16 replies to this topic

[Kazu Kun]

The only way to reproduce this is with these civilizations. (I don't know if the Persians won't respond to anyone or if it's just the Spartans)

Random (Not Campaign)
- Map (Deep Forest) - Please note that there was ABSOLUTELY NO WATER on this map.
- Map Size (Small) - 2 players
- Conquest,
Reveal Map [Enabled]
Team Lock [Enabled]
___________________________
Player 1 (Spartans)
Player 2 (Persians) - Qbot

- Details
37 Minutes into the game I decided to go ahead and destroy him since I had my base fully built, militia was well produced, and walls set up.
Throughout all the game all he had produced was.
8 Buildings ( 4 Houses, 1 Market, 3 Towers )
And only had 5 Units created, while I killed 13.

- Problem
Obviously the problem is the AI isn't responding very well on this map, it will kinda poop itself and sit there and waits to be changed.

[historic_bruno]

One problem I noticed is there are a bunch of females standing around. I investigated their UnitAI state (open the Developer overlay with Alt-D, check "Display selection state" and select something) and it said "gathering". From the gather target entity ID I was able to find what they wanted to gather. It turns out that map generates some chickens inside the foundation of the civil center (use Alt-W to toggle wireframe mode to see this), and because they are the closest target to the civil center dropsite, all these workers are trying to gather from it but failing. That in turn economically cripples the AI.

It's just a bug in the Deep Forest script, I'll create a ticket for this. Thanks for reporting it

[Geek377]

They also gather around the edges of the map at times. My guess is there might be some trees or other resources beyond the SOD and they try to reach it. Either that or they just love getting in a group and swinging axes near the map edge

[Pureon]

I thought the AIs would now ignore resources they cannot access?

[historic_bruno]

Pureon, on 25 June 2012 - 09:33 AM, said:

I thought the AIs would now ignore resources they cannot access?

Not in this case, at least I don't know about the edges of the map.

[FeXoR]

historic_bruno, on 25 June 2012 - 12:22 AM, said:

One problem I noticed is there are a bunch of females standing around. I investigated their UnitAI state (open the Developer overlay with Alt-D, check "Display selection state" and select something) and it said "gathering". From the gather target entity ID I was able to find what they wanted to gather. It turns out that map generates some chickens inside the foundation of the civil center (use Alt-W to toggle wireframe mode to see this), and because they are the closest target to the civil center dropsite, all these workers are trying to gather from it but failing. That in turn economically cripples the AI.

It's just a bug in the Deep Forest script, I'll create a ticket for this. Thanks for reporting it

Hm, yes, my fault. I noticed that before but didn't think that would cause problems. I'll fix it...

The random map script takes care of no entities are outside the walkable map.

Edited by FeXoR, 26 June 2012 - 09:40 AM.

[quantumstate]

Pureon, on 25 June 2012 - 09:33 AM, said:

I thought the AIs would now ignore resources they cannot access?

Working out if they can access a resource is tricky. I think I will need to monitor units trying to gather and set a time out value so the resource is marked as inaccessible if they have been trying to reach it for too long without managing to gather from it.

[Spahbod]

quantumstate, on 27 June 2012 - 10:50 AM, said:

Working out if they can access a resource is tricky. I think I will need to monitor units trying to gather and set a time out value so the resource is marked as inaccessible if they have been trying to reach it for too long without managing to gather from it.

Can't AI use pathfinding? In the end, we probably need it if we are going to add naval support.

[quantumstate]

Spahbod, on 27 June 2012 - 10:53 AM, said:

Can't AI use pathfinding? In the end, we probably need it if we are going to add naval support.

The AI doesn't have access to sufficiently detailed pathfinding information currently. The AI will need access to the standard pathfinder otherwise there will always be small inconsistency bugs, I am not sure how feasible this will be, Philip could probably say.

[FeXoR]

quantumstate, on 27 June 2012 - 10:50 AM, said:

Working out if they can access a resource is tricky. I think I will need to monitor units trying to gather and set a time out value so the resource is marked as inaccessible if they have been trying to reach it for too long without managing to gather from it.

I think this should be inside the unit AI. It should give back some "unreachable target" signal/state IMO.

[Kazu Kun]

Guess its a good thing i brought this up then

[historic_bruno]

quantumstate, on 27 June 2012 - 10:50 AM, said:

Working out if they can access a resource is tricky. I think I will need to monitor units trying to gather and set a time out value so the resource is marked as inaccessible if they have been trying to reach it for too long without managing to gather from it.

It should consider movement too, so if a unit is far away from the target or it takes some time to reach due to obstacles or slow speed, the unit won't get 95% to its destination, timeout and head back to base

quantumstate, on 27 June 2012 - 11:07 AM, said:

The AI doesn't have access to sufficiently detailed pathfinding information currently. The AI will need access to the standard pathfinder otherwise there will always be small inconsistency bugs, I am not sure how feasible this will be, Philip could probably say.

I seem to remember him saying reachability testing will be much more efficient than it is now, but I don't know if that was terrain only or all static obstacles.

[wraitii]

FeXoR, on 27 June 2012 - 11:15 AM, said:

I think this should be inside the unit AI. It should give back some "unreachable target" signal/state IMO.

That seems like the easiest solution for now.

[Zeta1127]

I have played two matches lately where Qbot hasn't done much in 30 minutes, and has a habit of hunting and/or foraging around my settlement. I probably wouldn't have loses if I included siege weapons in the assault.

In the first match, Athenians vs. Celts on Continent, Qbot only had about 30 units, while I had about 200 units, including an army of 40 hoplites, 20 peltasts, and 20 militia, and only lost 4 hoplites in the assault.

In the second match, Spartans vs. Persians on Continent, Qbot had a similar number of units to last time, while I again had about 200 units, including an army of 30 hoplites and 30 skirmishers, and only lost 3 hoplites and a skirmisher in the assault.

[spedgenius]

I have noticed in the maps that I have played, the first half hour or so Qbot will attack viciously, but hen drops off to nothing. when I finally get up enough troop strenght to venture out I meet littel to no resistance and end up killing buildings and workers.

[historic_bruno]

quantumstate, on 27 June 2012 - 11:07 AM, said:

The AI doesn't have access to sufficiently detailed pathfinding information currently. The AI will need access to the standard pathfinder otherwise there will always be small inconsistency bugs, I am not sure how feasible this will be, Philip could probably say.

I noticed http://trac.wildfire...changeset/12074 - is this marking the target resource as globally, permanently inaccessible? What if a few gatherers are (temporarily) trapped but any other unit could reach it? And is this inaccessibility data updated/invalidated at some point?

[quantumstate]

historic_bruno, on 05 July 2012 - 10:53 PM, said:

I noticed http://trac.wildfire...changeset/12074 - is this marking the target resource as globally, permanently inaccessible? What if a few gatherers are (temporarily) trapped but any other unit could reach it? And is this inaccessibility data updated/invalidated at some point?

This is permanent, each AI acts independently. Normally this will only occur when a unit spends 4 minutes trying to reach a resource and it only triggers if none of the resource has been gathered. I think it will be very rare for false positives to occur.