Why the narrow-sighted "AI" is a problem on Madden and NCAA

[LBzrule]

He is correct the A.I is too simplistic. Right now they can't even get field awareness on defense in. I'm tired of defenders in key situations pushing the offensive guy for the first down when he has the guy stopped or worse the TD. What the hell is that? Not only are they not aware of the field, as rhombic stated, they are not even aware of their own players. Even in pursuit sometimes your own player who is looking directly at you will get in your way and cut you off and bling the offensive player is gone, not because he did anything special, but because your own guy fighting against you. On 4th and 1 or on the Goal line, you can have the guy stopped only to have your own guy say NOPE!! NO!! Let him get this TD.

There has to be some kinda sphere(s) of awareness that they can incorporate to deal with these issues. One of the things that plagues NCAA so much is the defenders have only a 1 track mind on how to stop a guy and that's tackle him. So when you see a guy get going down the sideline your defenders all come running and when they get close, rather than push him out of bounds, they feel they need to get closer to tackle him, so they lose their angle and end up falling behind him. Why don't they just push the man out of bounds instead of getting out of position to do anything at all?

[Champ24]

Rhombic you make valid points, hence the separation from real life football and video game football. I'm assuming you have to have one bad a** processor to implement the kind of logic you are talking about or either scale down the defensive playbooks. There are just so many what if scenarios the developer has to account for to cover real life. Then each defensive player has to be "thinking" individually during each play based on what going on in front of them based on the play that was called. I think it would take a incredible amount of development time and resources to pull something like that off successfully.

[Qb]

Very good stuff, Rhombic. I think the points you raise are really highlighted by the game of football, but valid in many other sports games as well. AI players (both user & CPU) need to be able to read & react to other players on the field, what's happening at the time and the game situation. This should be the goal of every development team, but it is a truly ambitious aim. To program AI in a way that it can function on an almost human level is quite a task. Still, great stuff man and excellent food for thought...

[Joborule]

Quote: Originally Posted by rhombic21

I recently posted a video highlighting several problems with zone coverage on NCAA, http://files.filefront.com/ZoneDIssu.../fileinfo.html , which shows why zone coverage on NCAA (and presumably Madden) is fundamentally flawed. The flaws with zone coverage, while glaring in themselves, highlight an even broader problem with the "AI" on these games. The "AI" is extremely narrow-sighted. Each player is programmed to have a specific assignment, and he executes that assignment according to a collection of ratings. He is completely oblivious to what the rest of his team-mates are doing. As a result, he is incapable of understanding what the weaknesses in the defense/play are, and how he should be prepared to react accordingly. The problem with this approach is that it's completely contrary to real football. For example, in real football, corners in cover 2 will sink back to give help on vertical threats when there are no threats to the flat area, and the middle linebacker will sink back into the deep middle when there are no threats in front of him. Or if you bring a heavy blitz. Secondary players (particularly veteran secondary players), understanding that the blitz is going to probably force a quick throw, will often squat on short routes and force incompletions or turnovers. Another area where this becomes apparent is with the blitz and defensive line rush. In real life, you don't have the serious issues with containing QBs from rolling out that you do on the game, because the defenders rushing the passer have a basic understanding about what their rush lane is, and how it fits into the scheme. Thus, outside rush defenders (DEs or players blitzing off the edges) understand that they have containment responsibilities and take rush angles that force the QB to stay in the pocket. They do these things because the understand what the other 10 players on defense are doing, and know where the holes in the defense are that the offense is probably going to try and attack. On the game, there is no real capability for players to understand concepts. You can assign a player to guard a flat area, but he doesn't really have any capability to understand WHY he's covering the flat area. So you get things like the video posted above, where the players end up guarding nobody at all, instead of being able to make an adjustment on the fly, within the scheme, to adapt. On offense it's not as significant of a problem, since the offense can dictate the action, and thus the user can manually make adjustments for his players, knowing how the defense is going to react. But it still crops up. Notice on running plays where lead blockers will ignore a relevant defender and turn back to block somebody behind the play? I see this all the time. I turn the corner on the play, and there's one defender about 10 yards downfield, who (due to separate AI issues) is backpedaling away from the ball. Rather than realize that we have gotten outside the defense and I need him to seal this defender off so I can head up the sideline, my lead blocker will often turn back and try to block somebody back towards the middle, behind the play, forcing me to try and beat the defender 1-1 in the open field. That's because they don't really understand the play at all. They don't have any understanding to know where the runningback is attempting to run the ball, and be able to seal off a hole to clear a path for him. These games need significant AI advances in a number of areas. Adaptive AI is another issue. But until there's some ability for defenders to understand concepts or schemes, rather than just their specific assignment, the AI will always be significantly limited. As a result, smart players at positions other than QB will continue to be mostly useless compared to players with superior physical attributes (speed, strength, height, etc...) In reality, smart players matter a great deal, and not just because they can see play-art or because they don't bite on play action. Teams like the Patriots dominate on defense, despite not necessarily having tons of speed or athletes with elite physical abilities, because they have 11 defenders that understand the scheme and don't get confused or end up out of position. They are able to adapt to changing situations and make adjustments on the fly, something that players on Madden and NCAA have no ability to do.

