theta* pathfinding

An algorithm for the any-angle pathfinding problem:

For the Legacy world map I'd like to be able to find natural looking paths that avoid obstacles. I've got a lot of terrain and it's easy to turn that terrain into a potential field (bitmap) which can be used as a cost lookup for a pathfinding algorithm.. but.. there are some issues. In particular I don't want the paths to look like they are constrained to a grid. And if I make the grid cells small enough so that I get naturalish results, then the algorithm runs slowly. Theta* seems to be an interesting candidate; close to the speed of a grid-based A* but with more natural resulting paths, similar to results you'd get from path smoothing. I'd possible still want to round off the corners with post processing.. I don't know.

Assumedly, another approach would be based on steering, or a hybrid coarse grid A* with steering for local control. That sounds like it could take a long time to tune, but might give very good results. For the non-player actors, efficient paths are less important, and some kind of wandering/pheremone trail combination could be interesting. In particular, if you had such an emergent system, then you could use it to mark local paths, which then could become your nav grid. A lot of preprocessing I guess, but maybe that's ok.

You know what else sounds cool? Probabilistic roadmaps.

a dumb problem I had today

This is a technical post.

Today VisualVM stopped profiling my applications. No apps showed up in the bar on the left. After many false leads involving rival jvms and 32/64 bit confusion, I tracked this down to "not able to read the temp folder," which it turns out, is because my temp folder is now pointing to C:\Windows\Temp instead of the usual location under Users. Honestly no idea how that changed, perhaps an installer did it at some point? Who would do such a thing. Why.

Anyway by navigating to that folder in windows explorer it prompted me to elevate privileges for it, and that resolved the issue with VisualVM.

I still hate VisualVM. I don't know, an error message with the folder it was trying to access might have been nice. Would have saved me 3 or 4 hours today. Still, I'm reluctant to spend $500 for a professional profiler.


I'm making a lot of great progress on Legacy though. It's great fun most of the time. :-)


I accidentally looked up Celtic knots last night. Don't do that. It's a pretty amazing graphic design (and math?) rabbit hole. Also hey, Insular Art, the Book of Kells (the movie Secret of Kells is about a real book, I did not realize that), and the Picts, who seem to have directly inspired both Skyrim's Forsworn and Terry Pratchett's Nac Mac Feegle.

I found this software which is actually a really good starting point for the more regular knots. Where it appears to fall down is that the curves are not smoothed, so you can get a rough circle out of it, just not a smooth one, and getting to do complicated things with mixed weights sounds more difficult than drawing it myself. But it's very strong as a playground and as a place to sketch your design, from there bring it into illustrator or draw it by hand...

By taking the output and slicing it up into a tile and putting it into a mesh I was able to get what I was looking for.

(need a better bg tile obv)

...and I think I better just call it a win and stop right there. I don't want to get lost. To that end I've decided that every time I see the word "Gaelic", I'll shift a letter and internally read it as "Garlic." That should help things.

java development = the desert

Profiling java applications for free seems to be just barely possible. VisualVM is buggy, hard to use, and has the smallest feature set I've seen in a profiler. It's considerably worse than the tools that Adobe provides for actionscript. (?!)

Anyway if you have a memory leak and you're all "dag yo, what could be hanging on to that reference?" and you refuse to pay for a good profiler, then here's how you can do it with VisualVM:

fun with ai
or, how to be mostly right most of the time.

oh dear oh dear.

people must be using openCL for this sort of thing these days?

mm juices flowing. After Legacy, maybe something with a ton of realtime AI is next..