Hacker News Reader

Comment by mrcharles | Hacker News Reader

I call bullshit on your pathfinding AI excuses. A* is A*, it doesn't matter if you dynamically modify the terrain, and it's easy enough to check if a path would end at walls or impassable terrain, and have the AI deal with that accordingly.

View on HN

> and have the AI deal with that accordingly.

Yeah, that's kind of the problem I alluded to, isn't it?

But you're right, I don't know the state-of-the-art in pathfinding...as I admitted, the most recent RTS I played was Warcraft 3. It just seems to me to be a very common complaint in most modern games. I still play Team Fortress 2, and even though they've implemented AI, the AI only works on certain pre-designed levels. Yes, I know that a FPS is generally more path-computational than a RTS, but if you're proposing that an RTS have 3D terrain then, well, that makes it a 3D game, no?

And it's been awhile since I've designed a game but I think you're underestimating the computational power it would take for an AI to dynamically generate efficient paths with every change to the terrain. IIRC, a lot of early AI compensated by having pre-calculated paths for each level (and I think virtually all of them only pretended to be encumbered by a fog of war).

My point still stands though...these aren't really innovations...again, I'm not in game development, but I'm guessing that some of these are ideas that teams consider and then drop in lieu of their difficulty of implementation. In other words, the innovation would be the actual implementation of the idea, not the idea itself.

Using a tile or node based A* map, there's no additional cost for altering heuristics on the fly. All paths are always dynamically generated on the fly.

The only reason early AI had 'pre-calculated paths' (an assertion I am dubious about, to be honest) would be that it was running on processors that could barely push the graphics of the time to begin with.

Those days are loooooong behind us.

The trouble with dynamic terrain--and especially walls--is that it's hard to place a cost on destroying walls to get to where you want to get. A* is simple enough, but adding tactical considerations to it is much harder. It's a balance of "I will have a harder time hitting the opponent" vs. "I get to do damage right now, before my opponent gets to do any more tricksy maneuvering."

C&C was particularly terrible with walls; if there was any path at all, it wouldn't destroy walls, which made it easy to ambush.

I think it's pretty standard that if you issue a move command, units will not attack and will go around. Players have come to expect this since Starcraft I. The whole idea of tower defence that was made popular stands a testament to the fact that we can place walls and change terrain and units WILL change their path on the fly. If you have played a tower defence in Warcraft III, you know that units can be lured into taking the A* optimal path, and you can block it off just before they get there, forcing them to recalculate a new path (which may be a long way away from where they currently are). Then, you can destroy the new tower you just placed and block off the path they had just changed to, forcing them down the old path again. You can repeat this as long as you have resources to build your cheapest tower (this was an exploit that made certain user-created maps really easy). Tower Defence designers now know about this and most of the well-made maps prevent you from building during level.

TL;DR: Path recalculation has been around since Starcraft I. EDIT: And probably before that

You're missing the point of his post - he was saying that while path recalculation exists, it's hard to make an AI that makes proper choices when it comes to following the current path and creating a new path - especially when it's opponent can also modify the available paths in the future.

I'm saying that players have come to expect that issuing "move" will make AI take the shortest path without destruction, and we know that if there is a faster way, we have to tell the AI to attack through the wall if we want to break it. I guess the TL;DR didn't cover that though.

Reply on news.ycombinator.com