporally-expressive planners which do not support richer durative actions (that is with time intervals).

All possible solutions require concurrency of actions (temporally expressive problem).

Although many temporal planners have been compared in the International Planning Competitions (IPC), recent theoretical studies have brought to light the limitations of the current approaches to temporal planning [1]. [2] shows that the domains and problems which have been used up until now in the last competitions can always be solved with a sequential plan. They propose a method to prove that a domain can only be solved using concurrent actions. In fact, the winning planners in the IPC competitions, even if they are efficient in a restricted temporal framework, cannot solve problems for which all possible solutions require parallelism (temporally expressive problems) but only those for which there is at least a sequential solution (temporally simple problems). So, they are therefore far from being capable of solving real-world problems. The objective evaluation of these systems requires the setting up of new benchmarks corresponding to temporally expressive problems.

References:

[1] W.Cushing, S.Kambhampati, Mausam, D.S.Weld, "When is temporal planning really temporal ?", IJCAI, pp. 1852-1859, 2007.

[2] W.Cushing, S.Kambhampati, K.Talamadupula, D.S.Weld, Mausam, "Evaluating temporal planning domains", ICAPS, pp. 105-112, 2007.

[3] Maris F., Régnier P., 2008, "TLP-GP: New Results on Temporally-Expressive Planning Benchmarks", in Proceedings of 20th IEEE International Conference on Tools with Artificial Intelligence (ICTAI-2008), vol. 1, pp 507-514, Dayton OH, USA, November 2008.

[4] Maris F., Régnier P., 2008, "TLP-GP: Solving Temporally-Expressive Planning Problems", in Proceedings of 15th International Symposium on Temporal Representation and Reasoning (TIME-2008), pp 137-144, Montreal QC, Canada, June 2008.

Turn and Open

Author: Sergio Jiménez Celorrio

In this domain there are a number of robots, with two gripper hands, and a set of rooms containing balls. The goal is to find a plan to transport balls from a given room to another. There are doors that must be open to move from one room to another. In order to open a given door the robot must turn the doorknob and open the door at the same time.