Bot running speed

[Neoptolemus]

Have a good trip, wherever you're going. I'll keep posting progress. I have implemented bot vision using proper viewing frustums and using cylinders for player visibility tests and the results are pretty much perfect, but I need to look at occlusion as well. This bit is harder!

I've also implemented basic objectives, so the hostage starting points and the rescue point/zone are read from the BSP file and stored so the bots know where they have to go, but I delibrately avoid tracking hostages so that, as with humans, the bots have to search for them if they're no longer at their usual spot.

I'll try and release an initial version when it's at a point where they're fun to play with and against. It won't be near the level of ePODBot of course, but should be interesting to see how good I can get their navigation and movement without waypoints

I've also been scribbling down ideas for tactics such as weighting nodes if they're visible to a sniper to force the bot to take a different route and hug walls etc, and automatic identification of good sniping and camping spots. Can't wait to start implementing that!

[The Storm]

I got a bit of limited time with internet access until my next trip starts tomorrow.

Good work. About visibility, do you consider dark areas?

Also to be fun with bots 90% of the work is in the movements. So when you get this in good state you will be pretty much done.

[Neoptolemus]

Eventually I will handle dark areas, so bots can use nightvision, and possibly even use the dark to hide and ambush. Right now it just uses the view frustum and a trace to various parts of the body to check for visibility.

I've been trying numerous ways of handling movement to avoid bots getting stuck and I'm starting to get there I think. I can now have a bot run around cs_italy for a good 3 minutes before he gets caught in a corner, and his movement is pretty much perfect up until then.

I'm doing it by searching in a small radius (about 8 units) around the bot for the edge of the navmesh, taking the hit hormal of the nearest point to the edge from the bot's origin, and giving him a little nudge away, kind of like a magnet repelling another magnet with the same charge.

This works most of the time, but it seems like it's possible for the bot's origin to stray outside the navmesh, meaning in some cases the normal is inverted and ends up pushing them into the wall! I have added a sanity check that traces 20 units in the desired push direction, and if the trace hits a wall then it inverts it.

More testing is needed but so far, looking good! There are still plenty of minor issues to deal with: my technique does have the side effect of causing bots to rapidly zig-zag down narrow corridors where it can't help but touch the navmesh edge on either side, and if it gets stuck in a corner where it is blocked on one side AND in front, it will still be stuck as it's trying to move diagonally rather than strafing.

[Neoptolemus]

Bit of an update, managed to get the bot to spend 15 minutes running around cs_italy and de_dust2 without getting stuck. The movement isn't quite perfect yet, as they appear to bounce off walls a little, and their desire to hug corners conflicting with the repelling system gives them the appearance of being slightly drunk.

This is most noticeable in confined spaces (such as the wine cellar in cs_italy) where their movement is a little "wobbly". They still get around at a good speed, it just looks a little odd. I'll upload another youtube video when I have a chance to show it off.

The main thing though is that they're able to run smoothly and effectively