Same teams, different sides, different results.

pier4r
Skynet
Skynet
Posts: 3390

Same teams, different sides, different results.

Post#1 » 24 Mar 2017, 11:28

As discussed here: viewtopic.php?f=8&t=362

This time we have the confirmation that it is the same in career mode.

I and chris did not change the AI on kingmaker in those two matches.
476624

476628
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

harthag
Automaton
Automaton
Posts: 197

Re: Same teams, different sides, different results.

Post#2 » 24 Mar 2017, 14:39

The first noticeable change I could see was about 30 seconds in, when the shotguns are both going for the lone resource.

I don't see much variation in the actions of mrchris's shotgun in either replay, but it might just be too subtle for me to pick out without a debug window.

On pier's, however, the shotgun takes very different actions on tje 2 replays. On one, it stays almost level with mrchris's shotgun, while on the other it retreats way ... south? ... toward bottom right edge? ... the rest is "butterfly effect" from there

Wozza
Script
Script
Posts: 28

Re: Same teams, different sides, different results.

Post#3 » 27 Mar 2017, 04:16

I took a look at this while waiting for my pending matches to fill.

I can identify a difference within the first 4 seconds of the match.

When Pier plays white his shotgun moves in 3 different directions prior to taking its first shot. When Pier plays red his shotgun only moves in 2 directions prior to taking its first shot.

The linked thread pins the problem on collision handling so this is interesting because in the replay the closest to a collision I can see in the early seconds is mr chris's shotgun cutting in front of the sniper, but I can't actually see any change in movement so not sure whether they are actually colliding or not.

mcompany
Autonomous Entity
Autonomous Entity
Posts: 872

Re: Same teams, different sides, different results.

Post#4 » 27 Mar 2017, 10:10

Oh, the linked thread definitely isn't the cause of what is going on, and I've already said about the actual cause several times. The actual cause of the replays being different despite the same AIs is that currently each different side is reading nodes differently. The AI on one side might see a situation and say one more is valid, meanwhile, the other side might classify the same action as invalid. I currently don't know which is actually correct (nor do I have the time to make AIs to test this since I don't think I can play on my laptop anymore, nor do I simply have enough free time to test even if I had the AIs ready)

Wozza
Script
Script
Posts: 28

Re: Same teams, different sides, different results.

Post#5 » 28 Mar 2017, 01:17

I started digging into why when my Mind Game AI fights itself in a test match the 2 sides start behaving differently.

I found a situation where on the previous tick the shotguns from both sides are at medium range from each other and both moving toward the opposing machine gun.

On the next tick the white shotgun evaluates a node "if medium range, moving to ally, shotgun enemy exists" to true and further down the tree this results in the white shotgun switching to moving towards the red shotgun.

On the same tick the red shotgun evaluates "if medium range, moving to ally, shotgun enemy exists" to false - which to me looks like an error - and this leads to the red shotgun taking a different action from the white shotgun.

I put in a debug node to isolate the medium range piece of the logic and both the red and white shotgun evaluate "if medium ranged shotgun enemy exists" to true.

So based on what I can see I think the issue is in the "moving to ally" piece of logic and possibly the fact that the white shotgun is changing who it is moving to on the current tick is somehow impacting how the red AI evaluates the white shotguns action from the previous tick.

User avatar
GFX47
Dev
Dev
Posts: 2914

Re: Same teams, different sides, different results.

Post#6 » 28 Mar 2017, 11:29

Thank you for investigating it. I'll check the code ASAP to fix it.

User avatar
LuBeNo
Autonomous Entity
Autonomous Entity
Posts: 532

Re: Same teams, different sides, different results.

Post#7 » 17 Apr 2017, 16:55

Wozza wrote:On the next tick the white shotgun evaluates a node "if medium range, moving to ally, shotgun enemy exists" to true and further down the tree this results in the white shotgun switching to moving towards the red shotgun.

On the same tick the red shotgun evaluates "if medium range, moving to ally, shotgun enemy exists" to false - which to me looks like an error - and this leads to the red shotgun taking a different action from the white shotgun.

I could reproduce this behaviour in a simple AI here.

IMO the collision handling is perfectly fine.
Image
My algorithm of life: if(self.tired) sleep(); else if(self.hungry) eat(); else follow(Jesus);

User avatar
NullPointer
Autonomous Entity
Autonomous Entity
Posts: 539

Re: Same teams, different sides, different results.

Post#8 » 18 Apr 2017, 15:03

It's great this is in Alpha 8's TODO. Any improvements to any AI atm has a luck factor since you may be "improving" an AI that would have won if swapping sides.

TheKidPunisher
Autonomous Entity
Autonomous Entity
Posts: 565

Re: Same teams, different sides, different results.

Post#9 » 24 Apr 2017, 00:04

How can i share a battle between 2 of my AI to you GFX? because i got the problem of different result when i switch the side...

User avatar
GFX47
Dev
Dev
Posts: 2914

Re: Same teams, different sides, different results.

Post#10 » 24 Apr 2017, 21:17

That's OK, I have enough info to study it ;)

Return to “Bug Reports”

Who is online

Users browsing this forum: No registered users and 1 guest