Hi, Looking for some critiques from the HTN planning gurus out there. Let's say you have a simple hierarchical ordered planner (like SHOP, unsurprisingly), but everything is procedural (like in the PRS system). So your pre-conditions are not data-driven but hard-coded. The planner just does a greedy "execution" of the tree, super fast, but it doesn't necessarily know the difference between what's a pre-condition and what's the task itself. Great for modularity, but I'm wondering what this approach misses out on -- in terms of performance. What would having non-hardcoded pre-conditions buy you? Traditionally, having data means you can manipulate it better, but in the case of ordered planners, does that give you any extra speed somewhere? Comments welcome. Alex P.S. More specifics on this top secret project soon, stay tuned :-) AiGameDev.com [Edited by - alexjc on June 25, 2007 2:50:31 AM]

Some links, for those interested:

http://www.ai.sri.com/~prs/
http://www.cs.umd.edu/projects/shop/

Alex