Box2D Forums

Hi,

I'm just breaking ground on a little API for cloning a box2d world in order to run some speculative simulation for the AI players. I'm wondering, has anyone already done this before, specifically a b2World::clone() method or similar?

This will clone a world, but I doubt it would be fast enough to run many times per second on large worlds.
http://www.iforce2d.net/b2djson/
Just wondering, why does the AI need to clone the world? Does every AI entity do that?

The AI and some special moves require a predictive element. There is too much going on to predict the future accurately without simulating it directly. I can't reuse the main game b2World because it uses a tiny fixed timestep (8.3ms) and I'd rather let the AI and various predictive player elements have a full second of insight into the future using a 50 or 100ms timestep. The warm starting mechanism that creates very stable contacts works best when timesteps are fixed for a given b2World.

Incidentally, this will also make ball interception by the AI much easier to code.

The dumb AI inside this shadow simulation will use rand() to choose joystick direction and button presses, plus a few filters to keep their search space bounds a bit tighter. I've had very good luck using a brute force strategy in games where I had enough extra CPU around to use up.

After some testing this week, I've had to conclude that a second world for the AI is not a viable approach on the iPad.

I am testing to see if using raycasted collision reaction and simple velocity/position updates is enough to allow the AI the insight it needs. I don't need the full precision of Box2D, but I do need a chance to play out various scenarios. I'm lucky that all the dynamic elements (4 players, 2 goalies, 1 ball) are circles, so it is easy for me to calculate if a collision occurred.

Ok, this is pretty cool. I just copied the core integration code from b2Island::Solve and use that to advance my objects for the AI. It is enough for what I am trying to do. 8KHz.

Thought anyone seeing this thread might enjoy seeing a video of the AI in action (primitive so far, but visualization looks cool):

http://www.youtube.com/watch?v=2TAsrAbKwo4

Nice! It's all moving a bit too fast to catch what is going on, but definitely looks natural.