Auto / manual waypointing hybrid?

[botmeister]

We're now seeing some excellent work going on where bots can successfully generate their own waypoints.

Are manually generated waypoints history?

It would seem that a manually edited waypoint system will always beat an auto generated one if created by a skilled waypointer. There are simply too many things to consider for an auto waypointer to deal with effectively, and there are special cases that an auto waypointer cannot deal with, like having defensive waypoints set up to simulate, for example, a clan match where the CT bots defend bomb sites and never venture out from their assigned positions.

I think the auto waypoint generaters are a great idea. For example what I see with Realbot WIP #6 is amazingly good already, and what PMB has proposed should be even better.

What I am wondering, is it possible to still allow waypoints to be manually added to a bot like the new Realbot? Also, can the auto generated waypoints be manually edited to "enhance" them, by adding camping points, etc?
A bot with both auto and manual generated waypoints would make for an excellent bot. It would also be excellent if auto generated waypoints could be manually fine tuned and then "fixed" in place.

What do you guys think - is it doable?

The official SC-bots doing it`s succesfully !!!

but autowaypoint has some bugs ...

this bots have alternative navigation system unlike PODbot. He does not need camp-points, and this is good

Manually created PODbot-waypoints allow tune the bots-tactics strongly

[stefanhendriks]

i am planning an editor so you can fix bad connections, about specifying camp spots and such, i wonder if that will be enough. The problem i see with camping is not camping itself, but looking into the correct direction..

using 'look @ nodes for camping' is an idea though...

[Pierre-Marie Baty]

It is theoretically possible, with my navmesh implementation, to add arbitrarily specially-flagged navnodes, but they will be on "top" of existing ones, duplicates of an existing navnode (since the navmesh already includes ALL the walkable surface of the map).

The problem though, is that I wonder what sort of use they would be for the bot, since with my implementation, it's the bot itself which builds its own, ever-changing, paths (on a per bot basis), and the location of the navlinks (~= the waypoints) is ever changing too during the game ; and foremost it's the bot itself which decides, relatively to its personal knowledge and experience, what use it will make of such or such navlink (e.g, camp spot, passage to avoid, etc.) Thus it will be possible to ADD static nodes, but not to edit the existing ones, which are completely dynamic.

Hence I wonder if it's worth the try, for me at least.

[botmeister]

The official SC-bots doing it`s succesfully !!!

Yes that's right, I had forgotten about that, the official CS Bots are supposed to allow manual tuning.

Interesting what PMB posted. If I read it correctly, you are saying each bot will have its own waypoint network. If manual waypointing were allowed, that would mean allowing edits on a per bot basis. I can see the static waypoints as being usefull, if just to prevent the bots from doing silly things, or to force them to do smart things (more or less). You could have one common set of static waypoints which all the bots use, which would be easier to set up than haveing static waypoints for every bot.
One could argue that a lot of the intelligence is built into the waypoint / navigation system.

I personally love the idea of auto waypointing, even more than getting perfect waypoints from the internet. You can see the bot learn, and get better at navagation etc. It's great! Plus I know that my bots are different to any other bots

A suggestion for the camping problem, though I haven't programmed bots before so I don't know how viable it is. Perhaps you could look at the human player and see if he's stopped somewhere. If he stops for a decent length of time while still looking around, that spot could be labelled as a possible camping spot, with a look at vector - the average of where the human player is looking at.