Great post. The AI does need to be a lot more adaptable to it's surroundings and adapt to it. What makes great players great is by how adaptable they are. Great football sense in general. And not just in the play, but throughout the game and even other games throughout the season. Adaptable AI like this would be very advance AI, but I think it can be done and is necessary. When I see that NCAA video of the goal line run and everyone running in the endzone instead of to the ball carrier, I just laugh how brutal that is.

[GiantKingNy]

Quote: Originally Posted by Joborule

Great post. The AI does need to be a lot more adaptable to it's surroundings and adapt to it. What makes great players great is by how adaptable they are. Great football sense in general. And not just in the play, but throughout the game and even other games throughout the season. Adaptable AI like this would be very advance AI, but I think it can be done and is necessary. When I see that NCAA video of the goal line run and everyone running in the endzone instead of to the ball carrier, I just laugh how brutal that is.

Thanks, Great read.

But how do you translate duality into code, this is more complex then progression analyst horizontally. And for a video produced cyclically, we are dreaming.

[g2thecore]

I agree....Great post btw....Like the above poster stated, I doubt we'll be seeing these advances in AI any time soon. I hope that sometime in the near future though, we do get to see something that's remotely close in comparison.

[callmetaternuts]

Quote: Originally Posted by LBzrule

He is correct the A.I is too simplistic. Right now they can't even get field awareness on defense in. I'm tired of defenders in key situations pushing the offensive guy for the first down when he has the guy stopped or worse the TD. What the hell is that? Not only are they not aware of the field, as rhombic stated, they are not even aware of their own players. Even in pursuit sometimes your own player who is looking directly at you will get in your way and cut you off and bling the offensive player is gone, not because he did anything special, but because your own guy fighting against you. On 4th and 1 or on the Goal line, you can have the guy stopped only to have your own guy say NOPE!! NO!! Let him get this TD. There has to be some kinda sphere(s) of awareness that they can incorporate to deal with these issues. One of the things that plagues NCAA so much is the defenders have only a 1 track mind on how to stop a guy and that's tackle him. So when you see a guy get going down the sideline your defenders all come running and when they get close, rather than push him out of bounds, they feel they need to get closer to tackle him, so they lose their angle and end up falling behind him. Why don't they just push the man out of bounds instead of getting out of position to do anything at all?

Great comment here, about the D only going for the tackle. i HATE that. They take an angle to the "sure" tackle or to the pylon when they could just cut them, push them out of bounds, or slow them up enough for "help D" to be available.

[Ian_Cummings]

All good points.

Out of sheer curiosity (I'm not trying to deflect here), are there any other sports games out there that you all feel have best-of-breed AI? Before I joined EA I was always kind of in awe that they were able to correctly model football so well (since it's by far the most complicated sport), but as we all know as the visual fidelity increases in games, so do the expectations of every other part of the game (i.e. animation, AI, lighting, etc)

I'd like to see what everyone thinks so I am not missing anything and I can research them further.