Closed MKaysen2 closed 4 years ago
For the backward chaining in AStar, I'm pretty sure that the prior value v should be
argmin_v h(v) where v in V_k s.t. P(v)
where V is the domain of key k, P(x) is the truth value of precondition P for some x, and h(x) is the heuristic distance of value x, which should be a unique value for ordered keys (comparators other than eq are undefined for non-ordered keys)
I will probably hard code the cases in, but that's how I'm computing the values
edit: and I say pretty sure because I don't feel like proving it. Should hold if the heuristic is admissible and consistent, so don't mess up the edge costs
added improved effects to the planner, though currently it doesn't use the improved preconditions
currently only using SET and EQ operations, should add increment and decrement for the effects, and <, <=, >, >= for the preconditions