When a weak planning call fails for a complete state, it is typically not a DRDE, and we don't have a good generalization (making FSAP's useless). If the weak call is fast enough, do the generalization on (bounded) weak planning calls with auxillary actions added to allow variable re-assignment at the start.
If the bound is hit, assume (conservatively) that plan exists and remove the aux action. Otherwise, keep it in (signifying that the variable is relaxed in the partial deadend state), and move to the next variable. This may find us more powerful generalized deadends when DRDE's are rare.
When a weak planning call fails for a complete state, it is typically not a DRDE, and we don't have a good generalization (making FSAP's useless). If the weak call is fast enough, do the generalization on (bounded) weak planning calls with auxillary actions added to allow variable re-assignment at the start.
If the bound is hit, assume (conservatively) that plan exists and remove the aux action. Otherwise, keep it in (signifying that the variable is relaxed in the partial deadend state), and move to the next variable. This may find us more powerful generalized deadends when DRDE's are rare.