Action space reduction for planning domains

1 . A computer-implemented method (CIM) to save computer resources in execution of reinforcement learning artificial intelligence software, the CIM comprising:

receiving a domain-dependent artificial intelligence planning problem including definitions for a plurality of operators, the domain-dependent artificial intelligence planning problem a model-based approach relying on a symbolic model to guide a search for a solution to the domain-dependent artificial intelligence planning problem;

creating an initial version of a label set, which defines an initial version of an action space, with the label set including a plurality of labels, and with each label of the plurality of labels respectively corresponding to the operators of the plurality of operators;

performing, automatically and by machine logic, a label reduction on the initial version of the label set to obtain a reduced version of the label set that defines a reduced action space;

recasting the artificial planning problem as a first Markov decision process, the first Markov decision process using the reduced version of the label set generated by the machine logic; and

resolving the artificial intelligence planning problem relying on the symbolic model and outputting a planning recommendation by performing reinforcement learning using the first Markov decision process with the reduced version of the label set generated by the machine logic.

2 . The CIM of claim 1 wherein the performance of the label reduction includes:

determination, by machine logic, of a mutex group of operators from the plurality of operators; and

using the mutex group of state dependent operators to reduce a number of labels in the reduced version of the action space relative to a number of labels in the original version of the action space.

3 . The CIM of claim 1 further comprising:

translating the artificial intelligence planning problem to delete-free planning terms.

4 . The CIM of claim 1 further comprising:

exploring the space of plans to obtain a seed set of high quality.

5 . The CIM of claim 1 wherein the performance of label reduction includes:

finding a mutex group via reduction of operator parameters, such that the mutex group is found separately for each operator of the plurality of operators.

6 . A computer program product (CPP) to save computer resources in execution of reinforcement learning artificial intelligence software, the CIM comprising:

a set of storage device(s); and

computer code stored collectively in the set of storage device(s), with the computer code including data and instructions to cause a processor(s) set to perform at least the following operations:

receiving a domain-dependent artificial intelligence planning problem including definitions for a plurality of operators, the domain-dependent artificial intelligence planning problem a model-based approach relying on a symbolic model to guide a search for a solution to the domain-dependent artificial intelligence planning problem,

creating an initial version of a label set, which defines an initial version of an action space, with the label set including a plurality of labels, and with each label of the plurality of labels respectively corresponding to the operators of the plurality of operators,

performing, automatically and by machine logic, a label reduction on the initial version of the label set to obtain a reduced version of the label set that defines a reduced action space;

recasting the artificial planning problem as a first Markov decision process, the first Markov decision process using the reduced version of the label set generated by the machine logic; and

resolving the artificial intelligence planning problem relying on the symbolic model and outputting a planning recommendation by performing reinforcement learning using the first Markov decision process with the reduced version of the label set generated by the machine logic.

7 . The CPP of claim 6 wherein the performance of the label reduction includes:

determination, by machine logic, of a mutex group of operators from the plurality of operators; and

using the mutex group of state dependent operators to reduce a number of labels in the reduced version of the action space relative to a number of labels in the original version of the action space.

8 . The CPP of claim 6 wherein the computer code further includes instructions for causing the processor(s) set to perform the following operation(s):

translating the artificial intelligence planning problem to delete-free planning terms.

9 . The CPP of claim 6 wherein the computer code further includes instructions for causing the processor(s) set to perform the following operation(s):

exploring the space of plans to obtain a seed set of high quality.

10 . The CPP of claim 6 wherein the performance of label reduction includes:

finding a mutex group via reduction of operator parameters, such that the mutex group is found separately for each operator of the plurality of operators.

11 . A computer system (CS) to save computer resources in execution of reinforcement learning artificial intelligence software, comprising:

a processor(s) set;

a set of storage device(s); and

computer code stored collectively in the set of storage device(s), with the computer code including data and instructions to cause the processor(s) set to perform at least the following operations:

receiving a domain-dependent artificial intelligence planning problem including definitions for a plurality of operators, the domain-dependent artificial intelligence planning problem a model-based approach relying on a symbolic model to guide a search for a solution to the domain-dependent artificial intelligence planning problem,

creating an initial version of a label set, which defines an initial version of an action space, with the label set including a plurality of labels, and with each label of the plurality of labels respectively corresponding to the operators of the plurality of operators,

performing, automatically and by machine logic, a label reduction on the initial version of the label set to obtain a reduced version of the label set that defines a reduced action space;

recasting the artificial planning problem as a first Markov decision process, the first Markov decision process using the reduced version of the label set generated by the machine logic; and

resolving the artificial intelligence planning problem relying on the symbolic model and outputting a planning recommendation by performing reinforcement learning using the first Markov decision process with the reduced version of the label set generated by the machine logic.

12 . The CS of claim 11 wherein the performance of the label reduction includes:

determination, by machine logic, of a mutex group of operators from the plurality of operators; and

using the mutex group of state dependent operators to reduce a number of labels in the reduced version of the action space relative to a number of labels in the original version of the action space.

13 . The CS of claim 11 wherein the computer code further includes instructions for causing the processor(s) set to perform the following operation(s):

translating the artificial intelligence planning problem to delete-free planning terms.

14 . The CS of claim 11 wherein the computer code further includes instructions for causing the processor(s) set to perform the following operation(s):

exploring the space of plans to obtain a seed set of high quality.

15 . The CS of claim 11 wherein the performance of label reduction includes:

finding a mutex group via reduction of operator parameters, such that the mutex group is found separately for each operator of the plurality of operators.