Jordan Thayer and Wheeler Ruml and Jeff Kreis

Using Distance Estimates in Search: A Re-evaluation

More Results:

Revised Dynamically Weighted A*

Tie-breaking on d

Simple modicfications of improve performancedrastically.

In temporal planning,using distance estimates avoids catastrophe at high weights.

When using distanceestimates does notdrastically improve thesearch, it often does notsignificantly harm it.

Finds optimalsolutions in halfthe time of A*.

Surprising numberof nodes in last f(n) layer of standardbenchmarks.

Original algorithm rewardsdepth, not progress towards goal.

New formulation:

Greedy on distance-to-go,d,beats greedy oncost-to-go, h.

Searching on distance works because cheap paths may be longer interms of search effort.

Dynamically weighted A*decreases greediness asdepth increases.

expands the nodeclosest to the goal thatis within the suboptimalitybound.

Previous Work:

Can using distance-to-go estimates improve the performance of bounded suboptimal search?

Our Contributions:

Node orderings forinterfere with one another.

ggs

5 5

1 1

1 1 1